A single-elimination debate tournament billed at 32 contenders (the arithmetic says 30), covered touch-and-go. The individual rounds are specimens; the bracket is the experiment — what a format selects for when an audience votes. Rounds 1 and 2 complete; round 3 (the quarterfinals) has aired, two of four decided; the tree is fully reconstructed and the final is 2026-09-12 in Nashville.
Links: Debates index, Constructed ≠ Arbitrary, The Weighting Problem
32 contenders, single elimination, opening round branded “Thunder 32” — though the arithmetic of the completed round 1 gives 30 (see below). Two finalists advance to a live WW1 event. Format per round: 5-min openings → sponsor → ~15-min open floor → (sometimes) a moderated question round → closings. Moderators rotate. Advancement is decided by audience write-in ballot on a 48-hour window at wordwardebate.com — a mechanism that matters (see Cross-round findings).
Method: reviews are built from a cleaned auto-caption transcript plus real-time commentary captured while listening. That mode produced the densest discussion in the bracket (sixteen sections off one hour of tape) because reactions are caught at the moment they occur rather than reconstructed after. Its ceiling is length — it scales with runtime, not with signal — so the WW1 main-card debates (1:56 and 2:07) will need a different approach: timestamped notes on marked moments, or a selection pass before commenting.
The sampling fact behind the information channel, stated plainly (2026-08-17). > Chris: “I seem to know about 1/2 the contestants personally.” Named so far: Glenn Lawrence (military service), Tareyak (feminism rerun), one of the two in the hate-speech round, Kewl Vic (party loyalty), plus Charsky and Bunn from the AI round. This is worth stating as a finding rather than a footnote, because it cuts two ways. It is evidence for finding 2 — a bracket in which one observer personally knows half the field is, by construction, seeded from a single community, which is exactly the condition under which affiliation rather than reach decides a 30–105-ballot round. And it is a standing caveat on the predictions ledger: several calls are made with information no transcript contains, so a correct prediction is not automatically evidence that the stated reasoning was right.
Coverage policy: deliberately not exhaustive. Rounds get picked up when something structurally interesting happens, not to complete the bracket. Chris knows and has met a number of the contenders, so the series is partly followed as an audience member — which also means there is an information channel here beyond the transcripts.
The reason the bracket is worth tracking as a unit rather than as a pile of reviews. These only appear across rounds:
Moderator interference — the third data point breaks the pattern, and the real variable is attribution. After two rounds the read was “format property.” A targeted check of the Boomers round (moderator: Pisco) does not support that. Three moderators, three behaviours:
| Moderator | Round | Behaviour | Verdict |
|---|---|---|---|
| Molyneux | Masculinity | Argues a side at length; supplies the aff’s best case himself after closings | Interference |
| Kyla Turner | Feminism | Restates a debater’s losing answer as the far stronger one he never gave, attributed to him | Interference — and the worse kind, because it launders |
| Pisco | Boomers | Forces the aff to specify culpability vs. mere causality; isolates where the two already agree; offers a counter-argument explicitly flagged (“here’s a response that I’m kind of channeling”) | Not interference — arguably the correct technique |
So the earlier “questions vs. content” line was too crude. A moderator may supply content; what matters is whether it is labelled as the moderator’s own construction and put to both sides, or laundered into a debater’s mouth as something they said. Pisco supplies a demographic-collapse counter and names it as his; Kyla supplies a weight-equivalence argument and calls it Miles’s. That is the whole difference. Chris’s series-level read (“interfering too much so far”) survives as a claim about particular moderators, not the format.
Confirmed on the full tape (Boomers review §5) — the earlier caveat is retired. Pisco’s technique holds across the whole round, and the round’s structural spine (culpability vs. mere causality) exists only because he installed it. He also asks the best moderator question in the bracket so far: “What would it take for you to be convinced?” — the falsifiability question, put neutrally. The review logs one boundary case: he blocks a debater’s etymological move mid-round, which is a moderator adjudicating substance. It clears the rule only because he states it as his own rather than laundering it. A stricter rule would fail it — which is a useful stress test of where the line actually sits.
The etymology pair — the same intervention from two moderators, on opposite sides of the rule (2026-08-17). The stress test above now has its matching half. In the hate-speech round, the Aff uses discriminate in its descriptive sense — telling things apart — which is a real sense of the word and the one his equality argument depends on. Kyla blocks it, requiring him to clarify “what you mean which is not just to tell the difference between things” — a clause that does not ask which sense he intends but rules one out, and the sense ruled out is the one his case needs. The Neg’s definition survives and the segment proceeds on it.
Set against Pisco: identical move, opposite side of the line. Pisco blocked an etymological move and owned it as his own construction, which is why it cleared the attribution rule as a boundary case. Kyla’s arrives dressed as a request for clarification, so the room hears a neutral moderator asking a question rather than a moderator settling a contested term. That is a cleaner statement of where the line sits than either round alone: the rule is not “don’t supply content,” it is “don’t launder a ruling as a question.” Chris’s read of her across this round is blunter than the tape looks — “Kyla is constantly steering the narrative… she immediately cuts in and champions the left’s definition” — and it overturns the reviewer’s initial verdict that she was clear of the round-2 charge.
The scoring mechanism confounds every inference from the bracket. Time-boxed write-in ballots also measure reach, so advancement is over-determined: legibility and popularity both predict it. “This format selects for legibility over correctness” stays unfalsifiable against “it selects for audience size” until a round appears where the two point opposite ways. Kyla explicitly instructs voters against popularity — worth noting the organizers are aware of it.
Two sub-mechanisms, from Chris’s own experience as a debater rather than from the tape. “Reach” was too coarse — it was being read as raw audience size, and at least two distinct things hide inside it:
Rounds fail at different layers, and the layer is diagnostic. Arguments over a contested term have a stack — definitional → mechanism → empirical. Round 1 never escaped layer 1 (58 minutes on what masculinity is). Round 2 cleared it in ninety seconds and then failed at layer 3, running two parallel monologues against undefended metrics. Two distinct failure modes, both fatal, neither obvious from a summary.
The stack needs a floor — layer 0. The Boomers round never reaches layer 1, because the two sides never share a standard for what counts as evidence at all (one side’s causal mechanism is planetary transits, asserted at “almost always 100%”). This is diagnostically distinct from a layer-1 failure: layer 1 is two people who could adjudicate a definition and don’t; layer 0 has no adjudication procedure to reach. Neither debater is being unreasonable given their own standard, and the round is still dead on arrival. The tell is a position that cannot name its own defeater — Pisco asks Mariah directly what would change her mind and gets no condition back.
Topic-talk vs. resolution-work. The craft lesson, and it generalizes past this series: every minute is affirming the proposition, negating it, or wasted — however interesting. The test for any line of argument is whether you can state the sentence connecting it to the resolution. Round 2’s dating-app segment was the central argument in microcosm and still read as a digression, purely because neither debater tied it back. Framing is the debater’s job, not the moderator’s.
Superseded as best-in-bracket (2026-08-11). Chris’s read after nine rounds: “this was the best debate so far, even above the AI debate” — Turner vs. Bourdeau. Different virtue, though, and worth keeping both: the AI round is two well-matched debaters doing the resolution-work correctly; the fragility round has a live hinge that both sides keep fighting over once the term is pinned. One is a model of technique, the other of a real disagreement. The AI round remains the control case for this finding; the fragility round is the control case for The Load-Bearing Word.
Control case found — the AI round. Charsky and Bunn waste nothing: both weigh explicitly, Charsky’s closing opens “reasons why you should favor me winning a debate,” and both state the crux out loud (the counterfactual baseline). Which raises the confound: is the finding about the format, or simply about these two being better debaters? A format that permits both outcomes isn’t obviously selecting for either. Open on the review as seed 1.
True position, lost round. — substantially weakened by the results. The original claim: twice the side the vault read as correct lost on execution, so the series’ real finding would be about formats rather than topics. The round-1 results do not support it.
| Round | Position the vault reads as true | Outcome |
|---|---|---|
| Masculinity | Garcia | Lost, 25.8% — confirms |
| Rich-without-luck | Neg / The Aftermath (Chris’s read) | Won, 71.88% — counterexample |
| AI | Charsky’s side, on the vault’s own three-layer rule | Won, 79.05% — counterexample |
Third counterexample, and it retires the finding (2026-08-17). The military-service round was the designed test: the vault said the argument went to Brunet and the ballot would go to Lawrence on rhetoric and appeal. Brunet won 78.6 – 21.4. And the conditions favoured the wrong answer as hard as this bracket can arrange — the Aff was a USAF veteran arguing a veterans-first proposition to an audience that advertises patriot credentials, so rhetoric and audience prior both pointed at Lawrence. He lost by 57.
So the running tally is one confirmation (Garcia) against three counterexamples, and the three counterexamples are +43.8, +58.1 and +57.2. Garcia is better explained by finding 6 than by any bias against correctness. This finding is now retired rather than merely demoted; what replaces it is finding 15.
One confirmation and two counterexamples, both of them decisive margins. The honest revision: the format does not systematically punish the correct side. What round 1 actually shows is closer to the opposite — the two rounds where the vault had a clear view of who was right resolved in favour of that side, and the one loss (Garcia) is better explained by finding 6 (delivered ≠ received) than by any bias against correctness. Retained as a watch item for round 2, not as a finding.
Two ways to lose an audience — and a technique/format mismatch. “Delivered ≠ received” splits into distinct failure modes. Garcia lost his audience at the level of prose (a private vocabulary nobody could parse); Charsky loses his at the level of delivery — clear prose, announced structure, dense citation, all of it delivered fast and monotone in the manner of competitive policy debate. Those habits are optimized for a trained judge flowing arguments on paper, where dropped points score as conceded. Under an audience write-in vote the same technique inverts: density becomes noise and citations evaporate at speed. Not a deficiency — a mismatch between a tuned technique and a different scoring function. Extension (2026-08-14) — the mismatch also operates on form, not just delivery. The finding above is about delivery — Charsky’s fast, dense, citation-heavy competitive-policy manner inverting under an audience ballot. The military-service round shows the same mismatch one level up, in argumentative structure.
The correct technical answer to an inflating affirmative there is a counterplan: advocate national service — non-topical by construction, competitive because “prerequisite” is a necessary condition that a sufficient alternative destroys, and guaranteed to solve the harms because the affirmative attributed them to national service himself. Under a trained judge that is simply a negative ballot. Under an audience vote it is a liability, because the affirmative and most of the room read the negative’s agreement as concession:
Chris: “Glenn would think ‘Yeah, neg agrees with ME! so I win!’ not understanding that he is refuting the resolution… I don’t think many people would understand this tactic and it might be losing because of it.” … “structured debate lovers would enjoy this, but it alienates 90% of the audience.”
So the generalisation: the gap between winning a round and being seen to win it widens with the technical sophistication of the move. A format scored on perceived clash cannot register a win that consists of manufactured agreement. This is a real constraint on what a competent debater should do in this bracket, not merely on how fast he should talk — and it is the strongest version of finding 2’s worry, because it says the format doesn’t just fail to reward the correct response, it can actively punish it.
Untested escape: strip the machinery and keep the ballot question concrete — “neither of us is defending tonight’s resolution; should a firefighter who never enlisted be allowed to vote? He says yes, I say yes, the resolution says no.” Same structure, no jargon. Nobody in the bracket has tried it.
Second specimen, one round later — and it confirms the finding from the other direction. The military-service case was hypothetical: a technically-decisive move that wasn’t run, judged in advance as a probable vote-loser. The feminism rerun supplies the live comparison, because there the audience-facing counter was run and the tactical one wasn’t. Facing a squirrel affirmative (one instance satisfies me; you must refute every interpretation), Tareyak simply declined the burden framing and kept re-asking the general question, while the available tactical kill — pin the case to the single example in its constructive, refuse everything later as new argument — went untaken. Chris grades his own tactical counter down: “the neg’s approach has broader appeal because mine is ‘too tactical’, and people wanted to hear the Neg’s framing of the debate.”
The structural reason, now stated as a rule and promoted to The Negative’s Easy Burden: the tactical counter’s win condition is my opponent failed to discharge a burden, which requires the room to accept a burden allocation nobody announced; the framing counter’s win condition is my account of the subject is better than his, which is what the audience already thinks it is voting on. So the finding generalises past delivery and past argumentative form to win conditions themselves: a lay ballot can only register a victory whose condition it already recognises. That is the cleanest statement this bracket has produced of what an audience vote can and cannot measure.
The format punishes the disposition that makes discussion valuable. The AI round was the best conversation of the bracket — both men truth-seeking rather than playing bloodsports — and the debater who qualified himself honestly reads as unstable. Concessions that are intellectual virtues function as competitive liabilities, and a performance-scored vote cannot distinguish an honest qualification from a wobble. Chris: “I like both of them because they want discussion over blood-sports, but in this competition they need to play the game.”
Not every round has a proposition in contention — the verbal dispute. Guptill vs. Tejada is well-formed at every layer of finding 3’s stack and empty at the top: definitions get settled (twice), evidence standards are shared, both men reason competently — and they agree. Strip the contested word and the Neg (“society won’t speak hard truths and punishes those who do”) and the Aff (“society coddles, affirms delusions, rewards the unearned”) assert the same claim, which is why the Neg answers “Absolutely” when handed the Aff’s central evidence. The tell is a debater saying “it sounds like you agree with me — don’t you technically concede?”: he is right about the agreement and wrong that it’s a concession, because you cannot concede to someone asserting your own view under a different label. Diagnostic: state both sides without the contested term; if the disagreement disappears, it was verbal. This is orthogonal to the layer model rather than another rung on it — the question is there a proposition in contention? is asked before the stack is entered.
The ordering isn’t random — it’s a stable convention that systematically inverts, which is worse. Corrected 2026-08-12 after Chris pushed back on the “haphazard” reading; he was right and the checked version is sharper. Across all ten covered rounds:
| Signal | Pattern |
|---|---|
| Introduction order | The Neg is introduced first in 10 of 10. “Debater one” is always the side denying the prompt — a completely stable convention |
| Speaking order | The Aff opens in 7 of 10. The three exceptions (feminism, AI, too-sensitive) are all Kyla-moderated — but Kyla also runs Aff-first rounds, so it is inconsistency within a moderator, not a per-moderator rule |
| Closing order | ⚠ Not a single convention — see the correction below. The rule stated in Kyla-moderated rounds is whoever opens first closes first (“since Forest got the first word, Jose, you’re going to get the last word”), but Mullally runs a different rule, and runs it inconsistently |
Correction — closing order varies by moderator, and one moderator varies within himself (2026-08-17). Prompted by Chris’s suspicion that “the closings are not always the same,” checked against seven archived transcripts. His worry that speaker labels would make this unrecoverable turns out not to bind: the moderator announces the closing order out loud in every round, so the convention is readable straight from the text.
| Round | Moderator | Announced | Effective rule |
|---|---|---|---|
| Feminism (R2) | Kyla | “whoever got the beginning word, the opposite gets the final word” | opener closes first |
| AI | Kyla | “since Charsky got the first word, you get the last word” | opener closes first |
| Too sensitive | Kyla | “since Forest got the first word, Jose… the last word” | opener closes first |
| Algorithm | Kyla | “Ryan, you got the first word. Silvio will get the last word” | opener closes first |
| Hate speech | Kyla | “since Octavius, you got the first word. Chase, you get the last word” | opener closes first |
| Western therapy | Mullally | “Spencer as the affirmative, you have the last word” | Aff closes last |
| Feminism rerun | Mullally | “David started the debate. You go first and then David… the last word” | Aff closes last |
| Party loyalty | Mullally | “Joe, because you have the affirmative, you’re going to get a closing first” | Aff closes FIRST |
So: Kyla is perfectly consistent across five rounds on a positional rule (opener → first closing). Mullally uses a side-based rule instead — and applies it in opposite directions, giving the affirmative the last word twice and the first closing once, each time announced as though it followed from being the affirmative. Chris’s read is confirmed and sharpened: “most often Aff closes first, but a few times Neg closed first” is right, and the reason is that two incompatible conventions are in play and neither is the format’s.
⭐ Third correction (2026-09-07) — a third convention, and it is the exact inverse of the first. Monica Perez has now moderated four round-2 rounds and announces the order out loud in every one. She is 4 for 4 on a positional rule that is Kyla’s rule reversed:
| Round | Announced | Opener | Closes first | Effective rule |
|---|---|---|---|---|
| Suffrage | “Jose, you are up first because the Aftermath started the debate” | Aftermath | Tejeda | opener closes LAST |
| Iran | “Because Spencer went first in the beginning of the debate, it is Silvio’s time to go first” | Summerhays | Cruz | opener closes LAST |
| Amnesty | Rex closes, then “Owlish… this is all the last word” | Owlish | Jones | opener closes LAST |
| Abortion | “Cool, you’re up first since Octavius kicked us off” | Thunder | Vic | opener closes LAST |
Kyla: opener closes first, 5 for 5. Perez: opener closes last, 4 for 4. Mullally: side-based, and self-inconsistent. This kills the charitable reading that the format is merely sloppy. Neither woman is being careless — each applies a rule, correctly, every single time, and the two rules are opposite. Nine rounds of perfect internal consistency across two moderators who cancel each other out is a stronger indictment than nine rounds of randomness would be, because it shows the convention is being taken seriously at the moderator level and simply does not exist at the format level. The last word is a real advantage in a round scored by audience recall, and in this tournament it is allocated by which moderator you drew — in a bracket whose three narrowest rounds went by 12.4, 16.8 and 9.0.
Why it matters rather than being trivia: the last word is a real advantage in a round scored by audience recall, and formal systems allocate it deliberately — the side carrying the burden gets the bookends (The Negative’s Easy Burden § the institutional admission). Here the advantage is assigned by which moderator drew the round, and in Mullally’s case by which rule he reached for that night. That is a per-round coin-flip on a structural edge, in a bracket where rounds turn on a dozen ballots.
Second correction — the opening-order count looks wrong and needs a verification pass. This finding records the Aff opening in 7 of 10 with three exceptions (feminism R2, AI, too-sensitive). The transcripts support feminism R2 (Ruelas, listed first, gets the beginning word) and too sensitive (Guptill opens) — but not AI: Charsky takes the first word, and Charsky is the affirmative. If that holds the count is 8 of 10, not 7. Flagged rather than silently amended, because the Aff/Neg assignment for that round is inferred from its review rather than from the intro. Also note this refutes the stronger version Chris proposed — “we are 100% that Aff opens” is not right; two genuine exceptions survive the check.
So the intro order inverts the speaking order in the normal case — the Neg is named first and the Aff talks first. Two signals, pointing opposite ways, neither announced. A contender who reasons “the Aff opens, and I’m not opening, so I must be the Neg” is right 70% of the time and catastrophically wrong the rest — which is precisely what happened to Jose Tejada, who drew one of the three Neg-first rounds and was introduced second, so both cues misled him simultaneously. He argued his opponent’s case for a fifth of the round.
Cheap fix for the organizers: open with the same side every time, or say out loud which side is which before the clock starts. The moderators already state the position in the bios; the failure is that the ordering contradicts it. Debate convention says the side carrying the burden opens; here the order is effectively arbitrary and is never announced as arbitrary. A contender who reads it as a signal of his own assignment will be wrong about a third of the time, which is one plausible contributor to Jose Tejada spending a fifth of his round arguing his opponent’s case. Caveat: the assignment was stated explicitly in the intro, so this is a contributing defect, not an excuse. Cheap fix available to the organizers: fix the order, or say out loud that it’s random.
The audience prior is the specific pool’s, not the general public’s — and this pool is agency-affirming. The sharpest lesson from the one wrong prediction. Chris forecast the Aff in the wealth round on the reasoning that “people who are not rich want to believe the topic” and “it seems easier to blame something else than to solve the problem.” That is a sound intuition about the general population and it inverted against this audience: the Aff took 28%.
Look at the two largest margins in the bracket and what they have in common:
| Round | Winner’s position | Margin |
|---|---|---|
| Too sensitive | Tejada — people are coddled; struggle and merit are real | +89.1 |
| Rich-without-luck | The Aftermath — you can build wealth by your own action | +43.8 |
Both are agency-affirming: outcomes follow from what you do. Both won enormous. A bracket whose contender bios advertise “the working man’s resume,” “straight white Christian father patriot,” and Infowars credentials is not drawing a general-population sample — it draws people for whom the personal-responsibility answer is the flattering one.
So finding 2’s motivated-reasoning sub-mechanism needs a correction rather than a deletion: it is real, and its direction must be read off the actual voting pool. Applying a general-population prior to a self-selected audience will get the sign wrong, which is exactly what happened here. Caveat: two specimens, and the agency reading is cleanest for those two — Owlish’s win is better explained by delivery, and Charsky’s by argument quality. Offered as the leading hypothesis for round 2 to test, not as an established finding.
⭐ Round 2 tested it and forced a real amendment (2026-09-07): “the pool’s prior” is not one thing. Three round-2 rounds went against the coded prior — abortion, Biden/Trump, Iran — and the abortion round shows why, because it puts two strands of the “signature” directly against each other on tape and one of them wins:
| Axis | Content | Head-to-head result |
|---|---|---|
| Anti-establishment / bodily autonomy | don’t let the government decide what you do with your body; COVID-mandate memory; anti-institutional | wins |
| Social-conservative / traditionalist | life begins at conception; traditional marriage; demographic replacement | loses |
| Agency-affirming | outcomes follow from what you do | undefeated, but never yet opposed by another axis |
The collision is explicit in the transcript: the Aff asks “Did you take the COVID vac?”, the Neg answers honestly that he opposed the mandate, and the frame is gone. So the finding upgrades from “read the prior off the pool” to “read the axis off the resolution, then the prior off the axis” — and where a resolution loads two axes, the anti-establishment one has the only observed head-to-head win. Registered for round 3, where two of four resolutions are about war and the anti-establishment axis has a pure expression (non-intervention) available to one side in each: see the Iran double-header. ⚠ The competing explanation for the abortion round is plain reach — 3,495 views against a round-2 ceiling of 822 — and nothing in the remaining bracket separates them.
The neutral-dilution hypothesis — why later rounds should measure performance better. Chris’s structural prediction, registered before round 2 airs:
Chris: “I figured the early rounds were mostly popularity contests, but as we move up, they will attract larger audiences so a larger streamer doesn’t win by default against better performance.”
The mechanism is dilution. A round-1 ballot is decided largely by voters who arrived as partisans of one contender, so affiliation (finding 2) dominates. As the bracket advances, two things change: both contenders have already cleared a round, so neither has a raw-reach edge by default; and the series accumulates an audience that follows the tournament rather than any one debater. The neutral fraction rises, and with it the weight of performance.
This is a genuine, falsifiable prediction with a clean test: margins should compress and correlate less with follower counts as rounds progress. If round 2 produces another 89-point blowout for the better-known name, the hypothesis is in trouble.
The reach analysis — attempted 2026-08-11, and the attempt answered the question a different way.
Chris: “we would have to run some sort of reach analysis for each contestant to see if this was just a popularity contest.”
The plan was to collect each contender’s subscriber/follower counts and regress vote share on relative reach. It cannot be run as designed, for two reasons that are themselves the finding:
⚠ The search failure has a cause, identified by Chris 2026-08-12: “part of what makes the reach research hard is they are mostly using real names.” Contenders are billed as Josh Smith, Nic Turner, Silvio Cruz — names that return nothing, while their actual channels run under handles. Chris supplies the worked example: Josh Smith’s channel is “JS Urban Adventures,” with a decent following, plus the streaming show The Dreaded Conservative (200+ episodes). Neither is reachable from the name in the bio. So the reach data mostly exists and is simply not addressable by the identifier the bracket publishes — which downgrades the conclusion below from “there is nothing to measure” to “it cannot be measured this way.” A workable method would go handle-first: pull each contender’s channel from the debate description, their own socials, or by asking, rather than searching the legal name.
Confirmed 2026-08-17, and the hate-speech round supplies the identity mapping in both directions on one tape — which is the strongest evidence yet that the data exists and only the identifier is wrong. Kyla bills Chase McPherson by legal name and then reads out the handle (@_vigilante_tv, YouTube and Twitch); she bills Octavius Thunder by handle and then gives the legal name (Eli Halpern). Chris supplied the McPherson mapping independently before the tape was pulled. So the bracket is not consistently using real names or handles — it uses whichever the contender prefers, and the intros carry the cross-reference. The practical upshot: the reach data is recoverable by watching the first ninety seconds of each video, which is a far cheaper method than any search, and the bar to running finding 12’s analysis is now labour rather than availability.
So the correct reformulation: round 1 is not a popularity contest in the streamer sense — it is a turnout contest among personal networks. That is a meaningfully different claim, and it explains the 52–3 result better than reach ever could.
One case where the profile gap is legible, and it cuts the right way. In the AI round, the higher-profile figure is clearly Bunn — a credentialed academic (PhD Wisconsin–Madison, Assistant Professor at Covenant College, published in Comment and Providence, an ISI Top-20-Under-30 award, an established X presence). Charsky has no findable public footprint at all. Bunn lost 83–22 on the largest pool in the bracket. Whatever decided that round, it was not the participants’ relative standing — which is the strongest available evidence that the AI result is a genuine quality signal.
⚠ The decisive fact about this bracket: rounds are decided by 30–105 ballots. Discovered 2026-08-11 while attempting finding 12’s reach analysis, and it reframes every inference above.
Two independent derivations converge. (a) Reverse-engineering the reported percentages. Several results are published to two decimals, which constrains the denominator tightly — 71.88/28.12 admits only n=32 (23–9) or its multiples; 94.55/5.45 only n=55 (52–3); 79.05/20.95 only n=105 (83–22). (b) View counts on the source videos, pulled from YouTube: 163–1,136 per debate, against a channel with 6,360 subscribers. Each inferred n falls at 15–25% of its video’s views — ordinary turnout. The two methods were derived independently and agree.
| Round | Views | Reported | Inferred ballots |
|---|---|---|---|
| Masculinity | 703 | 74.2 / 25.8 | 23 – 8 (n=31) |
| Boomers | 1,136 | 61.7 / 38.3 | 29 – 18 (n=47) |
| Feminism | 591 | 60.5 / 39.5 | 23 – 15 (n=38) |
| AI | 810 | 79.05 / 20.95 | 83 – 22 (n=105) |
| Rich-without-luck | 210 | 71.88 / 28.12 | 23 – 9 (n=32) |
| Too sensitive | 276 | 94.55 / 5.45 | 52 – 3 (n=55) |
Caveat: multiples of each n fit equally well (n=64, 96… for the wealth round), so these are lower bounds; the view-count cross-check is what makes the smallest solutions most likely. Percentages could also be rounded in a way this reconstruction doesn’t model.
What this does to the findings:
⭐ The ballot-count reconstruction, redone for round 2 — and the turnout rate is higher than the obvious guess, because the denominator is wrong (2026-09-03).
Prompted by Chris: another streamer reviewing the data-centers round asked ChatGPT to estimate the vote split, which assumed a 10% conversion on ~589 views and preferred 52–7. Chris’s instinct was that 20% is likelier — “this is a fairly niche event and I think the retention rates are higher than normal.” He is right, and the vault could already have told him so.
First, the 52–7 estimate is circular. 52–7 is n=59, and 59 is exactly 10.0% of 589. The answer is the assumption restated with a plausible-looking split attached; the percentage did no work. What makes the split checkable is that a reported percentage constrains its own denominator — finding 13’s method — and 88.1% is consistent with n = 42, 59, 67, 84, 101, 109, 118, 126, 134, 143 … so 59 is one of at least ten candidates and nothing about the reported figure selects it.
Second, the vault has an empirically-derived turnout band and it is 15–25%, not 10%. Finding 13 got there from two independent directions: the two-decimal round-1 results pin n hard (79.05% is impossible below 105 ballots; 94.55% below 55; 71.88% below 32), and those floors sit at 15–25% of their videos’ views. So Chris’s 20% is inside the measured band and the 10% assumption is below every observed round.
Applying the band to the two published round-2 results:
| Round | Views | Reported | Feasible n in band | Implied split |
|---|---|---|---|---|
| Data centers | ~589 | 88.1 / 11.9 | 101, 109, 118, 126, 134, 143 | 89–126 to 12–17 |
| Deportation | ~740 | 79.3 / 20.7 | 111, 116, 121, 135, 140, 145, 150 | 88–146 to 23–38 |
At Chris’s 20% specifically: data centers ≈ 104–14, deportation ≈ 115–30.
Third — and this is the correction worth keeping — the denominator everyone is using is too big. Turnout is being computed against views now, but the ballot closed 24 or 48 hours after posting while views kept accumulating. The rate that matters is ballots ÷ views at ballot close, and that number is strictly smaller than any later view count. This hub logged the data-centers video at 435 views on 2026-09-01, shortly after its 24-hour ballot had closed; the streamer saw 589 later. Against ~435 rather than 589, an n of 101–143 is a turnout of 23–33%.
So the ordering is: 10% is too low, 20% is about right against late view counts, and the true conversion against the audience that could actually vote is plausibly 25–30%. That is a genuinely engaged niche audience, exactly as Chris supposed — and it also means finding 13’s 15–25% band is itself slightly understated, since it was computed against review-time views for round 1 too.
What it does not change: the absolute numbers stay tiny. Even at the top of the band these rounds are decided by roughly 100–150 people, so finding 13’s central point survives intact — mobilisation still outranks any analysis of debate quality. What improves is the precision: round 2’s pools (n≈100–150) look genuinely larger than round 1’s (n≈31–105), which is finding 11’s growth premise confirmed even as its compression prediction fails. The pools grew; the margins grew too.
⚠ Standing caveats, unchanged from finding 13: multiples of each n fit equally well, so these are lower bounds; the arithmetic assumes ordinary rounding of an exact vote share; and view counts are read at a single moment rather than at ballot close, which is precisely the error this entry corrects for.
Audience size is small and shows no clear trend within round 1. Views ran 703 → 1,136 → 591 → 810 → 210 → 276 → 239 → 163 across Aug 3–10. The release interval is now pinned: the channel has been posting on a 48-hour cycle (see the production note below), so each successive video in that series had two fewer days of exposure than the one before it. It is tempting to read a collapse, but later videos have had far less time to accumulate (the Aug 10 debate had one day; the Aug 3 debate had eight), so the sequence cannot establish a decline. What it does establish is the scale: no round-1 debate has cleared 1,200 views, against a 6,360-subscriber channel.
Subscriber datapoint, 2026-08-17: 6,410 — up ~50 over roughly two weeks of the bracket running. Worth recording because finding 11’s neutral-dilution mechanism requires the tournament to accumulate an audience, and a ~0.8% channel growth across the whole of round 1 is not the order-of-magnitude change finding 13 says would be needed for performance to outweigh a mobilised bloc. Early evidence against the hypothesis, from the one number that tracks it directly.
The ledger’s own calibration is the finding: the argument reads hold up, the ballot reads don’t — and they fail in one direction. With ten blind calls now resolved, the pattern is legible and it is not flattering to the cynical prior this hub started with.
| Resolved ballot call | Outcome |
|---|---|
| Owlish advances on delivery | ✅ |
| Charsky wins round 4 | ✅ |
| Tejada advances | ✅ |
| Calvin wins round 5 | ❌ |
| Hamm advances | ❌ |
| Turner (Aff) wins | ❌ |
| Lawrence (Aff) wins on rhetoric | ❌ |
| Summerhays/Gilkison — abstained | ✅ (the abstention was right; the vote diverged 43 points from the substance read) |
⚠ REWRITTEN 2026-08-18 — this finding was refuted by the next two results, one day after it was written. The claim above was that the misses are one-directional: that every wrong ballot call but Hamm came from expecting rhetoric, sympathy or affiliation to beat argument, and therefore that the fix was to stop discounting argument quality. Then the last two cards posted:
| Round | Chris’s read | Result |
|---|---|---|
| Party loyalty | Joe — better arguments, presented well | Kewl Vic +62.0 |
| Hate speech | McPherson — “had a plan… experience dealing with these types of debates” | Thunder +25.0, closing on “satanic pedophilic overlords” |
Both misses run the opposite way. Here Chris predicted the better arguer and the worse one won — in one case by 62 points, in the other over a closing statement about global enslavement. So the asymmetry claimed above does not exist; with nine calls resolved the misses run in both directions, and “predict the ballot from the argument” would have gone 0-for-2 on its first outing.
The lesson is about the vault’s method, not about the audience. Finding 15 was fitted to seven data points and broke on the eighth and ninth. That is precisely the error finding 13 warns about — small-n noise read as signal — committed here on the vault’s own ledger rather than on the debaters’. A pattern over seven binary outcomes is roughly what chance produces anyway; it should never have been promoted to a “posture going into round 2.” Standing correction: no predictor gets promoted off fewer than ~10 resolved calls, and any predictor that would have been contradicted by the two most recent results is not yet a predictor.
What actually survives, stated conservatively: the analytical reads still hold up — argument-quality verdicts, concession-tracking, layer diagnoses, and load-bearing-word calls have all survived contact with results. What remains unmodelled is the mapping from those verdicts to ballots, and three of nine is not distinguishable from a coin. The honest posture into round 2 is therefore agnostic rather than corrected: run the autopsy over all fourteen at once, and register a model only if it beats the base rate across the whole set rather than across a flattering slice of it.
The bloodsports penalty — a contender’s own diagnosis, and the most promising predictor on the table (2026-08-18).
Chase McPherson, to Chris after his loss: the format penalizes contestants who engage in “bloodsports” tactics.
Sourcing first: this is a defeated contender explaining his own defeat, which is exactly the kind of claim that needs discounting — the self-serving version attributes a loss to a format flaw rather than to performance. It is logged anyway because it is independently supported by material already in the vault, some of it recorded before this round aired, and because McPherson is well-placed to know: he is a practised debater and Chris’s read of him was positive (“had a plan… experience dealing with these types of debates”), so the thesis is not being used to paper over a bad performance.
The convergent evidence, none of it collected with this hypothesis in mind:
But the crude version is false, and the counterexamples locate the real line. Technical moves per se are not punished: Cruz won on a severance and Bourdeau won on a severance, both textbook technical manoeuvres, and Charsky won by 58 with fast, dense, citation-heavy competitive-policy delivery that finding 6 predicted should hurt him. So “the audience punishes technique” is refuted.
Sharpened hypothesis — the audience punishes making the opponent the subject. What separates Joe and McPherson from Cruz, Bourdeau and Charsky is not technicality but target: the winners argued about the resolution, the losers argued about the man across from them. This slots directly into finding 4’s ledger as a third category — every minute is affirming, negating, or wasted, and opponent-talk is wasted and costly, because a room that came to hear a question adjudicated reads it as evading the question. It also explains the counterexample pattern that “bloodsports tactics” cannot.
Standing relative to finding 17: finding 16 is now the junior hypothesis. Chris’s subject-prior model explains 7 of 11 codable rounds directly and the twelfth by affiliation, without appealing to posture at all — and it explains both rounds finding 16 was built on: hate speech fits the prior (a free-speech pool backing the free-speech side), and party loyalty fits affiliation (Vic is the larger channel). So opponent-talk may be doing no independent work. It survives only for the residual — the four rounds where the prior lost — and should be tested there rather than everywhere.
Falsifiable, and cheap to test in the autopsy: code each round for opponent-directed minutes and see whether it predicts the loser across all fourteen. If a debater who spent visibly more time on his opponent than on the resolution has lost every time, that is the first predictor in this hub with a real mechanism behind it. Registered before the coding is done, so it can fail honestly.
⭐ The accepted reductio is the most expensive single error in this format — two specimens, −65.2 and −74.8 (2026-09-07). A pattern that only became visible once two rounds had it.
In both cases the opponent offers a reductio of the debater’s own position; in both cases the debater accepts the characterisation rather than blocking it; and in both cases it becomes the spine of the opponent’s closing.
| Round | The reductio | Accepted by | Margin |
|---|---|---|---|
| Amnesty | “Can 500 million Indians just come here and it still be America?” → “yes… it’s just the name” | Owlish | −65.2 |
| DSA | “The DSA is only a force for good because it elects people who disagree with it” → “Yeah, sure. If you want to characterize it like that” | Bourdeau | −74.8 |
Two of the three largest margins in the bracket are rounds where the losing side accepted a reductio, and in both the block was one sentence and free. Owlish needed “no — every immigration system has a cap; we’re arguing about where it sits, not whether there is one.” Bourdeau needed “no — a platform is a bargaining position, not a legislative agenda; you don’t judge the Republicans by their most maximalist plank either.” Bourdeau reached for exactly that argument eight minutes late and never developed it.
Why it is so much more expensive than an ordinary dropped point: a reductio is designed to be quotable, so accepting one does not merely lose the exchange — it manufactures the opponent’s closing line and hands over the clip. It is the mirror image of finding 6: where a technically-correct move can fail because a lay ballot cannot register it, a reductio is the move a lay ballot registers best, which is exactly why conceding one costs so much.
Diagnostic, and it is a preparation task rather than an in-round one: for every position you intend to defend, write down the sentence an opponent would use to make it sound absurd, and prepare the limiting principle before you are asked for it. Both losers here were asked for a limiting principle and did not have one.
⚠ Two specimens is a pattern, not a law. The test is whether a round exists in which a reductio was blocked, and what that round cost — currently unchecked, and logged as an open question.
⭐⭐⭐ Two judges crossed their own priors in opposite directions on the same round — the counter-specimen to finding 19 (2026-09-14). Chris, on the SAVE Act main event:
“In the same judge group, a left-leaning judge leaned Pearson, while being more right-leaning, I gave it to Mockler. To me that shows how close this actually was. Impressive round.”
Pearson is the right-coded debater and Mockler the left-coded one, so both judges scored against their own side — and the official ballot was the closest of the night, “so close that the winner literally changed and flipped on the final review.” Three independent readings, two of them crossed, all landing near 50/50.
⚠ Finding 19 holds that a reviewer’s own politics supplies the “better position” verdict. This is its cleanest counter-case, and it is stronger than a single crossing would be: one crossing can be contrarianism; two crossings in opposite directions on one round cannot both be.
Chris, spitballing the cause, and the two judges gave different accounts of themselves:
“I think this means we tend to judge our own side too harshly. The other judge said it was probably because he didn’t want to look biased; for me I think it was knowing better responses to the other side and being disappointed that my side didn’t make them.”
| A — bias avoidance (the other judge) | B — informed disappointment (Chris) | |
|---|---|---|
| Kind | motivational — a reputational correction | ⭐ epistemic — asymmetric information, not asymmetric motive |
| Mechanism | discount your own side to appear fair | you hold a higher-resolution map of your own side’s argument space, so you see every argument that was available and unmade |
| ⭐ Testable difference | should weaken when judging privately or anonymously | should survive anonymity, and scale with expertise in your own side’s literature |
| Failure mode | over-correction | ⭐⭐ comparing your side to the steelman and theirs to the transcript |
⭐⭐⭐ B names a real and specific error, and this page committed it all season. “The strongest version of Thunder’s case is real and mostly unmade” is the review method’s standing move — and a steelman is only available for arguments you already know exist. So the method systematically holds the better-known side to a higher standard: its debater is scored against an ideal, the opponent against what they actually said. ⚠ The 4-of-4 steelman versus 2-of-4 forecast split is the same asymmetry showing up as a scoreboard.
⭐ A third candidate neither judge proposed: differential error detection. You catch your own side’s mistakes because you can check their claims; the other side’s claims you cannot check as well, so they pass unchallenged. This produces the crossing with no disappointment and no reputational motive at all — and it is the hardest of the three to introspect, which is a reason to doubt both self-reports.
You judge the side you know better favourably on position and harshly on execution.
Finding 19 is the first half; finding 28 is the second. They look opposite and pull against each other on a single ballot, so which one is visible depends on which is larger — and execution-harshness only wins when the round is close on position and your side underperformed its known best case.
⭐ Which derives Chris’ own caveat rather than merely agreeing with it: “it is still a decent metric that probably only shows up in very close fights.” ⚠ But the distinction matters: that is a property of the detector, not of the phenomenon. In a lopsided round the harshness is the same size and simply is not enough to flip the verdict, so nothing is observed. The crossing is the visible tip; the asymmetry is presumably always there. That makes the metric more useful, not less — a crossing is strong evidence, but its absence is no evidence at all.
⚠⚠ Honest limits, and they are severe. n = 2 judges, one round, with mechanisms self-reported after the fact — and introspective accounts of one’s own judgement are among the least reliable evidence there is. The behaviour (the crossing) is solid; every explanation above is speculation, including the one that reconciles two of this page’s own findings. 📌 The cheap test is available whenever there is a judge group: collect verdicts before anyone states their prior, then collect the priors.
⭐⭐ Which yields a usable quality metric the vault did not have: cross-prior agreement. When judges of opposing priors each award it to the other side’s debater, they are scoring the debate rather than the position — and that is the best available evidence that a round was genuinely balanced and genuinely well argued. 📌 Worth applying retrospectively: finding 5’s “better side loses” specimens should be re-checked for whether the reviewer’s prior and the verdict ever crossed.
⭐⭐ Burden inflation — volunteering a hard premise when an easier one would have done (2026-09-14). Chris, on the final: ”‘Embryos don’t have rights’ — to be fair this is consistent for him, but he doesn’t need it to beat consent, and it let Rex torture him for much too long.”
The mirror of finding 20. An accepted reductio is a hard premise forced on you by your own principle; burden inflation is a hard premise offered unprompted when the argument does not require it. Thunder had the sufficient answer to the consent objection — children cannot consent to chemotherapy; parents decide — which works whether or not an embryo is a person, and he delivered it in his closing where it could not be developed.
⚠ The tell is that it is consistent, which is why it is hard to see as an error. Volunteering it cost a second burden (fetal personhood), a large share of the round’s airtime, and standing with a pool that had already voted on abortion. ⭐ Diagnostic question for future rounds: does this premise appear in the shortest path to my conclusion? If not, holding it is a cost with no return.
⭐⭐⭐ The last two rounds were won by the same argument, delivered by the same move — and neither winner knew it was a thesis (2026-09-14). Topicality appeared in none of the first twelve rounds and then decided both of the last two.
| SF1 — drugs | Final — genetic engineering | |
|---|---|---|
| The technical move | legalization ≠ decriminalization | the cocaine proposition — illegal-with-carve-outs is not legal |
| What it neutralised | the entire sympathetic case (teenagers, dime bags) | the entire evidence base (sickle cell, SMA, hemophilia) |
| The substantive claim underneath | ⭐ the off switch — “that only existed because it was still technically illegal” | ⭐ keep the default prohibitory so carve-outs stay narrow and revocable |
| Winner | Medeiros (lost the ballot) | Jones (won the bracket) |
⭐⭐ Both are the same claim: the enforcement instrument exists only while the conduct is formally illegal — so the default allocation is the argument. That is the liability-default vs. conduct-command distinction, reached twice, independently, by debaters with no exposure to it.
⚠ And it reframes both rounds as default-allocation disputes disguised as yes/no disputes. In SF1 both men agreed about most drugs and fought over opioids; in the final both ended at “some yes, some no” — Chris caught it: “with the cocaine example he showed an instance when it should be legal, even by his own admission.” The winner in each case is the one who named his default and called it a position, rather than arriving at it as a retreat. 📌 Open: is topicality a late-round skill, or a strategy that only pays against an opponent with good evidence?
⭐⭐ The free concession — granting a premise your own evidence has already defeated (2026-09-09). A distinct error type from finding 20’s accepted reductio, and it took Chris to spot it: “the oxycontin example shows ‘clean’ drugs may actually be worse, and there is now incentive to make them more potent.”
In SF1 the Neg’s own opening establishes that Florida’s pill mills had every safety feature legalization promises — FDA drug, licensed prescriber, exact dose, no cartel, no adulterant — and produced the deadliest drug epidemic in American history. Twenty minutes later he concedes “the premise that drugs would be safer is 100% true.” He had already proved that clean supply is not safe supply, and then granted the opposite.
| Accepted reductio (finding 20) | Free concession (this one) | |
|---|---|---|
| What is granted | an absurd consequence of your own principle | a premise of your opponent’s case |
| Why | consistency — the honest debater takes the horn | generosity — it looks like grant-and-outweigh |
| Refuted by | the audience’s intuition | your own evidence, already on the record |
| Cost | catastrophic (−65.2, −74.8) | zero, if the opponent doesn’t notice |
⚠ It earns a separate number because it is invisible from the ballot. A reductio gets punished; a free concession is punished only if the opponent is paying attention, and Thunder was not. It is an error in the transcript that leaves no trace in the result — the first finding here that margin data can never surface, and a standing reason to read the tape even when the card is known.
⭐ The evidentiary double standard is this bracket’s most common losing move — three specimens in three consecutive rounds (2026-09-09). Each time it was the round’s turning point, and each time the opponent noticed.
| Round | The asymmetry | Cost |
|---|---|---|
| QF C | Tejeda claims future benefits and never prices a present cost — “we only have short-term views on this” | Jones: “what about the facts, man? What about what’s happening?” |
| QF D | Charsky holds market failure structural and inescapable, state failure a solvable design problem | The “ethical capitalist” move he ruled out on his own side |
| SF2 | Summerhays demands realized proof of harm, offers logical entailment for benefit — “I don’t need material events to have happened” | Asked for realized benefit: “No, I don’t have it.” / “Not yet.” Then dismisses a documented harm as “an edge case” |
One shape: a demanding standard for the opponent’s side of the ledger and a permissive one for your own. It is attractive because each half is individually defensible — scepticism about speculation is a virtue, and so is reasoning ahead of the data — and it is fatal because the two halves are only compatible if nobody puts them side by side. The counter is one sentence: state the standard once, then apply it to both columns.
⚠ It also explains why these rounds turn on burden rather than on evidence. In all three the loser had more material than the winner on at least one dimension; what beat them was being caught grading their own homework on a different curve. Related to but distinct from finding 19, which is about the reviewer’s standards rather than the debater’s.
⭐ Delivery rate is a fixed trait, and a constant cannot explain a variable — finding 6 narrowed (2026-09-08). Prompted by Chris: “Charsky speaks rapidly in his opening, and it would be fun to count WPM here too.” Measured off the timestamped cues:
| Words | Duration | WPM | |
|---|---|---|---|
| Charsky opening, QF D | 1,088 | 4.90 min | 222.0 |
| Charsky opening, round 2 | 1,083 | 4.87 min | 222.5 |
| Summerhays opening, QF D | 892 | 5.67 min | 157.4 |
Conversation and broadcast delivery run 130–160 WPM; full competitive-policy spreading clears 300. 222 is the top of the intelligible range. But the striking figure is not 222 — it is 222.0 against 222.5: two rounds, two opponents, two topics, and his rate is identical to within half a word per minute across ~1,085 words each. That is a locked-in motor habit, not a read of the room.
Finding 6 has to narrow accordingly. It says his delivery is a technique/format mismatch that costs him under an audience ballot. His three results at a constant 222 WPM are +58.1, +58.6, −9.0, and a constant cannot explain a variable — the prior moves and the delivery does not. So the defensible version is: the mismatch imposes a roughly fixed cost, which a favourable prior swamps and a hostile one does not. A handicap, not a verdict. ⚠ One contender, three rounds; measure a second before rewriting the finding outright.
⭐ There is no house style — two founders moderated round 3 and behave nothing alike (2026-09-08). This retires a hypothesis of Chris’s own, one day after he proposed it, on his own new fact.
The QF B reframe was that a promoter in the chair is a calibration instrument: a guest moderator’s conduct could be idiosyncrasy, but a founder’s shows “how he thinks the series should go.” It was a good hypothesis and it had one data point. Then Chris identified QF C’s moderator, Sam Tripoli, as a second founder — and the two diverge about as far as two moderators can:
| Nathan Paul (QF B) | Sam Tripoli (QF C) | |
|---|---|---|
| Content supplied | his own framing, as “more of a statement” | none |
| Questions | each sympathetic to its recipient; one labelled aloud as “kind of an anti-war question” | symmetric, and one gets turned against the man it was asked of |
| Structure | installs it — forces point-by-point rebuttal | stays out — “he stepped back and let them go” |
Two consequences, and the second is the useful one.
⭐ Two refinements from Chris, and the second one is a steelman this page had not made.
(a) It may not be neglect — it may be unresolved disagreement at the top. “We got 1 founder who likes to interrogate, and we got another who prefers they go at it themselves. This might be an indication that the reason there is no clarity is that the founders don’t even know.” That is a materially different diagnosis from nobody has got around to writing it down: it says the variance is live disagreement between the people who would have to settle it, which predicts the inconsistency persists rather than converging. Testable — if a house style is ever announced, it should look like one founder’s practice and not a blend.
(b) The interrogation round is a genuine format innovation with a real precedent, and this page has been treating it only as a hazard. Chris: “this is a unique segment to add to a debate. I can see in the lower rounds it can help get the debaters on track, but it is interesting to interrogate much like you see presidential debates, so maybe this was the blend they were going with.”
Everything finding 1 has built — attribution, symmetry, bite, matching the room — is a set of constraints on a risk. It is worth saying plainly what the segment is: a deliberate hybrid of two traditions that are normally kept apart. A debate clashes two advocates and the moderator holds the clock; an interview puts a questioner between the audience and the subject. The interrogation round splices them, and the American presidential debate is the obvious model — a format in which the moderator’s questions are half the content and nobody regards that as interference.
That reframes the variance without excusing it. If the segment is a deliberate blend, the founders’ divergence is them tuning a ratio rather than failing to have a rule — and Chris’s own observation about level fits: an interrogation round rescues a weak round (the Boomers round had no spine until Pisco installed one; QF A got its best philosophy from Perez) and intrudes on a strong one (QF C’s debaters “had a better grasp than the moderator”). So the ratio should vary by round quality, which is exactly what “match the intervention to the room” says — and it means the format’s inconsistency and its best feature are the same property seen from two sides.
⭐ The anti-war Neg keeps inflating his best argument into his worst one — twice in one round of the bracket (2026-09-08). Chris: “You don’t have to be against the war and be an antisemite at the same time!” and, on QF C, “the negative pushing that this helps Israel and not the US is a good thread, but then he goes anti-Israel, flattening the point out.”
| Round | The narrow claim — checkable, on-resolution, damaging | The inflated claim — unfalsifiable, off-resolution, dismissible |
|---|---|---|
| QF B | Israel’s interests drove the war’s timing | “Communism is a Jewish ideology”; media control, AIPAC, OnlyFans |
| QF C | “The war is not beneficial to America. The war is beneficial to Israel” — delivered with a factual reversal the Aff cannot answer | “Is Sharia law worse than Talmudic rule?”; “Talmudic rule says they’re all on the left, we have the right to kill them” |
The narrow version is the strongest argument available to the anti-war side: specific, about American interest, exactly on the resolution. The inflated version cannot be checked, changes the subject from American interest to ethnic claims, and hands the opponent a free dismissal — which is what happens both times. This is inflation operating on an argument rather than a term: widen it until no counterexample survives and it stops doing work. Two debaters, two rounds, one resolution, the same self-inflicted wound — which makes it a property of the position’s rhetorical gravity rather than of either man.
⭐ The reviewer’s own politics can silently supply the “better position” verdict — caught in the act, by Chris, on himself (2026-09-07). The single most important caveat on this page, because it applies to a column that appears in every round review and in the scorecard.
This hub records two independent judgments per round: who argued better (technique) and who had the better position (substance). The first is checkable against the tape. The second is not, and it has been treated as though it were neutral. It is not necessarily neutral, and the QF B round produced the demonstration:
Chris: “I guess I was coming at it that a libertarian should be against foreign wars, but it is fair to bring up the point that I judge foreign actions based on the net benefit for that country doing it — and who knows, maybe we can control the strait and the oil and get better off. Ty for reminding me of my tough lesson.”
The mechanism is worth stating precisely, because it is not simple bias. His “Thunder had the better position” was doing double duty: partly a substantive judgment about the Iran war, and partly an inference from a prior about what a libertarian’s position ought to be — an argument from the label rather than from the case. Those feel identical from the inside, and the second is not evidence.
It also interacts badly with the prompt mismatch in that round, which is what surfaced it: on the published prompt (“justified”) the Neg’s position is strong; on the spoken prompt (“beneficial to America”) it is an open empirical question about the Strait of Hormuz, oil, and the terminal cost of an intervention. Reading the Neg as obviously right required implicitly picking the prompt that suited the prior.
What follows for this page, and it is a real cost:
⚠ Registered as a caveat, not a retraction. No prior substance read is withdrawn on this evidence. What changes is that the column now carries a known failure mode and a check.
Chris’s subject-prior model — it outperforms everything else this hub has tried, and it is now the front-runner (2026-08-18).
Chris: “my thoughts were just that it was a combination of popularity (Kewl Vic) and voting for the belief they hold anyway… good rhetoric is a factor as it changes the margins — McPherson was 4th closest when most would disagree with him — but I think people are voting on the subject, and only the debaters if they know them.”
A three-part model, and each part is separately checkable: (a) the ballot mostly tracks the resolution, not the performance; (b) affiliation intrudes only where the voter knows a contender; (c) rhetoric moves the margin, not the winner.
Coded against all twelve decided rounds. “Prior” = the position this specific pool is expected to hold, read off the bracket’s own audience signature (agency-affirming, free-speech, anti-establishment, patriot-coded — finding 10):
| Round | Winner (side) | Pool’s prior favours | Fit |
|---|---|---|---|
| Feminism helping | Ruelas (Neg — not helping) | Neg | ✅ |
| Rich without luck | Aftermath (Neg — luck isn’t required) | Neg | ✅ |
| Too sensitive | Tejada (Aff — we are) | Aff | ✅ |
| Algorithm fair | Cruz (Neg — not fair) | Neg | ✅ |
| Therapy for men | Summerhays (Aff — doesn’t work) | Aff | ✅ |
| Hate speech | Thunder (Aff — protect it) | Aff | ✅ |
| Boomers responsible | Rex Jones (Aff) | Aff (weak — judgment call) | ✅ |
| Masculinity in crisis | Owlish (Con — not in crisis) | Pro | ❌ |
| AI making us dumber | Charsky (Neg — it isn’t) | Aff | ❌ |
| Fragile culture | Bourdeau (Neg — culture isn’t the cause) | Aff | ❌ |
| Military service to vote | Brunet (Neg — no) | Aff | ❌ |
| Party loyalty illogical | Kewl Vic (Neg) | unclear — affiliation | ➖ |
| Therapy culture | Josh Smith (Aff — it makes people weaker) | Aff | ✅ (added 2026-09-01) |
| Feminism rerun | Tareyak (Neg — not helping) | Neg | ✅ (added 2026-09-01) |
Seven of eleven codable rounds, with the twelfth explained by clause (b). Against the vault’s own 3-of-9 ballot record, that is a decisive improvement — and it was reached without watching a single tape, which is itself the point.
Update 2026-09-01 — the two recovered results both fit, taking it to nine of thirteen codable. Therapy culture and the feminism rerun were resolved by the round-of-16 bracket rather than by result cards, and the pool’s prior predicts both: an anti-establishment, agency-affirming audience votes therapy culture makes people weaker and feminism is not helping. This changes nothing about the model’s standing, because these two codings are exactly as post-hoc as the other eleven — the same fit-after-the-fact this finding’s own warning is about, and the reason its test was registered prospectively. What it does change is the comparator: the vault’s ballot record rose to 5-of-11, so the model must now beat 45%, not 33%.
And the four failures are not random: they are the four rounds where the vault recorded the largest execution gap. Owlish’s legibility against Garcia’s private vocabulary; Charsky, the best debater in the bracket; Bourdeau extracting three concessions; Brunet naming the inflation in thirty seconds. So the natural completion is a two-factor model:
Default to the audience’s prior on the resolution. The prior is overridden only by a large execution gap — and rhetoric otherwise moves the margin, not the winner.
That fits 11 or 12 of 12, and it subsumes rather than discards the existing findings: finding 10 supplies the prior’s direction, finding 2 supplies clause (b), and finding 6 supplies what counts as an execution gap.
Clause (c) has direct support, and Chris identified the specimen. McPherson argued a speech-restriction case, as a self-described center-left streamer, to a free-speech pool — the prior was against him about as hard as this bracket allows — and he held it to 25.0 points, the 4th-closest margin of twelve, against a median of 45.6. He lost the round and won the margin, which is exactly what clause (c) predicts.
A structural detail that fell out of the ranking and supports the two-factor split: the margins are bimodal. Four rounds land at 12.4–25.0, then there is a gap with nothing in it, and eight land at 42.8–89.1. A single noisy predictor would smear across the range; two regimes look like prior-aligned blowouts versus rounds where something contested the prior. Worth testing directly — the four close rounds should be the ones where prior and execution point in opposite directions.
⚠ The discipline this finding must obey, given finding 15 died yesterday of exactly this error. The prior codings above were assigned after the results were known, which is precisely how a model earns a fit it hasn’t earned. This is therefore a hypothesis, not a finding, and its test is registered in advance:
Chris falsified his own model within minutes of proposing it, and the objection is better than the model (2026-08-18).
Chris: “the Garcia debate already falsifies me :) It seems the audience leans right, and this would give Garcia the win, but here it seems rhetoric played the bigger part at least in this one. Of course this is a multivariate problem that the variables change each round (different voter block each round).”
The self-falsification is correct and was already coded ❌ above. But the second clause is the deeper point and it changes what kind of object the prior is. Finding 17 as written treats “the pool’s prior” as a fixed parameter of the bracket. It isn’t. At 30–105 ballots drawn largely from the contenders’ own networks, each round convenes a different electorate — so the operative quantity is that round’s pool’s prior, which moves with whoever mobilised. A resolution’s prior and a round’s prior are not the same number.
That is not a refutation so much as a specification of the error term, and it does real work:
Chris’s own framing of why it beat the vault’s model is worth keeping as the summary: the vault has been analysing debaters, and the audience has been voting on topics.
Chris’s standing complaint, and the reason this section exists: “a small pet peeve with the WW format — since it is not real time, they can’t keep the bracket updated, so it is hard to see how things are going.”
Exactly right, and it is structural rather than sloppy. A live event resolves on the night; this one resolves 48 hours after each release, on a staggered schedule, with results published as image cards on a page that carries no bracket. So at any moment the tournament is in a superposition — some rounds decided, some inside their window, some unaired — and nobody outside the organizers can see the standings. The rounds are all we get; the tournament is invisible. Since this hub exists to track the bracket as a unit, maintaining the state here is the fix.
⚠ Two things this section got wrong, and the corrections are the method working. (a) “Nobody outside the organizers can see the standings” was true and is now false in principle: the quarterfinal projection hit 4-for-4, so the bracket is reconstructed and the standings are derivable from the round-of-16 list alone. (b) The ballot window changed to 24 hours at round 2 and the rounds are now live-streamed. Round-1 standing below is unchanged; round 2 is here and round 3 is here.
Round-1 standing (as of 2026-09-01, plus one recovered margin) — round 1 complete, all 16 slots known. Two results were never published; they were recovered by subtraction from the round-of-16 bracket (below). One of the two margins was later recovered from a moderator intro; the other still is not.
| # | Round | Result | Margin | Advances |
|---|---|---|---|---|
| 1 | Masculinity | Owlish 74.2 – 25.8 Garcia | +48.4 | Owlish |
| 2 | Feminism | Ruelas 60.5 – 39.5 Anton | +21.0 | Ruelas |
| 3 | Boomers | Rex Jones 61.7 – 38.3 Meyers | +23.4 | Rex Jones |
| 4 | AI | Charsky 79.05 – 20.95 Bunn | +58.1 | Charsky |
| 5 | Rich without luck | The Aftermath 71.88 – 28.12 Constable | +43.8 | The Aftermath |
| 6 | Too sensitive | Tejada 94.55 – 5.45 Guptill | +89.1 | Tejada |
| 7 | Algorithm | Cruz 56.2 – 43.8 Hamm | +12.4 | Cruz |
| 8 | Therapy for men | Summerhays 71.4 – 28.6 Gilkison | +42.8 | Summerhays |
| 9 | Fragile culture | Bourdeau 73.7 – 26.3 Turner | +47.4 | Bourdeau |
| 10 | Military service | Brunet 78.6 – 21.4 Lawrence | +57.2 | Brunet |
| 11 | Therapy culture | Josh Smith 65% – 35% Ouedrago | +30.0 — recovered 2026-09-03 from a moderator intro | Josh Smith |
| 12 | Feminism rerun | Tareyak def. David S. | never published | Tareyak |
| 13 | Hate speech | Thunder 62.5 – 37.5 McPherson | +25.0 | Octavius Thunder |
| 14 | Party loyalty | Kewl Vic 81 – 19 Kung Fu Joe | +62.0 | Kewl Vic |
| 15 | (never listed) | no opponent on record — bye | — | Medeiros |
| 16 | (never listed) | no opponent on record — bye | — | Lee Horseradish |
Two rounds were never posted at all. Therapy culture (Smith vs. Ouedrago) and the feminism rerun (David S. vs. Tareyak) both closed before the two cards that published on 2026-08-18, and neither card ever appeared. So the pipeline does not merely lag — it publishes out of order and then drops rounds entirely, which rules out “they post as windows close.” Note what this cost: these are the two rounds this hub covered live and predicted blind, and their ballots are permanently unrecoverable — the bracket gives up the winner and nothing else.
What the margins say — this format produces blowouts. Across twelve decided rounds the mean margin is 44.2 points and the median 45.6. Eight of twelve were decided by 40+, and exactly one has been competitive (Cruz/Hamm, +12.4). Read against finding 13 — rounds turning on 30–105 ballots — a 45-point median means a typical round is decided by a dozen or so people breaking one way. This is the baseline finding 11 predicts should compress as the bracket advances: if round 2’s margins don’t shrink, neutral dilution is in trouble.
Three rounds are closed with no card posted, including hate speech, whose window has clearly expired. Given the two never-listed matchups and the round-2 filming already underway, the publication pipeline is running behind the tournament itself — a third production irregularity, consistent with reading the format as loosely administered.
The bracket is the result card. The organizers never published two round-1 results and never listed two matchups, but a list naming the sixteen survivors resolves all four gaps by subtraction: a name in round 2 won its round 1, and a name that isn’t there, didn’t. Defect 4 paying off backwards — the standings page they don’t publish would have been redundant with the bracket they finally did.
The eight pairings, as given (round-1 credentials attached):
| # | Round-of-16 matchup | Round-1 credentials |
|---|---|---|
| 1 | Medeiros vs. Lee Horseradish | neither has an aired round — the two byes drawn against each other |
| 2 | Josh Smith vs. Rowan Bourdeau | Smith def. Ouedrago (margin never published) · Bourdeau +47.4 |
| 3 | Chris Brunet vs. Tareyak | Brunet +57.2 · Tareyak def. David S. (margin never published) |
| 4 | Octavius Thunder vs. Kewl Vic | Thunder +25.0 · Kewl Vic +62.0 |
| 5 | Rex Jones vs. Owlish | Jones +23.4 · Owlish +48.4 |
| 6 | The Aftermath vs. Jose Tejada | Aftermath +43.8 · Tejada +89.1 |
| 7 | Silvio Cruz vs. Spencer Summerhays | Cruz +12.4 · Summerhays +42.8 |
| 8 | Martae Ruelas vs. Luke Charsky | Ruelas +21.0 · Charsky +58.1 |
The audition cross-reference was half right, which is the useful half. On 2026-08-18 this page cross-referenced the channel’s contender auditions against the known field and named Horseradish as the strong candidate for a missing slot. Confirmed. But it found one of two — Medeiros never surfaced by that method at all. So the auditions are a lead generator, not a census, and any future “who’s left” reconstruction must assume it under-counts.
The field was 30, not 32. Fourteen aired matchups = 28 contenders, plus two who advanced with no opponent on record = 30 names, against branding that says 32. Two readings, not equally likely:
/vote-now-1 never showed a ballot for either.Either way the consequential fact is the same: two of sixteen round-2 slots were filled without a single vote being cast. Against finding 13 — rounds decided by 30–105 ballots — two contenders reached the round of 16 at a price of zero, in a bracket where everyone else paid a few dozen.
Number the round-1 rounds by the order they aired (the standings table). The round-of-16 pairs are then:
{1,3} {2,4} · {5,6} {7,8} · {9,11} {10,12} · {13,14} {15,16}
Alternating blocks of four. In blocks 1 (rounds 1–4) and 3 (rounds 9–12) the bracket pairs odd-with-odd and even-with-even; in blocks 2 and 4 it pairs adjacents. That is not noise — it means release order interleaved two bracket slots within a block, then stopped doing so for the next block.
Two things follow, and both are worth more than the curiosity:
If the round-of-16 list is in bracket order and quarterfinals pair adjacent matchups (the standard reading, and the one the block structure supports), the QF is:
| QF | Feeds from |
|---|---|
| A | (Medeiros / Horseradish) vs. (Smith / Bourdeau) |
| B | (Brunet / Tareyak) vs. (Thunder / Kewl Vic) |
| C | (Jones / Owlish) vs. (Aftermath / Tejada) |
| D | (Cruz / Summerhays) vs. (Ruelas / Charsky) |
A falsifiable structural prediction, registered 2026-09-01, and it costs nothing to check when the QF is announced. If it holds, the bracket is fully reconstructed and every later round is predictable in shape before it is published.
It also pays off both matchups this hub said on 2026-08-18 that it wanted:
⚠ Superseded within hours of being written (2026-09-01). Three round-2 matches had already aired with resolutions and side assignments published in their video descriptions, so the model can be coded and now has been. The reasoning below is kept because it is the protocol that was followed, not because the blocker still stands.
Finding 17’s model is resolution-level: it predicts from the pool’s prior on the proposition, overridden only by a large execution gap. At the time of writing, no round-2 resolutions, side assignments, dates or moderators had been published — only the pairings — so the model had nothing to code. Anything registered today would be a person-level reputation guess, which is a different model, and reporting it as the same one would contaminate the test.
So the protocol is two-stage, and both stages must be dated in the ledger:
One bar moved, and it has to be said out loud rather than quietly re-set. The success test was pre-registered as beating the vault’s 3-of-9 ballot record. With Smith and Tareyak resolved, that record is now 5 of 11 (45%). The registered number stands as registered — moving a bar after seeing results is precisely the error finding 15 died of — but the honest comparator has risen, so a model clearing 3-of-7 while the vault’s own hit rate sits at 45% has not demonstrated anything. Report both numbers when the round resolves.
Most of this list closed between 2026-09-01 and 2026-09-07; struck items are kept rather than deleted so the closes stay visible.
⚠ Round-3 view pace, computed properly (Chris, 2026-09-07) — and it retracts what this page said hours earlier. The first version compared raw counts and read QF B as badly underperforming. Chris’s correction: QF A’s result is already published, so its 24-hour ballot has closed and YouTube’s “1 day ago” bounds it at 24–48 hours. Rate, not count, is the comparable quantity:
| Round | Views | Age | Views/hr |
|---|---|---|---|
| C Tejeda vs. Rex Jones | 2,600 | 23h | 113.0 |
| D Summerhays vs. Charsky | 1,600 | ~36h | 44.4 |
| B Tareyak vs. Thunder | 184 | 9h | 20.4 |
| A Medeiros vs. Bourdeau | 550 | ~36h | 15.3 |
Two corrections follow, and both cut against what this page believed this morning.
⭐ Scored on 2026-09-07, since round 2 is over and the answers are in. The first bullet’s own test resolved against the compression hypothesis on the aggregate and for it on the best single observation — worked through here — but the second half of that bullet, the clause-(c) prediction about which rounds land narrow, is the part that came good: the bracket’s three narrowest rounds are all rounds where prior and execution (or two audience axes) pull opposite ways. Best current read: the model predicts the margin band better than it predicts the winner, which is the reverse of how it was pitched. On format polish, two defects moved in round 2 — side assignments are now published (defect 1, fixed) and the ballot window shortened to 24 hours — while the closing-order defect got worse, since a third moderator convention appeared that is the exact inverse of the first.
Three round-2 matches are already posted, and the channel now publishes the resolution and the side assignment in the video description — which, if it holds, quietly fixes defect 1. That is everything finding 17’s model needs, so it is coded here before any transcript is read, which is the condition the finding pre-committed to.
| # | Match | Resolution as published | Aff | Neg | Aired |
|---|---|---|---|---|---|
| 1 | Medeiros vs. Horseradish | “America should build more data centers.” | Medeiros | Horseradish | 2026-08-31 · 1:11:43 |
| 6 | Tejeda vs. The Aftermath | “Women should have the right to vote.” | The Aftermath | Jose Tejeda | 2026-08-31 · 1:05:21 |
| 8 | Ruelas vs. Charsky | “Mass deportation is bad.” | Charsky (against deportation) | Ruelas (pro deportation) | 2026-08-31 · 1:11:27 |
Name correction: the channel bills him Jose Tejeda, not “Tejada” as this hub has had it since round 1. The channel’s spelling is adopted going forward.
Round 2 changed the format in five visible ways, all confirmed on the data-centers tape: the ballot window is 24 hours, not 48; rounds are live-streamed rather than posted as finished video; side assignments are published in the description (defect 1, fixed); there is a named sponsor read three times per round; and contenders appear to get ~12 hours’ notice of the resolution (the Neg says so on air) — which, if bracket-wide, reframes every “he didn’t know his topic” reading in the round-1 reviews.
⚠ The schedule explains the pipeline, and it came from Chris (2026-09-07): the live final is this weekend, 2026-09-12. So rounds 2, 3 and 4 were compressed into roughly twelve days to make the date. That reframes several things this page has been reading as sloppiness: the 48→24-hour ballot window, results posting late and out of order, matches airing before their predecessors’ cards, and — most consequentially — round 3 airing while round 2’s coverage was still in progress. It is schedule pressure against a fixed live event, not indifference. It also means finding 11’s compression test is confounded a third way: later rounds had less time between them, so a voter’s exposure window shrank as the bracket advanced. And it puts a clock on everything registered here — the semifinals and final must all air inside five days.
Two more facts recovered alongside the resolutions, neither previously in the vault: the championship prize is $5,000 (stated in every round-2 description), and round 2 is being run as live streams with live chat, where round 1 was posted as finished video. Both bear on turnout: a live audience and a named prize are different mobilisation conditions than round 1 had, which is a confound on finding 11’s dilution test that did not exist when the test was designed.
Model under test (finding 17): default to the audience’s prior on the resolution; override only on a large execution gap; rhetoric moves the margin, not the winner. The pool’s signature is agency-affirming, free-speech, anti-establishment, patriot-coded (finding 10).
| # | Prior favours | Predicted winner | Predicted margin band | Scored? |
|---|---|---|---|---|
| 6 | Aff — near-unanimous on any audience | The Aftermath | wide (≥ 40) if the prior dominates; narrow if affiliation compresses it | ✅ scored |
| 8 | Neg — this pool is patriot-coded | Martae Ruelas | narrow (≤ 25) | ✅ scored |
| 1 | split — the model is silent | no call (weak Neg lean, unscored) | narrow | ❌ excluded, as pre-registered |
⚠ UPDATE, same day — the round collapsed and the test changed shape. The tape opens with the affirmative declaring he is playing devil’s advocate and that he and his opponent “basically have identical worldviews.” Neither man defends the resolution: the Aff argues repeal is socially impossible (then proposes restricting the franchise), the Neg severs should from must and brackets repeal as “a whole separate topic.” The pre-tape call below stands exactly as filed — the model sees only resolution and side assignment, and weakening a registered condition after seeing the tape is the error finding 15 died of. But a declared devil’s advocate conceding his own side is the largest execution gap the bracket can produce, so the two forms of the model now split and the round becomes a discriminator rather than a single test: prior-only → Aftermath (filed pre-tape); two-factor → Tejeda (filed post-opening, labelled as such, and not claimable as a pre-tape prediction). One of the two dies whichever way the ballot goes.
Match 6 is the best test in the bracket, because it puts the model’s own two clauses against each other. Clause (a) says the ballot tracks the resolution, and “women should have the right to vote” has about as lopsided a prior as a proposition can have — for this pool or any other. Clause (b) says affiliation intrudes where voters know a contender, and Tejeda carries the largest mobilised bloc in the bracket (+89.1 in round 1, the biggest margin on the board) — drawn onto the indefensible side. So the two clauses point in opposite directions, and the round adjudicates between them:
Match 8 is the override clause’s test. The prior points at Ruelas, but Charsky is the bracket’s best-argued winner and the one contender who has already beaten legibility and prior simultaneously (+58.1 vs. Bunn). The prior-only call is filed as Ruelas so the prediction is falsifiable; a Charsky win is then not a refutation but a second instance of the override, and should be scored as such rather than quietly counted as a hit. Either way, clause (c) predicts this lands in the narrow mode — prior and execution are pulling apart.
Match 1 is excluded, and the exclusion was registered before the topics were known. No round-1 data on either contender, and the resolution splits the pool’s signature down the middle: anti-Big-Tech, anti-eminent-domain and rising-electricity-bill populism pull Neg, while pro-growth, anti-regulation and beat-China-on-AI pull Aff. A model whose default is “the prior” has no default here. That makes it the round where execution is most visible, which is worth watching even though it cannot score — see the round page.
Round 2 is over and round 3 has aired. Five of the round’s eight cards are published; the other three resolve by subtraction from the round-3 field, which is the bracket-as-result-card method paying off a second time. Cards are read off rendered images, as ever — the retrieval note still applies.
| Match | Result | Margin | Advances | Source |
|---|---|---|---|---|
| 1 | Medeiros 88.1 – 11.9 Horseradish | +76.2 | Medeiros | card |
| 2 | Bourdeau def. Josh Smith | never published | Rowan Bourdeau | round-3 field |
| 3 | (no debate) — bye | — | Tareyak | round-3 field |
| 4 | Octavius Thunder 58.4 – 41.6 Kewl Vic | +16.8 | Octavius Thunder | card |
| 5 | Rex Jones 82.6 – 17.4 Owlish | +65.2 | Rex Jones | card |
| 6 | Jose Tejeda 81.6 – 18.4 The Aftermath | +63.2 | Jose Tejeda | card |
| 7 | Summerhays def. Cruz | never published | Spencer Summerhays | round-3 field |
| 8 | Charsky 79.3 – 20.7 Ruelas | +58.6 | Luke Charsky | card |
⚠ A second bye, and it cost the bracket its best specimen. Match 3 — Chris Brunet vs. Tareyak — was never held. No video exists on the channel or on the PeerTube mirror, no ballot ever opened, and Tareyak appears in round 3. So Brunet is out of the tournament without losing a round. That is the contender the hub called “argument quality’s best specimen” — the man who beat rhetoric and audience prior simultaneously by 57 points and retired finding 15 by doing it. The bracket’s cleanest instrument for the question this whole page is about left the field without a vote being cast.
Running total: four of the tournament’s 16 round-2 slots and quarterfinal berths have now been decided by withdrawal rather than ballot (Medeiros and Horseradish in round 1, Tareyak in round 2), in a format where contested rounds turn on 30–105 ballots.
The pre-registration named match 6 as worth more than the other seven combined, and wrote its own death certificate in advance:
“Tejeda wins → finding 17 is dead. No re-coding of the prior will save it; a pool that votes down women’s suffrage is not a pool voting its priors on the resolution.”
Tejeda won 81.6 – 18.4. The card was published after all — this hub recorded it on 2026-09-03 as “closed, never published,” which was true of that day’s results page and is no longer true. The prior-only model is dead by its own registered condition, and the post-opening two-factor call (→ Tejeda) hits. Both were filed with their dates and their status labels, so the discrimination is clean:
| Model | Match 6 call | Filed | Outcome |
|---|---|---|---|
| Prior-only — the ballot tracks the resolution | The Aftermath | pre-tape | ❌ wrong by 63.2 |
| Two-factor — prior, overridden by a large execution gap | Jose Tejeda | post-opening (labelled) | ✅ correct |
What actually happened is worth stating plainly, because it is not “the audience opposes women’s suffrage.” The affirmative opened by announcing he was playing devil’s advocate and that the two men “basically have identical worldviews”; neither defended the resolution. So the ballot was never given a proposition to vote on, and it voted on the contenders instead — which is the two-factor model’s override clause with the largest execution gap the format can generate. The prior-only model died because it cannot see that the round had no proposition in it.
The pre-registration filed prior-only → Ruelas on match 8, while naming Charsky as “the single most likely override candidate in the bracket,” and specified the handling in advance: a Charsky win is not a refutation but a second instance of the two-factor override, and should be scored as such rather than quietly counted as a hit.
| Model | Match 8 call | Outcome |
|---|---|---|
| Prior-only (the ballot tracks the resolution) | Ruelas | ❌ wrong by 58.6 |
| Two-factor (prior, overridden by a large execution gap) | Charsky | ✅ correct |
So finding 17 survives in its two-factor form and its one-factor form is dead. That is the first time this hub has run a genuine pre-registered discrimination and had it come back clean. The override clause is now carrying two specimens — Charsky vs. Bunn (+58.1) and Charsky vs. Ruelas (+58.6) — and both are the same man, which is a real limitation on how much it establishes.
Still-unpublished margins: two. Bourdeau/Smith (match 2) and Summerhays/Cruz (match 7) give up winners and nothing else. Both are candidates for the moderator-intro retrieval route — check the round-3 intros for Bourdeau and for Summerhays, both of whom appear in quarterfinals A and D.
Match 4 is the round-2 tape this hub picked up last, and it is the one that costs the most to explain. “Abortion should be legal nationwide” in front of a patriot-coded, anti-establishment, right-leaning pool: every coding of the audience signature on this page points at the Neg, and the Neg is Kewl Vic, who had just won round 1 by +62.0. The affirmative won by 16.8.
The repair is not to abandon the prior. It is to notice that the “signature” is a bundle of at least two axes that can be put against each other, and the coding has never had to choose between them:
| Axis | Content | Result when they collide |
|---|---|---|
| Anti-establishment / bodily autonomy | don’t let the government decide what you do with your body; COVID-mandate memory | wins |
| Social-conservative / traditionalist | life begins at conception; traditional marriage; demographic replacement | loses |
The collision happens on tape, in one exchange: the Aff asks “Did you take the COVID vac?”, the Neg answers honestly that he opposed the mandate, and the frame is gone. So the operative rule going forward is code the axis before coding the prior — and where a resolution loads both, the anti-government axis is the one with an observed win.
That directly changes what the model says about round 3, where two of the four resolutions are about war and the anti-establishment axis has a pure expression (non-intervention) available to one side in each. It also retroactively puts the Iran round’s miss in play as a genuine anomaly rather than an execution override: the anti-interventionist axis was coded and lost. See the round-3 registration, which turns that into a designed test rather than a footnote.
Caveat, stated rather than buried: the alternative explanation is plain reach. The abortion tape drew 3,495 views against a round-2 ceiling of 822 and a round-1 ceiling of ~1,136, and Thunder’s register is this pool’s native one. Nothing in the remaining bracket separates the two readings.
Finding 11 predicted that margins compress as the bracket advances, because pools grow, both contenders have already cleared a round, and the neutral fraction rises. The test as registered: “if round 2’s margins don’t shrink, neutral dilution is in trouble.”
| Round 1 (n=13) | Round 2 (n=5) | Round 3 (n=2, partial) | |
|---|---|---|---|
| Mean margin | 43.1 | 56.0 | 41.9 |
| Median margin | 43.8 | 63.2 | — |
| Range | 12.4 – 89.1 | 16.8 – 76.2 | 9.0 – 74.8 |
The aggregate refutes it and the sharpest single observation supports it, which is an honest split rather than a wash.
⭐ Resolved 2026-09-07, and the answer is that the test was wrong, not the hypothesis. Finding 11’s mechanism is about pool size — more voters, more neutrals, margin pulled toward 50/50. Every test on this page has used round number as the proxy, and round number is a bad proxy because pool size varies enormously within a round: round 3 alone spans 184 to 2,600 views. Plotted against the variable the mechanism actually names:
| Round | Views | Margin |
|---|---|---|
| Biden/Trump* | 392 | +44.0 |
| Data centers | 435 | +76.2 |
| QF A — DSA | 550 | +74.8 |
| Deportation | 740 | +58.6 |
| Amnesty | 822 | +65.2 |
| QF D — Nationalism/Socialism | 1,600 | +9.0 |
| Abortion | 3,495 | +16.8 |
Pearson r = −0.74 on raw views, −0.77 on log views (n = 7). The three largest margins in rounds 2–3 sit on the three smallest audiences; the two narrowest sit on the two largest. Neutral dilution is confirmed on the variable it was always about, and the earlier verdict — “the mechanism was probably backwards” — is withdrawn.
⚠ Two caveats that keep this honest. (a) n = 7, and one margin (Biden/Trump) is a spoken figure carrying ±6. (b) A confound with the same signature: a lopsided or unengaging topic plausibly depresses both viewership and competitiveness, so topic quality could drive both variables without any dilution mechanism at all. The two readings are not separable from this data. What is settled is that the round-number test was measuring the wrong thing, and that finding 11 should never have been declared to be in trouble on it.
A better hypothesis than either, and it is testable. The three narrowest rounds in the tournament — Cruz/Hamm +12.4, Thunder/Vic +16.8, Summerhays/Charsky +9.0 — are not the three biggest pools. They are the three rounds where the pool’s prior is split or its axes conflict: an unfamiliar proposition, two audience axes pulling opposite ways, and a strong arguer drawn against a hostile prior. That is finding 17 clause (c) — prior decides the winner, execution and axis-conflict decide the margin — and it fits every narrow round on the board while dilution fits one. Registered prediction: margin band is predictable from axis-conflict, and pool size adds nothing once axis-conflict is coded.
Confounds, logged when they appeared rather than now: round 2 introduced live-streaming, a named $5,000 prize, and — as recorded — an inconsistent ballot window (24h for match 1, 48h for match 8). The window difference sits across the only two data points, so the comparison is contaminated at the source.
Josh Smith’s round-1 margin is recovered. This hub recorded it as “permanently unrecoverable” because no result card was ever posted — but the Biden/Trump round opens with the moderator crediting Smith with 65% of the audience, and Bourdeau with 73% (matching the published 73.7, which corroborates the channel). So therapy culture was Smith +30.0, and the round-1 margin set goes from n=12 to n=13.
This is a general retrieval method, not a one-off: the round-2 intros summarise each contender’s round-1 result, so a later round’s introduction can recover an earlier unpublished margin. Standing action — check the round-3 intros for Bourdeau and Summerhays, which would recover the two remaining unpublished round-2 margins, and for Tareyak, which would recover the feminism rerun. Caveat: intros round to whole percents and are spoken, so recovered figures are lower-precision than a card and should be marked as such.
⭐ The standing action paid twice, on the same intro (2026-09-07). Quarterfinal B’s moderator opens by summarising both contenders’ previous rounds, and it yields:
⚠ And finding 18 resolves it, which the first version of this entry failed to check before re-deriving the turnout rate from scratch. The measured band is 15–25% against late views, 25–30% against views at ballot close. Sibling round-2 tapes grew 16–22% over the four days after their ballots shut, so the abortion tape stood at roughly 2,100–2,600 views at close — giving n ≈ 525–780. That brackets the margin reading (~834) and puts the count reading (240) below every round the vault has measured. Arithmetic and turnout now agree: ~830 ballots, a 140-vote margin.
Which makes finding 13’s headline two rounds stale. “Rounds are decided by 30–105 ballots” is a round-1 statement; round 2 ran at n ≈ 100–150, and this figure puts the abortion match near 830. Pools are growing by roughly an order of magnitude across the bracket — finding 11’s growth premise confirmed again while its compression prediction keeps failing. The consequence survives regardless: a swing of ~70 people took a quarterfinal berth in a $5,000 tournament. Standing action — log every raw vote figure a moderator says aloud.
⚠ Third payoff, and a six-point error that downgrades the whole method (2026-09-07). Quarterfinal A’s intro yields two figures and one of them is wrong:
| Intro figure | Actual | Status |
|---|---|---|
| Bourdeau “won with 72% of the votes” on Biden vs. Trump | never published | ⭐ Recovers round-2 match 2: Bourdeau +44.0 |
| Medeiros “got 82%” on data centers | 88.1% (published card) | ❌ off by 6.1 |
⚠ Corrected 2026-09-08 — the blanket downgrade below was an over-correction. Quarterfinal C’s intro adds two more checkable figures and both land: Rex Jones at “83%” against a published 82.6, and Tejeda at “81%” against 81.6. That takes the record to six accurate within ±0.6 and one wrong by 6.1. So the honest statement is not ±6: spoken figures are good to ±1 about six times in seven, with roughly a one-in-seven chance of being badly wrong — a usable instrument with a known failure rate. The three margins resting on spoken figures alone (Smith, Tareyak, Bourdeau) carry that caveat rather than a flat ±6 band.
That is a bigger deal than the recovery. This route was logged as reliable to ±0.5 on four checkable figures that all landed. Here the same moderator misses by six points on a card published four days earlier — the single largest margin in the tournament she could have quoted. So the three margins that exist only as spoken figures — Smith +30.0 (“65%”), Tareyak +50.0 (“75%”) and Bourdeau +44.0 (“72%”) — carry a ±6 band, not ±0.5, and should be read as spoken estimates rather than results. They are currently doing work in the margin distribution and in finding 11’s compression test; the direction of that test does not change, but the precision does.
⭐ CLOSED 2026-09-08 — and it is the model’s worst miss. QF D’s intro credits Summerhays with “92% of the votes” on the Iran round, so round-2 match 7 was Summerhays 92 – 8 Cruz, +84.0. Every round-1 and round-2 margin in this bracket is now known — and four of the last five came from spoken intros rather than published cards.
The registered call there was Cruz, narrow (≤ 25), on an anti-interventionist prior: wrong on the winner and wrong on the band by roughly sixty points. ⚠ It also bears directly on the Iran double-header, whose two open calls both back the anti-war Neg. A +84 for the pro-war side is the strongest evidence available that the coded axis has the sign wrong — unless it is Summerhays’ reach, which is Chris’s reading. The two unread cards separate them.
(Perez’s fourth checkable figure also lands: Charsky at “79%” against a published 79.3. Running record: seven accurate within ±0.6, one wrong by 6.1.)
Corroborated a second time, and the rounding is confirmed. The abortion round opens with Perez crediting Thunder with “63% of the votes” (published: 62.5) and Kewl Vic with “a whopping 81%” (published: 81.0). Both land inside a point, both round to whole percents exactly as flagged. The channel’s spoken figures are reliable to ±0.5.
Logged as a reuse-over-rebuild failure, because that is what it was. This session hit the data-centers page’s recorded blockers — yt-dlp and curl cannot reach youtube.com from this machine (TLS handshake failure on every request, channel listing included), and an in-page fetch of the caption track’s baseUrl returns zero bytes — and built a new route instead of checking what the vault already had. tools/fetch-youtube-transcript.py was sitting there and works: run against this round’s video it returned 1,552 cues through 62:14 on the first try. It drives headless Chrome over QUIC via CDP and lets the real player mint the POT token, which is precisely the thing a naive in-page fetch lacks.
So the standing guidance is unchanged: py -3 tools/fetch-youtube-transcript.py "<url-or-id>" "<out>" is the route. The zero-byte caption response is not a network block and not evidence that captions don’t exist — it is a missing POT token, exactly as the tool’s own notes say.
Two things are genuinely new and worth keeping.
transcript-segment-view-model elements — not virtualised, so a full hour is present at once with no scrolling. To move the text out of the browser, POST it to a throwaway local HTTP listener on 127.0.0.1; YouTube’s CSP permits that, while both clipboard routes fail (execCommand('copy') returns false, navigator.clipboard.writeText throws “document is not focused”). Lower fidelity than the tool — the panel merges cues, 511 against 1,552 — and worth having only when CDP is unavailable.vid.samtripoli.com re-hosts the WWD rounds, and /api/v1/videos?count=25&sort=-publishedAt plus /api/v1/videos/<shortUUID> return titles, full descriptions, durations, publish timestamps and view counts as machine-readable JSON — which is to say the resolutions and side assignments, the exact input the model needs before a tape is watched, with no browser at all. It reaches this machine when YouTube does not. ⚠ It is a partial mirror (three of round 3’s four matches, none of round 1), so it is a fast first look and never a census.One unexpected dividend: the two captures cross-check each other. Every quotation in the abortion review appears in both the panel scrape and the tool’s independent caption fetch. That is a second oracle on the transcript rather than a single ASR pass taken on trust — cheap to repeat whenever a round turns on an exact wording.
The only thing still requiring a rendered browser is the results page, which remains images with no percentages anywhere in the markup.
Three further round-2 tapes are up, so six of eight have aired. Resolutions and side assignments again come from the video descriptions.
| # | Match | Resolution | Aff / for | Neg / against | Views |
|---|---|---|---|---|---|
| 2 | Bourdeau vs. Josh Smith | “Biden was a better president than Trump.” | Bourdeau | Smith | 392 |
| 5 | Rex Jones vs. Owlish | “Amnesty vs. Mass Deportation” | Jones — for mass deportation | Owlish — for amnesty | 822 |
| 7 | Summerhays vs. Cruz | “The Iran war is in America’s best interest.” | Summerhays | Cruz | 262 |
Filed under the surviving two-factor model — default to the pool’s prior on the resolution; override only on a large execution gap. Prior direction read off the bracket’s audience signature: patriot-coded, anti-establishment, agency-affirming, right-leaning.
| # | Prior favours | Predicted winner | Margin band | Basis |
|---|---|---|---|---|
| 5 | Jones (mass deportation) — strongly | Rex Jones | wide (≥ 45) | Pure prior play; the pool just gave a pro-deportation position 20.7% against Charsky, but that reads as an execution override rather than a prior reversal |
| 2 | Smith (anti-Biden) — strongly | Josh Smith | wide (≥ 45) | ⚠ post-opening, not pre-tape — openings were read before filing, so this is labelled accordingly and is the weaker registration |
| 7 | Cruz (war not in US interest) — moderately | Silvio Cruz | narrow (≤ 25) | The anti-establishment/anti-interventionist prior points at Cruz. But Cruz advanced on the bracket’s only competitive margin (+12.4) while Summerhays won by +42.8, so execution points the other way and the two factors conflict |
Match 7 is the one worth watching, because prior and execution point in opposite directions — exactly the condition finding 17’s clause (c) says produces a narrow margin. It is also the round-2 pairing with the weakest mandates on both sides.
Match 7 is the one worth watching, because prior and execution point in opposite directions — exactly the condition finding 17’s clause (c) says produces a narrow margin. It is also the round-2 pairing with the weakest mandates on both sides.
⚠ Registration hygiene. Matches 5 and 7 are filed pre-tape, sight unseen. Match 2 is filed post-opening and is not claimable as a pre-tape call — the openings had already been read when the resolution was identified. Scored separately.
| # | Model call | Chris’s call | Result | Model | Chris |
|---|---|---|---|---|---|
| 5 | Rex Jones, wide (≥45), pre-tape | Rex Jones, post-tape | Jones +65.2 | ✅ hit, and the band held | ✅ |
| 2 | Josh Smith, wide, post-opening | same | Bourdeau advances (margin unpublished) | ❌ | ❌ |
| 7 | Cruz, narrow (≤25), pre-tape | Summerhays, post-tape | Summerhays advances (margin unpublished) | ❌ | ✅ |
Two prior misses in one wave, and they share a shape: both were resolutions where the anti-establishment axis was coded as decisive and the ballot went the other way. Read alongside the abortion round’s reversal, the pattern is that this pool’s axes are not reliably ranked, and the model has been assuming they are. That is the single most important thing round 2 produced, and it is what the round-3 registration is built to test.
And the standing bar, restated so it cannot drift: the pre-registered success test was beat the vault’s own 3-of-9 ballot record on round 2. Round 2 is now complete, so it can be scored. Six of the eight matches carried a registered model call (match 3 was a bye; match 1 was pre-excluded):
| Model | Registered round-2 record | Hits | Misses |
|---|---|---|---|
| Two-factor (the surviving form) | 3 – 2 (60%) | 5 Jones · 6 Tejeda · 8 Charsky | 2 Bourdeau · 7 Summerhays |
| Prior-only (dead) | 0 – 2 | — | 6 Aftermath · 8 Ruelas |
⚠ Three caveats that stop this being a victory lap.
Against the honest comparator: the vault’s own ballot record moves from 7-of-13 to 9 of 16 (56%) with wave 2 scored (Jones ✅, Summerhays ✅, Smith ❌). So a 60% model against a 56% forecaster is not a demonstration of anything yet. Round 3 is where it has to separate.
On 2026-09-01 this page registered the quarterfinal shape before any of it was announced, on the reasoning that the round-of-16 list was in bracket order and that quarterfinals pair adjacent matchups. All four pairings are exactly as registered:
| QF | Registered as | Actual |
|---|---|---|
| A | (Medeiros / Horseradish) vs. (Smith / Bourdeau) | Medeiros vs. Bourdeau ✅ |
| B | (Brunet / Tareyak) vs. (Thunder / Kewl Vic) | Tareyak vs. Thunder ✅ |
| C | (Jones / Owlish) vs. (Aftermath / Tejeda) | Rex Jones vs. Jose Tejeda ✅ |
| D | (Cruz / Summerhays) vs. (Ruelas / Charsky) | Summerhays vs. Charsky ✅ |
So the bracket is fully reconstructed, the tree was fixed at seeding, and every remaining round is predictable in shape before it is published. The round-3 field is also what settled three round-2 results the organizers never published — the bracket is the result card for the second round running.
⚠ Both matchups this hub said it wanted are now impossible. On 2026-08-18 the page named two rounds as the most informative the bracket could produce, and the projection noted each was one or two wins away. Brunet vs. Thunder died when Brunet’s round-2 match became a bye and he left the tournament without losing. Charsky vs. Tejeda died in QF D. The bracket has now eliminated, without a contested loss between them, the two contenders whose rounds carried the most signal about whether argument quality does anything.
Resolutions and side assignments from the video descriptions; the channel has kept publishing them, so defect 1 stays fixed. Ballot window is 24 hours throughout.
| QF | Match | Resolution as published | Aff | Neg | Views | Status |
|---|---|---|---|---|---|---|
| A | Medeiros vs. Bourdeau | “The DSA is a force for good in America.” | Bourdeau | Medeiros | 550 | ✅ Medeiros 87.4 – 12.6 (+74.8) · reviewed |
| D | Summerhays vs. Charsky | “Nationalism vs. Socialism” ⚠ not a pair of opposites | Summerhays — nationalism | Charsky — socialism | 1.6K | ✅ Summerhays 54.5 – 45.5 (+9.0) · reviewed |
| C | Tejeda vs. Rex Jones | “War with Iran is beneficial for America.” | Jose Tejeda | Rex Jones | 2.6K (113/hr) | ⏳ reviewed blind — card deliberately not read |
| B | Tareyak vs. Thunder | “The Iran war is justified.” ⚠ spoken on air as “beneficial to America” | Tareyak | Octavius Thunder | 148 | ⏳ voting · reviewed |
⚠⚠ Correction (2026-09-07, from the QF B tape): the two Iran resolutions may be the same resolution. The published prompt for QF B is “the Iran war is justified”, but the moderator states it on air as “war with Iran is beneficial to America” — which is QF C’s published prompt almost verbatim. Defect 9 firing again, and this time it lands on a registered test: the “moral vs. prudential burden” reading below survives only on the published text, and on the spoken text the double-header is the same proposition twice. That makes the controlled experiment tighter and for an entirely defective reason. The registered calls are unaffected — both were filed on side assignment and axis, not on the burden distinction.
⚠ Two of the four resolutions are nearly the same question, and Chris flagged it as a disappointment. “The Iran war is justified” and “war with Iran is beneficial for America” are a moral claim and a prudential one — genuinely different burdens — but they will read as one topic to a viewer watching both, and a bracket of four quarterfinals spending half of itself on Iran is a thin slate. It is almost certainly a symptom of the compressed schedule: four topics had to be produced fast. The accident is that it handed this page a controlled experiment it could never have arranged.
⚠ Unscoreable and labelled as such — the card (+9.0) was published before this page saw the lineup, so nothing below is a prediction. It is registered anyway because a pre-tape read of the prompt has twice been the most useful thing this hub produced about a round (data centers, the feminism rerun), and the review is still to come.
Chris, before watching: “2 good debaters on a decent topic — I am not sure nationalism and socialism are opposites!!”
They are not, and the defect is structural. Nationalism answers what is the relevant political community? Socialism answers who should own productive property? Orthogonal axes, occupants in all four quadrants; the real opposite of nationalism is internationalism, and of socialism, capitalism. So each contender was assigned a position on a different axis and told to collide — now logged as defect 14.
Three failure modes this predicts, in descending likelihood:
What the result already tells us, and it is the interesting part. The pool’s prior on this pairing is about as lopsided as the bracket can produce: finding 10’s signature is patriot-coded and anti-establishment, so nationalism is close to a free win and socialism is the most hostile side anyone has been dealt. Summerhays won by +9.0 — the narrowest margin in the entire tournament. So either Charsky closed a near-unwinnable gap by execution, which is clause (c) at maximum amplitude, or the prompt’s incoherence prevented the prior from attaching to anything. The review’s job is to say which.
⚠ Hygiene: A and D are unscoreable. Both cards were published before this page saw the lineup, so no call on either is claimable and neither earns the model anything. They are read below as evidence, not as tests.
Two things A and D establish anyway:
Both ballots are open. Neither transcript has been read and neither result exists. Filed under the surviving two-factor model, with the axis correction applied — code the axis the resolution loads, then the prior on that axis, then override only on a large execution gap.
| QF | Prior favours | Predicted winner | Margin band | Basis |
|---|---|---|---|---|
| C | Rex Jones (anti-war) — strongly | Rex Jones | wide (≥ 40) | Prior, affiliation and current form all point the same way. Jones is Alex Jones’s son and the anti-interventionist axis is the Infowars audience’s purest position; the tape has 2.6K views, second-largest in the bracket, which reads as his network turning out. Tejeda carries the bracket’s other large bloc (+89.1, then +63.2), so the two blocs partly cancel and the prior is what is left |
| B | Octavius Thunder (anti-war) — moderately | Octavius Thunder | narrow (≤ 30) | Same axis, weaker on every dimension. Thunder’s register — anti-establishment, Epstein-file, “satanic overlords” — is the anti-war position in this pool’s dialect, and he has ridden it to two wins. But he has won by +25.0 and +16.8, the two narrowest wins of any advancing contender, so the band is set low. Tareyak has the opposite problem: he won round 1 on straightforwardness and then sat out round 2 on a bye |
🎯 Why these two together are the best test on the board — the Iran double-header. Round 3 runs the same proposition twice, in the same week, with the pro-war side as the affirmative in both. That is a controlled experiment the format has never offered before, and the hub did not have to engineer it:
🎯 A second hypothesis this pair tests for free — does a bye cost you the next round? If reach is really mobilisation rather than raw audience (finding 2’s sub-mechanism), then a contender who skipped a round has a cold list and should underperform his form. Tareyak is the first clean case: he advanced on a bye, has not been on screen since round 1, and his QF tape has 148 views against Tejeda/Jones’s 2,600. Medeiros does not test it — his round-1 bye was against Horseradish, who also had one. Registered: the bye is a liability, not a gift, and QF B is where it shows.
Reported by Chris on 2026-09-07, after the quarterfinal A and D cards were known — so those two are held reads reported against himself, not filed predictions, and score nothing. The B and C halves are live and unresolved, and one of them disagrees with the model.
| QF | Chris’s pick | His reasoning | Outcome |
|---|---|---|---|
| A | Bourdeau | round-2 form | ❌ Medeiros +74.8 |
| D | Charsky | “he is the strongest” | ❌ Summerhays +9.0 |
| C | Tejeda | biggest demonstrated bloc, two rounds running | ⏳ open — disagrees with the model |
| B | Thunder | “Tareyak really hasn’t been tested, and he still beat Vic despite a poor performance” | ⏳ open — agrees with the model, by a different route |
His onward reads, now moot: Bourdeau beats Thunder (“Thunder is the weakest of the final 4”) and Charsky beats Tejeda.
⭐ What the two misses actually show — the cleanest quantification of clause (c) this bracket has produced. Chris’s model here is form: who has been performing. It went 0-for-2, and look at what the two picks had in common — both were drawn onto the most hostile prior this pool can generate, and neither reading required a judgment call:
| QF | Chris’s pick | Side he had to defend | Prior | Margin |
|---|---|---|---|---|
| D | Charsky | socialism | maximally hostile | −9.0 |
| A | Bourdeau | the DSA is a force for good in America | maximally hostile | −74.8 |
Same condition, and the margins differ by 65.8 points. That is exactly the shape finding 17 clause (c) asserts — the prior picks the winner, execution moves the margin — and it puts a number on it for the first time: in this bracket, on a hostile prior, execution appears to be worth something like sixty-odd points of margin and zero points of outcome. Charsky, the best arguer in the field, converted his entire edge into losing narrowly.
Read that against Charsky’s own three-round line — +58.1, +58.6, −9.0 — and the picture is consistent: he does not lose his ability to argue between rounds, he loses the prior, and the sign follows the prior every time.
⚠ Three things that stop this being cleaner than it is.
Model: Rex Jones, wide. Chris: Tejeda. Both filed pre-result. This is the first time the two routes have pointed at different contenders on an open round, and the routes are clean:
And Chris has the stronger precedent, which is worth saying plainly since it cuts against the model. The suffrage round was this exact test: Tejeda drawn against a near-unanimous prior. He won by 63. If that generalises, the model is about to lose the same way twice.
The model’s answer — that the suffrage round was collapsed, with the Aff declaring himself a devil’s advocate so the ballot never faced the proposition at all — is a real distinction and also exactly the kind of after-the-fact rescue that should make anyone suspicious. So the round is a genuine discriminator: if Tejeda wins a contested round against a hostile prior, “the prior picks the winner” is finished, and the bloc is the whole story.
Both open quarterfinals are decided, and the answer comes from a source that did not exist yesterday. The /view-results page no longer carries result cards at all — it now contains exactly two images: the site logo and a Final Four bracket graphic. Every one of the ~19 percentage cards this hub has been reading since August, round 1 included, is gone.
| Result | Source | |
|---|---|---|
| QF B — Tareyak vs. Thunder | Octavius Thunder advances | bracket |
| QF C — Tejeda vs. Rex Jones | Rex Jones advances | bracket |
Final Four: Medeiros · Thunder · Rex Jones · Summerhays.
This is a format finding with teeth. The margins for QF B and QF C were never published and now cannot be, and every previously published card has been removed with them. Had this hub not transcribed each card at the time it appeared, the entire margin distribution — the instrument behind findings 11, 13, 17 and 18 — would have been lost. It survives only because it was copied out on the day.
Promoted to defect 5, which already said results post late, out of order, and sometimes not at all. The stronger statement: the page has no memory. Standing rule for the rest of this tournament and any future one — transcribe every card the day it appears; the source is not durable.
| Call | Filed | Result | Verdict |
|---|---|---|---|
| Model → Thunder (QF B), narrow (≤30) | pre-tape, pre-result | Thunder ✅ | winner ✅ · band untestable — no margin published |
| Chris → Thunder (QF B) | pre-result | Thunder ✅ | ✅ |
| Model → Rex Jones (QF C), wide (≥40) | pre-tape, pre-result | Jones ✅ | winner ✅ · band untestable |
| Chris → Tejeda (QF C) | pre-result | Jones | ❌ |
| Model → QF C margin under +40 (dilution) | blind, 2026-09-08 | — | ⚠ unresolvable |
⭐ The Iran double-header resolved, and it resolved cleanly. The test was registered as: both Negs win → the pool’s Iran prior is anti-war; both Affs win → the model has had the sign wrong; a split → the axis buys nothing. Both anti-war Negs won. So the coded axis is right, and Summerhays’ +84.0 in round 2 was reach and execution, not prior — which is exactly how Chris read it (“Spencer has the bigger reach”). The round-2 Iran miss is therefore an execution override, not a sign error, and finding 17’s axis coding survives its hardest test.
⚠ And the page’s self-contradiction escapes judgement on a technicality, which is the least satisfying possible outcome. Jones wide (≥40) and Jones under +40 were both filed; both needed a margin; no margin exists. The winner is right on both, and the discriminator between finding 17 and finding 11 that this round was going to settle is dead — killed by the organisers deleting the evidence. Logged as unresolved rather than quietly dropped.
Chris’s two-stage popularity test, leg 1: ✅. “If Thunder wins this, this is a popularity contest.” Thunder won. Leg 2 is now live and the projection puts him against Medeiros.
The projection registered before announcement had SF1 = Medeiros vs. (Tareyak/Thunder) and SF2 = (Tejeda/Jones) vs. Summerhays. The live ballot reads:
REX JONES (Negative) vs. SPENCER SUMMERHAYS (Affirmative) — “Flock cameras: beneficial or harmful?”
SF2 exactly as projected, which makes the structural model 4-for-4 on quarterfinals and 1-for-1 on semifinals. The bracket has been fully solved from the round-of-16 list since 2026-09-01.
⚠ And the ballot window has been cut again — to SIX HOURS. Round 1 ran 48; round 2 halved it to 24; the semifinal runs 6. Against finding 11’s confirmed dilution result — smaller pool, wider margin — the format has now compressed the ballot by 8× heading into its most consequential rounds. Defect 12 upgraded accordingly: this is no longer a scheduling inconvenience, it is a mechanical push toward blowouts at the sharp end of the tournament.
⭐ Note the resolution, too: “Flock cameras: beneficial or harmful?” is a surveillance topic — and it is the first prompt in the bracket that cleanly loads the anti-establishment axis for both men, since neither Jones nor Summerhays has an obvious pro-surveillance side to defend. Worth watching whether that produces the verbal-dispute failure (finding 8) rather than a contest.
| SF | Match | Resolution | Aff | Neg | Status |
|---|---|---|---|---|---|
| 2 | Summerhays vs. Rex Jones | “Flock cameras are beneficial to Americans, not harmful.” | Summerhays — beneficial, opens first | Rex Jones — harmful | ✅ Jones 62.8 – 37.2 (+25.6) · 662 views at 20h (33.1/hr) · reviewed blind |
| 1 | Thunder vs. Medeiros | “All drugs should be legal.” | Thunder — for, opens first | Medeiros — against | ✅ Thunder 52.7 – 47.3 (+5.4) — bracket minimum · 325 views at 16h (20.3/hr) · reviewed blind |
Both exactly as projected on 2026-09-01. Only one Final Four tape exists so far; SF1 has not been posted.
| SF | Predicted winner | Margin | Basis |
|---|---|---|---|
| 2 | Rex Jones | narrow | ⭐ “Flock cameras are beneficial” is the worst side anyone has drawn since Charsky got socialism. The pool’s axis is anti-establishment and anti-surveillance; Summerhays is defending mass ALPR surveillance to an Infowars-adjacent room. Jones also carries the better-debater read from both Chris and this page, the network, and a win over the bracket’s largest bloc. Prior + execution + form aligned — the same configuration that produced QF C. Narrow because it is the round-3-and-beyond tape with the largest pool (115 views/hr) and dilution compresses with pool size |
| 1 | Medeiros | wide | ⚠ Filed without a topic, so the axis is uncoded and this is an execution-and-reach call only — weaker than a normal filing and labelled so. Medeiros is the best audience-facing prosecutor reviewed and is 2-0 by +76.2 and +74.8; Thunder is 3-0 on the three narrowest wins of any advancing contender (+25.0, +16.8, and QF B unknown). Wide because both men have low reach, so the pool should be small — and small pools produce big margins |
⚠ Chris’s two-stage test has a confound that is now visible. Leg 2 was “Thunder goes down easily when faced against a similar-reach opponent.” But Medeiros has the lowest reach on the board, and Thunder’s one big audience — the abortion round’s 3,495 — came with the topic, not with him: his QF B ran at roughly 11 views/hour. So SF1 is two low-reach contenders, and a Thunder loss there would be evidence about execution, not audience size. The test’s mechanism does not apply to the round it landed on.
SF1’s resolution is “All drugs should be legal,” with Thunder on the affirmative. The registered call was Medeiros, wide, filed explicitly without a topic and labelled as an execution-and-reach call only. The topic is now known and the axis is codeable, and it points the other way.
“All drugs should be legal” is the bodily-autonomy / anti-government-overreach axis in its purest available form — and that is the axis finding 10’s correction says beats every other strand of this pool’s signature. It is also Thunder’s own axis: he won the abortion round on it against a hostile social-conservative prior, and his opening here is the same case structure — “humans should have full body autonomy… it is our God-given right to ingest these substances” — including the same pre-emptive line-drawing move (cyanide is legal, it is in apple seeds; suicide is covered too).
The call is not revised — moving a registration after new information is the error finding 15 died of. What this does instead is turn SF1 into an accidental discriminator between two of the hub’s own instruments, in the same way QF C was:
| Instrument | Says | |
|---|---|---|
| Execution + reach (the filed call) | Medeiros — the best audience-facing prosecutor reviewed, 2-0 by 75+ | ⏳ |
| Axis model (finding 10, as corrected) | Thunder — the anti-government axis, run by the man who already won on it | ⏳ |
A Medeiros win says execution beats the axis on a topic the axis owns. A Thunder win confirms the axis correction a second time and costs the execution call. Either way something learns.
📌 And Chris files a content prediction on SF1, before the tape (2026-09-09): “It will [be] interesting how the bans will be defended — predicting crime/homeless and religion.”
✅ Scored 2026-09-09 — 3 of 3. Crime ✅ heavily (public disorder, a 96-study meta-analysis linking cocaine to partner violence, stimulant violence in the home, car crashes). Homelessness ✅ but moderator-sourced — Binkley raises it first and Medeiros then makes it central. Religion ✅ — “religion did make a showing, but it wasn’t a main point” (Chris, correcting this hub, which had first scored it ❌ absent).
⭐ The religion item is the most interesting of the three, because it enters from the wrong side. The Affirmative grounds his case in it — “we have a god-given right to use all plants,” Genesis 1:29 — and the Neg, who says “I know the Bible very well,” takes the ground away with it (“in the same Bible you’re quoting, God tells Israel to outlaw all shamans who make drugs”), forcing “Correct.” to “I asked you to prove that, and you couldn’t.” That demotes the Aff’s whole case from a right to a stated preference, at minute eighteen. The Neg then uses religion twice as his own answer — Christian ethics as the reason alcohol is the legal exception, offered again when the moderator asks him for the strongest argument against his own position. So the prediction’s reasoning was right as well as its content: a prohibitionist boxed off libertarian ground reaches for public-order harms and moral grounds, and reaches for the second exactly where consequentialism runs out. The model under-predicted his range (topicality, a natural experiment, a reversibility argument) but not his floor.
A different species of call from the ones in the ledger — it forecasts which arguments appear, not who wins, which makes it checkable against the transcript rather than against a ballot. It is also a sharp test of the axis model from the other side: if Medeiros is boxed onto the anti-autonomy side of this pool’s dominant axis, the question is what he reaches for instead, and Chris’s answer is public-order harms and moral/religious grounds — the two places a prohibitionist can stand once libertarian ground is conceded. Scored on the tape when SF1 is reviewed.
⭐ And Chris frames SF1 as a consistency test on the audience, which is sharper than either call (2026-09-09):
“Interesting thing is we now go from ‘I don’t like cameras because they restrict us’ to ‘I don’t care if drug laws are restricting, we are better off with them’… this one serves as an example of if we can take liberty too far. What does Uncle Ben think?”
The same pool votes both rounds inside 24 hours, and the two resolutions pull opposite ways on the liberty axis. Finding 10’s correction says the anti-establishment / bodily-autonomy strand dominates this pool — if that is right, it should be anti-camera and pro-legalisation, and the two ballots should agree. If it goes anti-camera and pro-prohibition, the axis is not dominant but selective, and the operative principle is narrower than the finding claims — something closer to “anti-establishment when the state is watching me, pro-order when the subject is other people’s behaviour.”
The Franklin test makes it concrete. Rex Jones quoted “those who would give up essential liberty to purchase a little temporary safety deserve neither” twice in SF2, against the cameras. Does the same room apply Uncle Ben to drug prohibition? A pool that cites Franklin against surveillance and ignores him on prohibition is telling you the quotation is a weapon, not a premise — which is a finding about the audience rather than about either contender, and it is checkable across the two cards.
Posted overnight, with margins, exactly as Chris said. The final is Rex Jones vs. Octavius Thunder, live in Nashville on 12 September.
| SF | Result | Margin | Registered call | Score |
|---|---|---|---|---|
| 2 — flock cameras | Jones 62.8 – 37.2 Summerhays | +25.6 | Jones, narrow (≤ 25) | ✅ winner · ❌ band by 0.6 points |
| 1 — all drugs legal | Thunder 52.7 – 47.3 Medeiros | +5.4 | Medeiros, wide (≥ 45) | ❌ wrong winner, wrong band |
⚠ N1 also resolves: ❌. The result arrived as a results card, not as a card announcement — the page was rebuilt overnight with a championship graphic and two Final Four cards. And the page is still a current-state display: yesterday’s bracket image is gone, and the QF B and QF C margins did not come back. Defect 5 confirmed a third time — those two numbers are permanently lost.
SF1 was set up on 2026-09-09 as an explicit contest between two of this page’s own instruments, with the call deliberately not revised after the topic was published:
| Instrument | Said | Outcome |
|---|---|---|
| Execution + reach (the filed call) | Medeiros — best audience-facing prosecutor in the bracket, 2-0 by 75+ | ❌ |
| Axis model (finding 10, as corrected) | Thunder — the anti-government axis, run by the man who already won on it | ✅ |
⭐⭐ The axis model beat the execution call on the round the axis owned, and it did so against the best debater in the tournament. This page’s own review of SF1 concluded “Medeiros, and it is not close.” He lost.
This is the second confirmation of finding 10’s correction and the strongest specimen of finding 5 — the better-argued side losing — in fourteen rounds. It also vindicates the discipline: the call was left standing after new information made it look wrong, which is the only reason the test was clean. Revising it would have bought a hit and destroyed the evidence.
⚠ And it costs the execution instrument badly. “Best prosecutor in the bracket” was not a lazy read — Medeiros ran topicality, a natural experiment, grant-and-outweigh, and a reversibility argument nobody else attempted. He was better and it did not matter. Any future call that rests on execution alone should now carry an explicit discount.
Chris framed the two semifinals as a test of the room, not the contenders:
“We now go from ‘I don’t like cameras because they restrict us’ to ‘I don’t care if drug laws are restricting, we are better off with them’… what does Uncle Ben think?”
Answer: the room is consistent, and emphatically. It voted anti-surveillance (Jones +25.6) and pro-legalisation (Thunder +5.4) inside the same six-hour window. The Franklin quotation Jones deployed twice against the cameras was not a weapon of convenience — the same pool applied it to prohibition. ✅ Anti-establishment/bodily-autonomy is confirmed as this pool’s dominant axis for the third and fourth consecutive rounds, and the SF1/SF2 mirror resolves in favour of liberty on both readings of enforcement capability.
SF1 and SF2 aired the same day, ran the same six-hour window, and faced the same electorate. Nothing else in the tournament is this clean:
| Pool | Margin | |
|---|---|---|
| SF1 — drugs | 325 views @16h (20.3/hr) | +5.4 |
| SF2 — cameras | 662 views @20h (33.1/hr) | +25.6 |
The bigger pool produced the wider margin — the opposite of finding 11’s direction (r = −0.74, n=7). ⚠ n=2, but perfectly controlled, which is worth more than two more uncontrolled points. Finding 11 should now be treated as unresolved rather than confirmed, and re-run at n=9.
⭐ The hypothesis it was losing to explains both rounds instead. Registered 2026-09-07: “the narrowest rounds are the ones where the pool’s prior is split or its axes conflict.” SF1’s axes were documented as conflicting the day before the tape — this page said so in writing — and it produced +5.4, the narrowest margin in the tournament, displacing Summerhays/Charsky’s +9.0. SF2’s axis was clean (anti-surveillance to an anti-establishment room) and it produced a comfortable but not blowout +25.6, with Summerhays’ execution plausibly absorbing the rest. ✅ Fourth confirmation, and the first one predicted in advance.
Rex Jones vs. Octavius Thunder, 12 September, Woolworth Theatre, Nashville — the undercard of WWD II.
| Call | Confidence |
|---|---|
| Rex Jones, narrow | ⚠ Weak, and labelled so. No topic published, so the axis cannot be coded — which means this is another execution-and-reach call, filed immediately after that instrument lost SF1. It is the only instrument available, not a good one |
⚠⚠ The larger caveat, and it may invalidate the whole ledger for this round: the electorate probably changes. Every instrument on this page is calibrated to an online ballot of a niche podcast audience. A live theatre crowd in Nashville — who paid, travelled, and are watching three matches — is a different pool, and defect 16 already established that a changed window is a changed electorate. A changed venue is a bigger change than a changed window.
⭐ And reach may invert. Thunder carries the tournament’s largest online audience (3,495 on the abortion round, ~4× anything Jones has drawn) — an advantage that means one thing for an online ballot and something quite different for a room of ticket-holders. Whether the ballot is even cast by the room is unknown.
Live from the Woolworth Theatre, 12 September. Full review. ⚠ No margin was published, and view-results now 404s with the voting section deleted from the site — defect 17 in its terminal form. The result was recovered from the tape: the announcement is a 23-second gap, but the stage interview and the closing recap (“You know who won the first one? There’s Rex Jones”) both confirm it.
| Call | Result |
|---|---|
| Post-topic: Rex Jones, narrow — on axis-conflict + framing burden | ✅ winner correct; margin unpublished, band unscoreable |
| Pre-topic: Jones, narrow — on execution + reach, then self-refuted by its own condition | ⚠ see below — the condition was wrong, not the call |
| Margin narrow / possible bracket minimum | ⏳ permanently unscoreable |
| N2 — four semifinalists travel, losers fill the undercard | ❌ the undercard was Angels vs. Demons (James Williams def. Dr. Heather Lynn) |
| N3 — the Contender final is not the main event | ✅ and understated: it was the opening debate; the main event was Mockler vs. Pearson on the SAVE Act |
⚠⚠ The most useful failure is the self-refutation, and it was mis-specified. I filed “if the resolution lands on liberty-versus-state ground, treat Jones-narrow as already refuted” — on the reasoning that such ground makes the axis model point at Thunder. It landed there, I honoured the condition, and Thunder lost anyway. ⭐ So the naive axis read was wrong and the refined one written the same day — the axis splits, the disgust default sits with the Neg, and Thunder carries the framing burden — was right. The lesson is not that registering conditions is bad; it is that a condition is a prediction too, and this one was worse than the analysis it was meant to discipline.
Filed pre-tape as “what Rex has beyond icky” — all four appeared:
| Steelman argument | |
|---|---|
| Consent of the edited party | ✅ his lead, anchored on the Nuremberg Code, Article One |
| Irreversibility / germline permanence | ✅ the gene pool, “a thousand generations” |
| Positional compulsion | ✅ “if your child doesn’t get the shot… you’re going to end up a second-class citizen” |
| The weighting problem | ✅ verbatim — “Who are you to say what’s better and what’s worse?” |
The content predictions did much worse:
| Prediction | |
|---|---|
| Thunder supplies no limiting principle | ✅✅ and it decides the round — “I don’t have a line right now” |
| Thunder does not run “you can’t put it back in the bottle” | ✅ never makes it; gets adjacent via China and does not convert |
| Thunder runs pure bodily autonomy for a 4th round | ❌ miss, and a credit to him — he ran a regulated-technology case instead |
| Jones does not run positional compulsion | ❌ miss, and it was my most confident content call |
⭐⭐ The page predicted which arguments were good far better than it predicted which would be made. That asymmetry is worth keeping: steelmanning is a different and apparently more reliable instrument than forecasting behaviour, and this hub has been spending most of its effort on the weaker one.
Medeiros opened SF1 with legalization ≠ decriminalization. Jones won the final with the same shape — his “cocaine proposition,” deployed six times: cocaine is illegal yet has medical carve-outs, therefore illegal-with-carve-outs is not legal, and every gene therapy the Affirmative cites becomes evidence for the Negative.
⭐ It appeared in none of the first twelve rounds and then decided both of the last two. Both users were late-round contenders facing opponents with strong concrete evidence, and in both cases topicality neutralised that evidence without contesting a single fact. 📌 Open: is this a skill that only survives selection, or a strategy that only pays when the opponent’s evidence is good?
⚠ And the loser’s opponent named the winning line aloud, on stage: “I would have taken the position that you just support somatic. That’s what I would have done.” Somatic-only is defensible, matches the existing-therapy evidence, and dodges every consent and gene-pool argument — and it concedes the resolution as written, which is presumably why Thunder refused it.
The margin series ends here. No figure for the final, none for either semifinal beyond the two captured on 2026-09-10, and the two quarterfinal margins lost on 2026-09-08 are not coming back. ⚠ Finding 11 is therefore permanently unresolvable on this dataset, and the axes-conflict hypothesis keeps its four confirmations without a fifth.
⚠ One variable arrived with no precedent and no way to measure it: voting stayed open through two subsequent debates while the commentary desk praised Jones on air, and the desk itself noted that slots later in the night “get a shorter window to get the votes in.” Defect 16 — the window is the electorate — now applies within a single event.
Off the WWD X account, via Chris: “Genetic Engineering on Humans Should Be Legal.” Thunder affirmative, Rex Jones negative. Live, Woolworth Theatre, Nashville, 12 September.
Filed yesterday, before the topic existed: “If the resolution lands on liberty-versus-state ground, the call is wrong on its own logic and should be treated as already refuted.”
It landed on liberty-versus-state ground. “X should be legal,” Thunder affirmative — the identical construction to “All drugs should be legal.” So the filed call of Jones, narrow is refuted by its own stated condition, and it stays on the ledger as refuted. No reinterpretation of “axis-favourable” after seeing the topic; that is the exact move SF1 was kept clean by refusing.
| Round | Resolution | Thunder’s side |
|---|---|---|
| R2 | abortion | bodily autonomy |
| SF1 | “All drugs should be legal” | Aff |
| Final | “Genetic engineering on humans should be legal” | Aff |
Three rounds, one axis, always affirmative, always “the state may not forbid what you do with your body.” ⭐ This is the strongest evidence yet for the prompt-negotiation hypothesis — a contender does not land on his own axis three times running by lottery. Chris’s “they rattle off topics until they find disagreement” predicts exactly this: the topic search terminates wherever Thunder’s opponent will take the other side of Thunder’s standing position.
Every prior round had the pool’s signature pointing one way. This one splits it, because two strands of the same anti-establishment signature give opposite answers:
| Strand | Reads the topic as | Points at |
|---|---|---|
| Bodily autonomy / anti-prohibition | the state banning what you may do with your own reproduction | Thunder |
| ⭐ Anti-technocracy — anti-pharma, anti-WEF, anti-expert, naturalist, religious | laboratories and corporations redesigning human beings | Jones |
⚠ And the default reading is Jones’s. Chris: “the gut reaction is usually disgust for the idea.” The topic is not a “my body” question until somebody makes it one — it arrives sounding like labs doing things to babies, not like the government in my bedroom. So the round is a framing fight, and whoever wins the framing wins the ballot.
⭐ Which is the worst possible round for Thunder’s documented flaw. He has now twice found the correct frame and let go of it inside a minute — “is it financially beneficial for me to rob an old lady?” in the quarterfinal, and the closed-course answer in SF1. A round decided by holding a frame is a round decided by the one thing he does not do.
📌 By the axes-conflict hypothesis, a split axis predicts the narrowest margins. SF1 holds the record at +5.4 on a conflicted axis. This one is more conflicted.
| Call | Basis |
|---|---|
| Rex Jones, narrow (≤ 25) | ⚠ Same winner as the refuted call, reached by a different route — and that must be stated rather than hidden. The original was execution-and-reach. This one is axis-conflict plus framing burden: the disgust default sits with Jones, Thunder must convert the topic before he can argue it, and converting-and-holding is his documented weakness |
| Margin: narrow, and possibly a new bracket minimum | the most internally conflicted axis in fourteen rounds |
Content calls, filed before the tape (the species Chris scored 3-of-3 on SF1):
“I usually argue that humans naturally ‘genetically engineer’ with mate selection, so the principle itself isn’t absurd. I also like to describe the movie GATTACA as evidence that it isn’t as bad as people think… the selectors don’t always get it right, and even people without selection can be competitive.”
Mate selection is a non-uniqueness argument, and it is now Chris’s signature move — third time in three rounds (DSA, drugs, here). It works against the strong claim — selecting for traits in offspring is inherently wrong — because humans have always done it and nobody calls mate choice eugenics.
⚠ But it under-delivers against the claim actually on the table, and the gap is worth having before the tape:
| Mate selection | Germline editing | |
|---|---|---|
| Who decides | distributed across millions | centralised — a clinic, a protocol, a fashion |
| Speed | generations | one generation |
| Direction | undirected; no one is optimising | directed toward a chosen target |
| Consent of the selector | both parties choose each other | a third party chooses for the edited |
| Error correction | continuous, cheap | ⭐ none — the change is in the germ line |
So the analogy licenses the principle and not the mechanism. ⭐ And note the shape: the right-hand column is centralised, fast, directed, and uncorrectable — which is the common-law-versus-legislation distinction applied to the genome. Mate selection is nomos; germline editing is thesis. That is a much stronger version of Chris’s own point than “it’s natural,” because it says why the distributed version is safer without needing nature to be a value.
On GATTACA: ⭐ the inversion is genuinely clever — the canonical anti-engineering film is set in a world where the unengineered man wins, so the standard citation undercuts the standard fear. ⚠ But it is fiction, so it is evidence about what people fear, not about what happens. Its real force is rhetorical and reductive: it denies the other side its own best-known illustration. Worth deploying exactly that way and not as data.
“In GATTACA the parents do the selection and it is consensual… I don’t think Thunder has to argue for state-mandated eugenics with the state deciding outcomes. What if the parents just go to a clinic and pick from a menu? This still contains all of the evolutionary pressures that exist if it was just randomly determined by nature. Lots of parents might choose blue eyes, but I bet a fair amount of parents choose brown.”
⚠ This is right, and it breaks a row of the table above. That table listed germline editing as “centralised — a clinic, a protocol”, which conflated two different things:
| Who supplies the technology | Who chooses the target | |
|---|---|---|
| Mate selection | — | distributed |
| Germline editing via a clinic menu | centralised | ⭐ also distributed — one couple at a time |
| State eugenics programme | centralised | centralised |
A menu at a clinic is much closer to a market than to a plan, and my nomos/thesis label was wrong on that axis: millions of uncoordinated parents with heterogeneous preferences is a price system for traits, not a central plan. Chris’s version is the stronger one, and the correction stands.
⭐ Two residues survive it, though, and they are where the real argument is:
1. The menu is the plan. Parents choose from the menu; they do not write it. Which traits are on offer is decided by clinics, regulators and reimbursement — a small, centralised, professionally-captured committee — and distributed choice over a centrally curated option set is not a market. ⭐ It is an election with a candidate list. So the weighting problem is relocated to the supply side, not dissolved — and that is precisely where the filter-versus-quota test starts to matter.
2. ⭐⭐ Distributed choice only preserves diversity when preferences are uncorrelated. Chris’s blue-eyes/brown-eyes case is right for neutral traits, where taste genuinely varies. It fails exactly where selection bites: height, IQ, disease resistance, freedom from disability — traits with a perceived universal ranking, where parental preferences are not diverse but strongly correlated.
⚠ And this has already been run as an experiment. Sex-selective abortion in India and China produced badly skewed sex ratios entirely from distributed, private, parental choices — no state mandate required. Distributed + correlated = coordination without a coordinator, which is the aggregation-versus-sorting failure mode: choice that normally sorts instead aggregates onto one answer.
“I don’t think evolution can be broken that easily. Just because we can choose traits doesn’t mean we select properly.”
Half of this is a strong defence and half of it is the objection. He is right that recombination keeps regenerating variance, that edits at a few loci are a small perturbation on a vast genome, and that selection feedback is slow — so “we will engineer ourselves into a corner” is overstated.
⚠ But “we don’t select properly” is not reassurance — it is the argument, once you add correlation. Random variation is not noise that evolution tolerates; it is a diversification strategy. Evolution does not optimise, it maintains a portfolio, and the variance is what pays out when the environment moves. Correlated directed selection concentrates the portfolio: a population tuned to the last generation’s idea of better, with the variance that would have handled the next one deliberately removed. The failure mode is monoculture, and it is fully naturalistic — no disgust required.
⭐ So the honest form of the worry is a speed mismatch, not a break: editing acts in one generation, and the feedback that would tell you it was a mistake arrives over many. You find out you were wrong after it is already in the pool — which is the off-switch argument again, arrived at from population genetics instead of from law.
The sharpest thing in this exchange is buried in Chris’s own concession, and it is bad news for the affirmative case rather than good:
“I agree there is a level of positional compulsion here, and that was also a key theme in the movie.”
⭐⭐ GATTACA is not a film about state eugenics. There is no law in it compelling anyone. Vincent is not arrested; he is denied insurance and denied employment. The coercion is private, emergent, and produced by exactly the distributed voluntary choices Chris is defending — actuaries pricing risk and employers screening applicants, each acting rationally on available information.
⚠ Which means Chris’s favourite illustration undercuts his own dismissal of “who controls.” Nobody controls it. That is the problem. “Legalise it and keep the state out” is a solution to a form of coercion the film does not contain — and a libertarian framework has no tool against emergent private discrimination on true information. You cannot shrink your way out of an actuary.
So the Neg’s best line is not icky. It is: the thing you are afraid of losing is the thing legalisation takes away. Positional compulsion converts permitted into required with no statute, and the enforcement is done by insurers and employers rather than police — mode change without text change, in the purest specimen the vault has yet found.
Chris: “…it turns over to market forces, which has the ‘citadel’ effect (Tacit Collusion) you describe.”
Searched: there is no page, note or memory containing citadel or tacit collusion anywhere in the vault. I am not going to reconstruct what I supposedly said — that is the grounding rule’s first mask. ⚠ This is the vault’s documented rot pattern happening in real time: a concept discussed, referred back to with confidence, and never persisted — the same failure the OS manual records for positional goods, “a topic discussed twice and never persisted.” It got caught this time only because someone cited it and the citation failed.
Tacit collusion is real and standard — firms in a concentrated market converge on supra-competitive behaviour with no agreement, purely by observing each other’s incentives. Applied here it is strong: no clinic, insurer or employer needs to coordinate for a genetic tier to form; each acts rationally on available information and the citadel assembles itself. ⭐ That is the GATTACA mechanism stated in economic rather than cinematic terms, and it is the same coordination-without-a-coordinator shape as correlated parental preference. 📌 Owed a page.
“Let’s assume the only control of trait selection is what the genetic engineers can accomplish — this removes the ‘who decides’ argument.”
Largely accepted. If the menu is bounded by the capability frontier rather than by a committee’s taste, then the option set is not curated in the objectionable sense, and “an election with a candidate list” overstates it. ⚠ The residue is only that the frontier is itself endogenous — research funds follow demand and reimbursement — so capability is shaped by preference on a lag. But that is a much weaker objection than the one I made, and the correction is his.
“This feels like evolution++ decisions in low generations instead of millions of years… maybe humans can’t handle that, maybe they can.”
This is the crux, and it is a control problem rather than a moral one.
| Generates variation | Evaluates it | |
|---|---|---|
| Natural | one generation | one generation — same clock |
| Engineered | ⭐ instant, deliberate | still one generation |
⭐⭐ Editing speeds up the actuator without speeding up the sensor. Variation becomes fast and directed; the feedback that tells you a choice was wrong still arrives at the reproductive rate of human beings. A control loop with a fast actuator and a slow sensor overshoots and oscillates — that is the precise, non-mystical content of “maybe humans can’t handle that.” Not we will break evolution; we will raise the gain without raising the sampling rate.
⚠ And it answers the other half — “the market should find answers to both.” ⭐ It can’t be the independent corrective, because it shares the lag. A market prices what it can observe, and a trait whose fitness consequences appear in forty years cannot be priced today. Markets and evolution are not two insurance policies here; they are one feedback loop counted twice, and the binding constraint on both is the same interval.
The argument above is trait-dependent, and that is a feature. Monogenic disease correction has fast, legible feedback — it works or it doesn’t, within a life. Polygenic edits for cognition, temperament or longevity have feedback measured in generations, and that is exactly where the control problem bites.
⭐ So the non-arbitrary line is not “genetic engineering,” which is a blob, but how long it takes to find out you were wrong. Regulate by feedback latency.
And note the shape — it is the vector schedule, arriving unprompted in a third domain. Not more rungs on a “how genetic is it” ladder, but a different ladder: latency, reversibility, and whether the change is somatic or germline. ⭐ This is also the first time in this session that a policy dimension has fallen out of an argument rather than being proposed and then defended — which is usually the sign it is the right one.
Chris: “Humans interfere with ‘naturalism’ all the time. How is medicine not genetic engineering? Technology has been modifying our evolutionary fitness for millennia and this is just a more direct method… we need to be cautious and understanding, but like any other tech, you can’t put it back in the bottle.”
The premise is correct and stronger than it usually gets credit for. Medicine does not merely coexist with the gene pool, it changes it — by relaxing selection. Eyeglasses, insulin and Caesarean section all keep alleles in circulation that would previously have been removed. We have been altering human evolutionary trajectory for as long as we have had medicine, so “don’t interfere with nature” is not a position anyone actually holds.
⚠ But it does not transfer the safety record, and the reason is a sign change:
| Effect on population variance | Propagates? | |
|---|---|---|
| Medicine | ⭐ increases it — relaxes selection, retains alleles | no — the treatment must be re-administered every generation |
| Directed germline editing | ⭐ decreases it — correlated preference removes variance | yes — the intervention itself is inherited |
⭐⭐⭐ Medicine is a rented phenotype; a germline edit is an owned one. Insulin’s effect must be re-purchased daily and stops the moment you stop paying — which makes it continuously revocable and error-correctable. That is the entire source of medicine’s good track record, and it is the property germline editing does not have.
So the non-uniqueness argument establishes exactly what Chris claims for it — that interference is not novel and “unnatural” is not an objection — and no more. ⚠ It does not establish that the new interference inherits the old one’s safety, because the old one was safe in the specific way of being undoable. Which is the off-switch again, for the fourth time in three days.
⭐ And his own hedge is the honest middle: “I’m not sure there is a gene for aging, but we can reduce risk to allow more people to reach that age.” Exactly — the intervention is probabilistic and indirect, which cuts both ways at once: it is far less powerful than the designer-baby framing assumes, and it is correspondingly harder to tell whether it worked. Weak effects with generational feedback is the worst combination for the control problem above, not the best.
📌 “You can’t put it back in the bottle” is Thunder’s best available argument, and I predict he does not make it. If the capability exists, a ban relocates rather than prevents — clinics move offshore, and the prohibition becomes a rule that binds only the people who cannot afford a plane ticket. ⭐ That is the drug-legalisation playbook applied to this topic, by the man who just won the drug round, and it has a further edge he is unlikely to see: an unenforceable ban is a liability default wearing a conduct command’s clothes — which would make it the first argument in the tournament to connect a resolution to the enforcement question underneath it.
⭐ The citadel concept is now written up — Coordination Without a Coordinator, created 2026-09-11 in response to the failed citation above. It carries the GATTACA reading, the correlated-preference mechanism, the state/non-state blind spot in both directions, and a discriminating test (would a defector profit, and does anyone defect?) so the idea cannot be used as a universal explanation for durable inequality.
Chris: “This stuff is interesting, but the debate will probably only scratch the surface.”
Filed as a prediction, and it is checkable. ⏳ Scored against the tape on four items — does either contender reach (a) consent of the edited party, (b) irreversibility / germline permanence, (c) positional compulsion, (d) the private, non-state character of the coercion? ⚠ Prior expectation: (a) yes, (b) probably, (c) no, (d) no — and if (d) appears from either man it is the single most impressive moment in the tournament, because it is the argument that beats a libertarian affirmative on its own terms.
“I also agree these are advanced concepts, and ‘icky’ and fear might win the day.”
Filed as his call on the round, and it agrees with the model’s post-topic call by an independent route: Chris expects the disgust default to hold; the model expects the framing burden to sink Thunder. ⭐ Same prediction, two mechanisms — which is the configuration that has produced this page’s best calls and also its most confident misses. ⚠ Note the shared blind spot: both routes assume the room does not convert to bodily autonomy. If Thunder frames it early and holds the frame — the one thing he has not done in three rounds — both of us are wrong together.
Chris: “it will be interesting to see Rex’s arguments beyond ‘it feels icky/unnatural.’” There are at least four that are not disgust arguments, and three of them turn Thunder’s own axis:
⚠ The eugenics card cuts toward the Neg in this room specifically, and not by squeamishness: legalise it and the state, insurers and employers will apply pressure. That is an anti-state argument for prohibition, which is the rarest and most effective shape available against a libertarian affirmative — and the same structure as Medeiros’s socialised-cost turn, which is the move that went closest to beating Thunder.
Chris went and looked Thunder up rather than trusting the view counts:
“I was doing research on Thunder’s reach. Only 4.45k on YouTube with 2.2k videos (mostly shorts), but Instagram has him at 37.1k, and maybe he has a large TikTok following — I don’t know his main platform.”
⚠ This page has used YouTube view counts as the reach proxy throughout, and that is a category error. Thunder’s YouTube presence is negligible — 4,450 subscribers across 2,200 uploads is a shorts channel, not an audience — while his Instagram following is roughly eight times larger. A contender’s reach lives wherever their audience lives, and this bracket only ever measured one platform.
⭐ What survives, and what does not:
Finding 11 was rescued by correlating margin against pool size (r = −0.74, n=7), with the abortion round’s 3,495 views as its most influential point. But pool size is not exogenous. It is produced by two things that independently predict margin:
⭐⭐ So the correlation may be spurious, with topic salience as the common cause of both the big pool and the narrow margin. That is the same variable the axes-conflict hypothesis already names — and it now has a mechanism for why it also inflates the audience. Finding 11 stays unresolved, and the honest verdict is stronger than “n is small”: the independent variable is endogenous.
“TBH, I don’t think [SF1] was as close as the results showed it to be. We basically had concession by the Aff that they don’t want opioids legalized. OTOH, the Neg kinda conceded it was OK for non-opioids to be free, though he was less set against it, citing even marijuana has its troubles.”
Both readings check out against the tape, and together they are worse for the Aff than the ballot shows:
| Conceded | Where | |
|---|---|---|
| Aff (Thunder) | the opioid ground — “worst drug, and I’m trying to talk about the psychedelics” | 46:42, and again when he answers the pill-mill case with “well, they have the right to do that” |
| Neg (Medeiros) | non-opioids, with a hedge — “I’d 100% be, yeah, legalize the psychedelics”, but marijuana daily users “1 million to 18 million” | 49:00 / 19:12 |
⭐⭐ The two concessions collapse the round onto opioids — and that is the ground the Aff had already given up. On a resolution reading “all drugs should be legal,” the Affirmative conceded the only live question and won the ballot anyway. So the merit/ballot gap is not the 5.4 points the card shows; the card understates it, and this is a sharper version of finding 5 than the raw margin conveys.
“I think he rates high in likeability even if his arguments are not so sound :)”
This is a real rival to the axis model and it has not been separated from it once. Thunder’s two wins — abortion and drugs — are both on axis-favourable ground, so every observation to date is consistent with either the axis carries him or the room likes him. ⚠ Two instruments, zero discriminating evidence.
⭐ The final is the discriminator, and only if the topic cooperates. The test, filed now so it cannot be fitted afterwards:
| If the published topic is… | And Thunder wins | And Thunder loses |
|---|---|---|
| axis-favourable to Thunder (liberty/anti-state) | ⏳ tells us nothing — both instruments predicted it | axis model damaged |
| ⭐ axis-neutral, or axis-favourable to Jones | ✅ likeability confirmed as a distinct force | ✅ axis model clean, likeability unnecessary |
📌 Registered consequence: the filed call of Jones, narrow should be read as conditional on the topic being axis-neutral. If the resolution lands on liberty-versus-state ground, the call is wrong on its own logic and should be treated as already refuted — the axis model just beat the execution instrument once, and it would point at Thunder again.
He expects the ballot to survive the venue change: “I think they still use a public voting system in the live event… while I am sure supporters from each side will be there, it should also attract a non-partisan audience, so reach will matter less.” If that holds, it cuts in a useful direction — a less partisan room is a room where execution should matter more, which is the instrument that just lost SF1. ⭐ So the final is a second, independent test of execution-versus-axis, on a pool selected to favour execution.
⚠ And the variable with no precedent in fourteen rounds: “they also have commentators that might sway opinions.” No online round had live commentary between the argument and the ballot. A commentator supplies a frame to a room that has not yet voted — which is the moderator-interference question (finding 21) with the volume turned up and the ballot still open. Nothing on this page is calibrated for it, and it should be watched as a format change rather than a detail.
Checked at Chris’s request. The ballot is closed and the numbers are not posted, by design.
| What was checked | State on 2026-09-09 |
|---|---|
wordwardebate.com/voting-results |
404. The URL this hub had recorded is dead; the page moved to /view-results |
/view-results |
A single Final Four bracket image and nothing else. The four semifinalists are shown; neither semifinal winner is advanced, and there is not one percentage card anywhere on the page |
/vote-now-1 |
“WHO WILL BE IN NASHVILLE??? CHECK BACK SOON TO SEE THE RESULTS!” — the six-hour window is long shut and the outcome is deliberately held |
| Championship | Word War Debate II — LIVE, Friday 12 September, Woolworth Theatre, Nashville. The final is Rex Jones vs. Octavius Thunder |
⭐⭐ This reinterprets defect 5, and not in the organisers’ favour. The 2026-09-08 wipe was recorded here as a current-state display overwriting its own history. It now reads as deliberate: the per-round cards were cleared to make room for a reveal-driven bracket, and the finalists are being kept secret to sell a live show. That is a defensible entertainment decision and a total loss for the record — the difference between carelessness and policy is that policy will do it again.
⚠ And the prior season shows what “archived” means here. /word-war-i-results proves they do keep past seasons — and the format is winner’s name only: a matchup graphic, the word WINNER, a name. No percentages, no ballot counts, and several of the images are already broken placeholders.
🚫 So the working assumption must now be that the semifinal and final margins will never be published at all — not wiped, never issued. If the Contender Series is archived the way WWI was, we get four names and no numbers.
Two consequences for this hub:
📌 Filed as defect 17. The reveal is 12 September; the finalists’ identity will be public then whether or not any number is.
Chris: “Also they have to fly the winners out to Nashville… but let’s hope they can get the winners out on 3 days’ notice :)”
This is the observation that dissolves the withheld-results puzzle. The semifinal ballot closed on 8 September; the live show is 12 September at the Woolworth Theatre. Nobody gets a contender from wherever they live to Nashville without telling them, and nobody books that flight the morning of. So:
🔓 The embargo is public-facing only. The result exists, the organisers hold it, and the finalists were told within hours of the ballot closing. What is being withheld is not the outcome but the announcement — which is a marketing decision, not an unfinished count.
⚠ And it is the calendar defect cashing out in cash — the idle fortnight was the missing airfare discount. Retracted the same day, on Chris’s correction: “The extra paid for the plane tickets is probably justified. It is the cram session due to that big gap that isn’t.”
He is right, and the retracted version confused two different things. A late semifinal is required by the product: the reveal has to be close to the show, and the alternative — running the semifinals live in Nashville — fails on its own terms. Chris: “It might make sense to have the SF and finals on the same day, but they are already planning a 6-hour event… this would make it 8+ with no prep time for the finals.” So the four-day booking window is a deliberate purchase of marketing tension, not an accident, and the premium is the price of the format working. It is not the defect.
Two grounded facts point the same way. First, /word-war-i-results shows the previous live event ran a three-match card: WWI UNDERCARD → WWI PRIME SHOWCASE → WWI MAIN EVENT. Second, flying all four semifinalists is the cheap hedge — it fills an undercard, it keeps the reveal intact until showtime, and it removes the need to know the winner before booking.
So the prediction:
| # | Call | Resolves |
|---|---|---|
| N1 | The Nashville card is announced before 12 September, and announcing it reveals both semifinal winners — the moment a pairing is published, the bracket is solved | on announcement |
| N2 | All four Final Four contenders travel, with the two semifinal losers used on the undercard rather than sent home | 12 Sep |
| N3 | ⚠ The Contender Series final is not the main event. WWI’s main event was the marquee outside name (Kestenbaum) while the in-house pairing (Mullally/Pisco) opened. If that pattern holds, the championship this hub has tracked for six weeks is the middle slot on someone else’s card | 12 Sep |
“Ohh, this contender series is actually the undercard :) They show this on the WW2 card.”
✅ N3 hits. The championship this hub has tracked for six weeks is the opening slot on someone else’s bill. “Road to WWD II” is literal: the Contender Series is a feeder, not a season — its job is to deliver a name by a date.
⭐ Which explains the results handling better than carelessness ever did. A feeder does not need margins, a permanent archive, or a long ballot window; it needs a body in a chair on the 12th. Every item in the record-keeping cluster — the wiped cards, the withheld semifinals, the winner-name-only WWI archive — is exactly what you would build if the bracket’s output were a person rather than a record. That is not a defence of it, but it is a reason, and reasons predict better than sloppiness does.
⚠ N2 is now doubtful and should be downgraded, not quietly dropped. It predicted the two semifinal losers would fill the undercard — but the undercard is the Contender final itself, so that slot is taken and the losers have no obvious place on a three-match card. Probable miss. N1 stays open: the card apparently shows the slot rather than the names, or Chris would have had the result in front of him.
Take Chris’s correction seriously and the diagnosis gets sharper rather than softer. Three dates were genuinely constrained — the live show (12 Sep, venue booked), the semifinal (late, for the reveal), and the gap between them (short, for travel and tension). None of those is the problem.
The problem is the eleven debates upstream of the semifinal. Round 2 (7) and the quarterfinals (4) had the whole of 12 August – 5 September available — 24 days, one debate every 2.2 days, with 48-hour ballots throughout. Instead they ran in nine days with the ballot cut to 24 hours and then to six.
⭐ So the corrected fix moves no fixed point at all. Same live date, same semifinal date, same four-day reveal, same airfare. It only spreads the middle across days that were already empty — which means the schedule that repairs six of the seventeen defects costs nothing, not even the marketing tension the late reveal was buying.
That is the memo. Not “stop compressing” in the abstract, and not “book flights earlier” — “the fortnight you left empty in August is the fortnight rounds 2 and 3 needed, and using it would have cost you nothing you cared about.”
Both semifinal intros were checked for the two missing margins. Neither gives a number. What they give instead is a characterisation, and the characterisations do not survive contact with the cards we already hold:
| Moderator | Says about | Characterisation | Known truth |
|---|---|---|---|
| Binkley (SF1) | Thunder’s QF B | “he won a vast amount of the votes” | unknown |
| Perez (SF2) | Jones’ QF C | “he won the last debate, but it was pretty close” | unknown |
| Perez (SF2) | Summerhays’ QF D | “he crushed in his last debate” | +9.0 — the narrowest margin in the tournament, and she moderated it herself |
That last row is decisive. A +9.0 described as “crushed” is not a rounding error or a misremembering — it is promotional copy. Which means the two unknown characterisations carry no information: “a vast amount” and “pretty close” cannot be converted into margins, and it would be reasoning backwards to treat “pretty close” as support for the dilution reading of QF C simply because that is the answer this page would prefer.
So the route splits, and the split is the finding:
Consequence for the bracket record: the QF B and QF C margins are almost certainly lost for good. The cards are deleted, the semifinal intros gave no figures, and only the championship broadcast remains as a possible source. Standing action unchanged and narrowed: transcribe numbers, discard adjectives.
Moderators: SF1 is Brad Binkley (Propaganda Report), his second round after Biden/Trump; SF2 is Monica Perez, her seventh — she has now moderated more of this tournament than anyone.
1. View counts are revised downward. QF C read 2.6K at ~23h on 2026-09-07 and reads 2.1K at ~2 days today — a ~19% reduction, which is YouTube stripping inflated or bot traffic after the fact. Any view snapshot is provisional, and the tape that looked like the bracket’s runaway hit lost 500 views. This bites the turnout-band work directly: denominators measured early are too big.
2. Views-per-hour decays with age, so rates measured at different ages are not comparable — and this page has now made the opposite error twice on the same question.
| Read | Method | Verdict on QF B |
|---|---|---|
| First pass (2026-09-07) | raw counts at different ages | “badly underperforming” — wrong method |
| Correction (2026-09-07) | rates at different ages — 20/hr at 9h vs 15/hr at 36h | “running faster than QF A” — also wrong method |
| Now, matched maturity | QF A 550 @ ~36h = 15.3/hr · QF B 310 @ ~28h = ~11/hr · QF C 2,100 @ ~48h = 43.8/hr | QF B is the slowest of the four |
So the retraction was itself wrong, and the original direction was right for the wrong reason. The rule that prevents a third error: compare cumulative views at a matched age, never rates measured at different ages. The bye-cost hypothesis is back in play — weakly, since a topic effect explains it just as well.
With the card archive deleted, Chris’s observation is the whole plan: “maybe we will glean the margins from the intros when we discuss the final 2 debates.” Correct, and it is no longer a supplementary channel — it is the last one. The QF B and QF C margins exist nowhere else, and the semifinal intros are the only place they can still surface.
Standing action, upgraded: transcribe every number a moderator speaks aloud in the Final Four and the final. Current reliability of that route: seven accurate within ±0.6, one wrong by 6.1.
Same reasoning as the quarterfinal projection, which is now 4-for-4. Pairing adjacent quarterfinals gives:
| SF | Feeds from | Known half |
|---|---|---|
| 1 | QF A vs. QF B | Medeiros vs. (Tareyak / Thunder) |
| 2 | QF C vs. QF D | (Tejeda / Jones) vs. Summerhays |
The results page reads “FINAL FOUR BEGINS SOON!”, and the championship is live in Nashville on 2026-09-12 — so the semifinals and final must all air inside five days. If this projection holds, the bracket has been fully solved from the round-of-16 list alone, and the only thing the organizers have published that the hub could not derive is the resolutions.
Compact view of the ledger; the ledger rows carry the reasoning, this carries the record.
| Round | Chris’s call | Result | Verdict | Note |
|---|---|---|---|---|
| Masculinity | Owlish, on delivery not substance | Owlish +48.4 | ✅ | Called the mechanism, not just the winner |
| Masculinity | Garcia keeps losing on legibility | 25.8% | ✅ | Open for future appearances |
| AI | Charsky | Charsky +58.1 | ✅ | The load-bearing hit: argument quality beat legibility and audience prior at once |
| Too sensitive | Tejada | Tejada +89.1 | ✅ | Called before the tape’s second half was credited |
| Rich without luck | Calvin (Aff) | Aftermath +43.8 | ❌ | Prior was general-population, pool is agency-affirming → finding 10 |
| Algorithm | Hamm — “might be close” | Cruz +12.4 | ❌ | Wrong winner, exactly right on shape — the only competitive round in the bracket |
| Therapy for men | declined to call | Summerhays +42.8 | ✅ | Abstention vindicated: substance read and ballot diverged 43 points |
| Fragile culture | Turner (Aff) | Bourdeau +47.4 | ❌ | Pre-registered meaning holds: the concessions were invisible |
| Military service | Lawrence on rhetoric; argument to Brunet | Brunet +57.2 | ❌ ballot / ✅ argument | The designed test of rhetoric-beats-argument. It failed by 57 |
| Meta | Bracket selects legibility over correctness | — | ❌ refuted | The bracket’s most valuable single result |
| Hate speech | McPherson (Neg) | Thunder +25.0 | ❌ | Called against Chris’s own position; the better debater lost to a closing about “satanic pedophilic overlords” |
| Party loyalty | Kung Fu Joe (Aff) | Kewl Vic +62.0 | ❌ | Called against a personal acquaintance — and the acquaintance won by 62 |
| Therapy culture | Smith, on reach alone | Smith advances — margin never published | ✅ | The ledger’s one pure affiliation call, made against the merits read — and it paid |
| Feminism rerun | Tareyak (Neg) | Tareyak advances — margin never published | ✅ | The burden-play Aff loses; the “conceded the room” predictor gets its first observation |
| R2 — data centers | Medeiros, on polish and reach | Medeiros +76.2 | ✅ (unscored) | Round pre-excluded, so it earns the model nothing — but the mechanism call was right, and +76.2 between two unknowns is finding 2’s cleanest specimen. ⚠ Filed against a personal connection |
| R2 — women’s suffrage | Aftermath (prior-only, pre-tape) | Tejeda +63.2 | ❌ | ⭐ The decisive test, resolved. The card was published, later than this page checked. Prior-only dies by its own registered condition |
| R2 — women’s suffrage | Tejeda (two-factor, post-opening) | Tejeda +63.2 | ✅ | Filed and dated separately so the discrimination is clean. The second pre-registered discrimination the ledger has run, and the second one two-factor won |
| R2 — mass deportation | Ruelas (prior-only, pre-tape) | Charsky +58.6 | ❌ | The prior-only model’s registered loss |
| R2 — mass deportation | Charsky (two-factor override, pre-tape) | Charsky +58.6 | ✅ | Pre-named as the override candidate, with the scoring rule set in advance. The ledger’s first clean pre-registered discrimination |
| R2 — mass deportation | Chris’s lean: Charsky on presentation and reach | Charsky +58.6 | ✅ (lean) | Filed as a lean, not a call — and against his own read that Charsky’s logic and sources both fail |
| R2 — Iran war | Model: Cruz, narrow (pre-tape) | Summerhays advances | ❌ | The anti-interventionist axis coded and beaten. Margin never published, so the narrow band is untestable |
| R2 — Iran war | Chris: Summerhays (post-tape) | Summerhays advances | ✅ | ⚠ The two calls disagreed and Chris won. His third call against his own position — “I don’t agree with the aff position, but I think he argued better” — now 2-for-3 on that pattern |
| R2 — amnesty | Model: Rex Jones, wide (pre-tape) | Jones +65.2 | ✅ | Pure prior play, filed sight unseen — and the margin band held |
| R2 — amnesty | Chris: Rex Jones (post-tape) | Jones +65.2 | ✅ | ⭐ The ledger’s strongest joint prediction, and it paid. Two independent routes to the same call: the model used the pool’s prior and never saw the tape; Chris used execution (“Aff was too scattered”) and never used the prior |
| R2 — Biden/Trump | Chris + model: Josh Smith (post-opening) | Bourdeau advances | ❌ | A straight prior miss on the model’s strongest coding, with no margin published to diagnose it. Prior and reach agreed here, so the ballot discriminates nothing — it just says both were wrong |
| R2 — abortion | no call — result published first | Thunder +16.8 | — | ⚠ Unscored, and the page says so. But this page’s own audience coding points at Vic, so it is a prior-only failure the model was never made to register. Occasioned the axis correction |
| R3 — DSA | no call — result published first | Medeiros +74.8 | — | Unscoreable. Medeiros is 2-0 by 75+ and has never been modelled |
| R3 — nationalism/socialism | no call — result published first | Summerhays +9.0 | — | Unscoreable, but consistent with clause (c): hostile prior picks the winner, execution gap eats the margin to the bracket’s narrowest |
| R3 — DSA | Chris: Bourdeau, on round-2 form — held, reported after the card was known | Medeiros +74.8 | ❌ (unscored) | Reported against himself. Not filed, so it scores nothing — but it is the form model losing to a hostile prior by 75 |
| R3 — nationalism/socialism | Chris: Charsky, “he is the strongest” — held, reported after the card was known | Summerhays +9.0 | ❌ (unscored) | Same shape, same result, 65.8 points of margin apart — the pair that quantifies clause (c) |
| R3 — Iran (Tejeda/Jones) | Model: Rex Jones, wide (pre-tape, pre-result) | Jones advances | ✅ winner / ⚠ band unresolvable | Prior + affiliation aligned. Half of the Iran double-header |
| R3 — Iran (Tejeda/Jones) | Chris: Tejeda (pre-result) | Jones advances | ❌ | ⚠ The ledger’s first live disagreement on an open round. He reads the contenders (largest demonstrated bloc, +89.1 then +63.2); the model reads the resolution (anti-war axis + the Jones network). He has the better precedent — Tejeda already beat a near-unanimous prior by 63 in the suffrage round |
| R3 — Iran (Tareyak/Thunder) | Model: Octavius Thunder, narrow (pre-tape, pre-result) | Thunder advances | ✅ winner / ⚠ band unresolvable | Same axis, weaker on every dimension; also the bye-cost test |
| R3 — Iran (Tareyak/Thunder) | Chris: Thunder (pre-result) | Thunder advances | ✅ | ✅ Agreement by independent routes again — the model on the anti-war prior, Chris on form (“Tareyak really hasn’t been tested, and he still beat Vic despite a poor performance”). Neither used the other’s input |
| Structural | Chris: Thunder is the weakest of the final four | ⏳ — he is in it, drawn against Medeiros | — | Still testable if Thunder advances — the projection puts him against Medeiros, who is 2-0 by 75+ |
| ⚠ Filing failure | The model’s two calls on QF C contradict each other — Rex Jones, wide (≥40) (finding 17, filed 2026-09-07) against margin under +40 (finding 11, filed 2026-09-08) | ⏳ | — | ⭐ Filed a day apart off two different findings and the conflict went unnoticed until the tape was opened. Sloppy filing that accidentally produces the best test on the board: the round now discriminates between two of this page’s own findings. Handling set before the card was read — Jones by 40+ kills dilution, Jones under 40 kills the wide band, Tejeda kills both |
| R3 — Iran (Tejeda/Jones) | Model: the narrowest margin of round 3 (filed blind 2026-09-08; result deliberately not looked up, but the 24h ballot has probably closed — so blind, not pre-result) | ⏳ | — | ⭐ Independent of who wins, which is what makes it worth filing. Finding 11’s dilution result says margin falls with pool size; QF C is running at 113 views/hr against QF A’s 15 and QF B’s 20, so it should have the largest pool in the round by a wide margin. Predicted: narrower than QF A’s +74.8 and QF D’s +9.0 is the hard case — the falsifiable form is QF C comes in under +40. A blowout kills the dilution reading outright |
| SF2 — flock cameras | Model: Rex Jones, narrow (pre-tape, pre-result) | ⏳ | — | “Beneficial” is the worst side drawn since socialism — Summerhays defends mass surveillance to an Infowars-adjacent room. Prior + execution + form aligned, as in QF C |
| SF1 — topic unpublished | Model: Medeiros, wide (pre-tape; ⚠ no topic, so execution-and-reach only) | Thunder +5.4 | ❌❌ | Best prosecutor in the bracket against the narrowest-winning contender; both low-reach, so a small pool and a wide margin |
| The final | Chris + model: Rex Jones vs. Medeiros (both filed before either semifinal resolves; SF1 has not even aired) | Jones vs. Thunder | ⚠ half | |
| The championship | Model: Rex Jones, narrow (post-topic, on axis-conflict + framing burden) | ✅ Jones champion — margin never published | ✅ | ⭐ Agreement by two routes again. Both semifinal calls held jointly, so the final is a joint prediction with a single failure point — if either leg misses, both are wrong at once. The strongest form: it forecasts the championship pairing two rounds and one unaired tape ahead |
| Structural | ✅ leg 1 hit — Chris: the QF B winner loses the semifinal on audience size — “if Thunder wins this, this is a popularity contest, and Thunder goes down easily when faced against a similar-reach opponent” (pre-result) | ⏳ | — | ⭐ A two-stage test that survives either quarterfinal outcome. Both legs landing turns “this is a popularity contest” from a complaint into the model; the QF B winner beating Medeiros falsifies it. Filed 2026-09-07 |
| Structural | The quarterfinal bracket, registered 2026-09-01 before announcement | 4 of 4 pairings correct | ✅ | The page’s first purely structural prediction, and it solved the tree |
Record: 9 of 16 ballot forecasts correct (56%), one correct abstention, one meta-claim refuted, and one structural prediction that hit 4-for-4. Updated 2026-09-07 — round 2 is complete, and wave 2 resolves as Jones ✅, Summerhays ✅, Smith ❌. The model ledger is separate and now stands at two-factor 3–2, prior-only 0–2 on pre-registered round-2 calls, with the caveats on the denominator stated where they belong. Two round-3 calls are open. Updated 2026-09-01 — the last two were resolved by the round-of-16 bracket, not by result cards. Both late hits were non-argument calls that paid: Smith on reach alone, Tareyak on confidence and straightforwardness. That does not revive the dead finding 15 — the misses still run in both directions — but it does move the bar the round-2 model has to clear, from 33% to 45%.
Open bookkeeping: an Aff/Neg tally would be worth having — does either side enjoy presumption with this audience? — but three of the ten side assignments need verifying against their tapes before the count means anything, so it is deliberately not asserted here.
⚠ Retrieval note (2026-08-17): results are published as images, so they cannot be fetched programmatically. A pass over /voting-results returns navigation, a copyright line, and lazy-loaded image placeholders; the only machine-readable result on the whole page is one image’s alt text (“Rex Jones wins Round 1 of Word War Debate Contender Series”). No percentages appear in the markup at all. So every tally in the table below was — and must be — read off a rendered page by a human or a browser session, and the finding 13 reconstruction depends on those two-decimal figures being transcribed accurately. Attempting the fetch again without a rendering browser is wasted effort.
Status check the same day: /vote-now-1 shows only Kung Fu Joe vs. Kewl Vic open among contender rounds (plus a non-bracket “Candace vs. Andrew” event), so the military-service, feminism-rerun and hate-speech ballots have closed and their results exist but are not yet transcribed here. It also independently corroborates the production note above: no ballot has opened for either unlisted matchup, so neither has aired.
Source: wordwardebate.com/voting-results. Six of the seven covered rounds have posted; Cruz vs. Hamm and Summerhays vs. Gilkison are still inside their 48-hour windows, so the blind protocol continues for those two.
| Round | Result | Margin |
|---|---|---|
| Masculinity — Garcia vs. Owlish | Owlish 74.2% – 25.8% | +48.4 |
| Feminism — Ruelas vs. Anton | Ruelas 60.5% – 39.5% | +21.0 |
| Boomers — Rex Jones vs. Meyers | Rex 61.7% – 38.3% | +23.4 |
| AI — Charsky vs. Bunn | Charsky 79.05% – 20.95% | +58.1 |
| Rich-without-luck — Constable vs. The Aftermath | Aftermath 71.88% – 28.12% | +43.8 |
| Too sensitive — Guptill vs. Tejada | Tejada 94.55% – 5.45% | +89.1 |
| Algorithm — Cruz vs. Hamm | Cruz 56.2% – 43.8% | +12.4 — the closest in the bracket |
| Therapy for men — Summerhays vs. Gilkison | Summerhays 71.4% – 28.6% | +42.8 |
| Fragile culture — Turner vs. Bourdeau | Bourdeau 73.7% – 26.3% | +47.4 |
| Military service — Lawrence vs. Brunet | Brunet 78.6% – 21.4% | +57.2 |
Retrieved 2026-08-17. Ten result cards are archived at raw/debates/wordwar-results-2026-08-17/ — the site publishes results only as images, so those files are the primary source. Never posted: therapy culture and the feminism rerun — hate speech and party loyalty published on 2026-08-18, but those two cards never appeared, and their winners were recovered from the round-of-16 bracket instead.
⚠ Note the precision drop. The two newest results are reported to one decimal, not two. That materially weakens the finding 13 reconstruction, because a single decimal admits far more denominators: 78.6/21.4 fits any multiple of 14 (smallest 11–3), and 73.7/26.3 any multiple of 19 (smallest 14–5). Against the military-service round’s 224 views, the 15–25% turnout band observed elsewhere would favour n≈42 (18.8%) over n=14 (6.3%) — but that is an inference from the band, not from the figures, and is flagged as such.
(Turner vs. Bourdeau still inside its window as of 2026-08-12.)
| Prediction | Made | Status |
|---|---|---|
| Medeiros (Aff) wins, R2 data centers | R2 M1 discussion, 2026-09-01 | open — unscored by design, since this is the match pre-registered as excluded (both contenders advanced on byes, and the pool’s prior splits on the resolution). Chris: “I am going with Alex here mostly due to polish and reach, though Lee can muster up quite a zealous crowd too.” ⚠ Disclosure, and it runs against the call: Lee Horseradish owns the show Chris debates on, and Chris is calling for his opponent while also disagreeing with Lee’s side — the second ledger entry where the affiliation channel points opposite to the prediction (the first was Kewl Vic, where the acquaintance won by 62). The round’s value is diagnostic rather than predictive: it is the first resolution in sixteen that asks “should” about a decision private parties make, which is the finding it produced |
| Owlish advances on delivery, not substance | R1 §7 | ✅ CORRECT — 74.2%, the second-largest margin on the board |
| Garcia keeps losing on legibility wherever he appears | R1 OQ4 | ✅ so far — 25.8%. Stays open for future appearances |
| Bracket selects for legibility over correctness | R1 OQ4 | ❌ REFUTED — and this is the bracket’s most valuable single result. The AI round is exactly the case finding 2 said was needed: legibility (finding 6) and audience priors both pointed at Bunn, and only argument quality pointed at Charsky. Charsky took 79.05%. Argument quality beat legibility and sympathy simultaneously and decisively |
| Round 2’s winner “should go far” | R2 §6 | open — Ruelas took R2 at 60.5%; resolves in round 2 |
| Charsky wins round 4 | R4 §12, §16 | ✅ CORRECT — 79.05%. The test resolved the informative way. Three predictors, two directions: argument quality → Charsky; delivery/legibility (finding 6) → Bunn; audience priors (AI skepticism is the popular position) → Bunn. Charsky won by 58 points, so argument quality beat both legibility and sympathy at once. The single most load-bearing result in the bracket |
| Kung Fu Joe (Aff) wins, Joe/Vic | R14 discussion | ❌ WRONG by 62 points — Kewl Vic took it 81–19, the second-largest margin in the bracket, against a call giving the Aff both better arguments and better presentation. Its value: the round holds argument quality nearly constant (the two agreed), leaving posture as the visible variable — Joe’s closing is about his opponent, Vic’s about the question. Best specimen for finding 16, with the confound that Vic is both the larger channel and the acquaintance. Original registration: registered blind. Chris: “Aff has the better arguments, and he presented well, I think he wins.” Convergent (better debater and predicted winner), and made with a personal connection to the opponent — Chris knows Vic — which is the first call in the ledger where the affiliation channel runs against the prediction rather than with it |
| McPherson (Neg) wins, McPherson/Thunder | R13 discussion | ❌ WRONG — Thunder took it 62.5–37.5. The better debater lost to an opponent closing on “satanic pedophilic overlords” and “global enslavement” — and lost while arguing the side Chris disagrees with, so Chris and the room reached the same winner-side by opposite routes. McPherson’s own post-mortem became finding 16. Original registration: registered blind, and a call against Chris’s own position on the resolution, which is what makes it worth logging. He sides with the Aff on the merits and still calls the round for the Neg: “the neg performed better here as he had a plan and he has experience dealing with these types of debates… Despite all of the ‘bad’ arguments, I think this goes to the Neg.” So this is the second clean instance of the Garcia pattern — the side the vault reads as correct losing on execution — and it refines the demoted finding 5 in a useful direction: not the format punishes correctness, but a correct position with no plan loses to an incorrect position with one |
| Tareyak (Neg) wins, David/Tareyak | R12 discussion | ✅ CORRECT — Tareyak advances. Recovered from the round-of-16 bracket on 2026-09-01; the card was never published, so the winner is known and the margin never will be. Registered blind, and notable for being convergent where the previous call was split. Chris: “the Neg did well to steer the debate away from the very narrow framing… I think the neg pulls this as he was more straight-forward with his arguments and was more confident.. the Aff’s closing almost sounded like a concession speech.” The Aff himself models the pool correctly on air (“I do know sort of how the audience probably skews here”) and then trashes the debate in his exit interview, drawing the moderator’s “we’ll see if the strategy of trashing the debate quality is rewarded by the audience” — so a David win would be the bracket’s first case of a procedural/burden-play affirmative beating a substantive negative before a lay ballot, which is the direct test of finding 6’s extension. It did not happen. The burden-play affirmative lost, so that test returns no — and the moderator’s on-air “we’ll see if the strategy of trashing the debate quality is rewarded by the audience” is answered in the negative |
| Lawrence (Aff) likely wins, Lawrence/Brunet — but the argument goes to Brunet | R11 discussion | ❌ WRONG — and it is the most important result in the bracket, because it was the designed test and it failed cleanly. Brunet took it 78.6 – 21.4. The call was the ledger’s first explicit split: argument to Brunet, ballot to Lawrence on rhetoric. The side named as having the better argument won by 57 points — the second-largest margin on the board. So the pattern the demoted finding 5 was built on does not merely lack support; the one round staged to test it produced the opposite, decisively. Note what makes it strong evidence: the Aff was a USAF veteran arguing a veterans-first proposition to an audience that skews patriotic, i.e. audience prior and rhetoric both pointed at Lawrence, and he still lost by 57. Together with Charsky (+58) this is now two decisive rounds where argument quality beat rhetoric and sympathy simultaneously. Original registration below. Chris: “the argument goes to Chris [Brunet], but I don’t think he was aggressive enough… Glenn is likely to take this based on appeal and rhetoric alone. He was the better ‘debater’ even though he had the worse argument.” This is the exact pattern the demoted finding 5 was built on, now stated as a prediction rather than noticed afterwards — so it is the cleanest available retest of whether the format rewards rhetoric over argument |
| Smith (Aff) wins, Smith/Ouedrago | R10 discussion | ✅ CORRECT — Smith advances. Recovered from the round-of-16 bracket on 2026-09-01; no card was ever published, so the margin is permanently unknown. Registered blind, and explicitly a reach call rather than a merits call. Chris: “I think JS takes this… JS has a channel called ‘JS Urban Adventures’, and has a decent sized following. I think this will carry him.” This is the first prediction in the ledger made on the affiliation channel alone, and it is testable against the merits read: the Aff conceded his own causal claim and the Neg — in his first formal debate — had a defensible case. A Smith win is consistent with finding 2; an Ouedrago win would have been the bracket’s first clear instance of an unknown beating a following. Smith won, which leaves finding 2 holding its single best data point — the one round where reach was the only predictor pointing that way, and it was the one that was right |
| Turner (Aff) wins, Turner/Bourdeau | R9 discussion | ❌ WRONG — Bourdeau took it 73.7 – 26.3, and this is the informative failure the registration anticipated. The call rested on execution: the Neg conceded the data, the mechanism, and the causal principle (“they’re going to be a product of whatever that environment is”), and the Aff never cashed any of it. The page said in advance that a Neg win “would say the concessions were invisible to the audience, which is itself worth knowing.” It says exactly that, by 47 points. Concessions extracted but never converted are worth nothing on a lay ballot — an audience scores what a debater does with an admission, not the admission. Pairs with the Ouedrago round, where the same failure to collect went unpunished. Original registration below. Unlike most calls in this ledger it rests on execution rather than sympathy: the Neg conceded the data, the mechanism, and the causal principle, and the Aff still declined to cash it — so a Neg win would say the concessions were invisible to the audience, which is itself worth knowing |
| Summerhays/Gilkison — no confident call | R8 discussion | ✅ the abstention was correct, and that is the point. Chris read Neg (Gilkison) as winning on substance — “does this rise to the level of ‘doesn’t work’.. I don’t think it does” — while explicitly declining to predict the vote: “this might be a victim of small sample size :)”. The vote went the other way: Summerhays 71.4–28.6. So the substance read and the ballot diverged by 43 points, exactly the gap the abstention anticipated. Finding 13 earning its keep: a round turning on ~21–49 ballots cannot be forecast from the merits, and the correct move was to say so rather than guess |
| Hamm (Aff) advances, Cruz/Hamm | R7 discussion | ❌ WRONG — Cruz took it 56.2–43.8. But Chris called the shape exactly: “might be close.” At +12.4 it is by far the tightest margin in the bracket, on an inferred n of roughly 16–73 — i.e. a swing of a few votes. This was flagged in advance as the informative outcome, and what it informs is finding 13 rather than anything about argument quality: on the tape the Aff had the cleaner case (the severance held; the Neg conceded the inside standard in his own closing), the predictors all agreed, and it still went the other way by single digits of ballots. At this sample size, “the better case lost” and “four people voted differently” are the same sentence |
| Tejada (Aff) advances, Guptill/Tejada | R6 discussion | ✅ CORRECT — 94.55%, the largest margin in the bracket. Tejada spent his opening and a third of the crossfire arguing his opponent’s case and was corrected by that opponent rather than the moderator. Qualified by Chris after the result: the second half was genuinely good — the struggle/duty case is organised and he closes well — so the margin is partly earned, and this page’s original “whatever the ballot measures, it isn’t performance” is withdrawn as too strong. An 89-point margin still exceeds what a strong second half explains |
| Calvin (Aff) wins round 5 | R5 discussion | ❌ WRONG — and instructively so. The Aftermath took 71.88%. Chris’s two premises: (a) audience priors favour the luck thesis because “people who are not rich want to believe the topic”; (b) the Neg didn’t make his case. (b) still looks right on the tape. (a) is backwards for this audience — see finding 10 |
✅ = reviewed · ◐ = captured, not reviewed · ◻ = candidate · blank = not planned
Note: the channel title for the Boomers round bills “Miriah Headbangs”; the moderator introduces her as Mariah Meyers (master herbalist, co-host of Poppy’s Field Project). Rex Jones is introduced as a former Infowars reporter/contributor “before 2016.”
| Resolution | Contenders | Why / vault hook | |
|---|---|---|---|
| ✅ | Is Masculinity in Crisis? | Owlish vs. Gabriel Garcia | → review; produced Constructed ≠ Arbitrary |
| ✅ | Is Feminism Helping Modern Relationships? | Martae Ruelas vs. Miles Anton | → review; scope+time, survey-reason≠cause |
| ✅ | Is Feminism Helping Modern Relationships? | David S. vs. Tareyak | → review; moderator Ryan Mullally (his third). The rerun — same resolution as round 2, different pair, still the designed test of finding 2. Decided in the prompt negotiation: only feminism is defined, so the Aff runs a squirrel (one instance satisfies him; the Neg must defeat every interpretation) — an affirmative annexing the negative’s easy burden, the mirror of Guptill forfeiting it. Also asymmetric latitude, and the unbundling pincer. Result pending |
| ✅ | Should Military Service Be a Prerequisite for Voting? | Glenn Lawrence vs. Chris Brunet | → review; 1:13, moderator Ryan Mullally (his second). The bracket’s cleanest inflation specimen — the Aff widens military service to national service in his opening and is named for it in the Neg’s first thirty seconds — and the inflation refutes him, since granting that firefighting demonstrates commitment concedes commitment isn’t military-specific. Also the round where the Aff’s answer to the electorate-capture charge (“most lefties don’t serve”) is the charge. Both vault suffrage pages pre-adjudicate it and neither debater reaches the scope move. Result pending |
| ✅ | Is Political Party Loyalty Illogical? | Kung Fu Joe (Aff) vs. Kewl Vic (Neg) | → review; ~1:10, moderator Ryan Mullally (his fourth). Both debaters admit on air they didn’t research why the two-party system exists — the one fact that decides the resolution, and the vault has it on the shelf (Duverger). Largely a verbal dispute written into the Neg’s own definition of allegiance (conditional, “not a blood oath”), which the Aff eventually names himself; underneath it a dual-standard failure — two live definitions of logical (instrumental vs. internal-coherence), each side scoring under his own. Finding 9 at 12 of 12 |
| ✅ | Is It Impossible to Get Rich Without Luck? | Calvin Constable vs. The Aftermath | → review; the bracket’s first universal quantifier resolution, and the Aff wins by emptying “luck.” Harvest: irreducible ≠ decisive, buy the ticket, the determinism pincer |
| ✅ | Is AI Making People Dumber? | Luke Charsky vs. Dr. Philip Bunn | → review; the control case for finding 4. Extended Mind vs. phronesis; the vault’s three-layer rule cuts between them |
| ✅ | Should Free Speech Protect Hate Speech? | Chase McPherson (Neg) vs. Octavius Thunder (Aff) | → review; 1:09, moderator Kyla Turner (her sixth). The resolution gets vacated from both ends — the Neg specifies down into existing doctrine, the Aff concedes that doctrine, and the live fight migrates to hate-crime enhancement, which is sentencing rather than speech protection. Confirms finding 9 at 11 of 11; supplies the finding-1 etymology pair (Kyla blocks a definitional move without owning it, where Pisco owned his); and the two cases that would have settled it — R.A.V. and Mitchell — go unnamed. Result pending |
| ✅ | Is the Algorithm Fair? | Silvio Cruz vs. Ryan Hamm | → review; 0:57, shortest round. The pre-registered hook — “‘fair’ is a weighting question” — was confirmed independently by Chris’s compression argument. Aff wins on a severance (algorithm = automated program, so all moderation evidence is off-topic); Neg names the winning inside/outside frame in his first answer and abandons it. Result pending at review time |
| ✅ | Are Boomers Responsible for Today’s Economy? | Rex Jones vs. Mariah Meyers | → review — analyst read, not discussed (Chris opted out of this one). Answers open question 3 with layer 0; full-tape confirmation of Pisco on finding 1; already spawned Generational Attribution, which now has it as a specimen |
| ✅ | Is Modern Western Culture Making People Fragile? | Nic Turner vs. Rowan Bourdeau | → review; 1:11, moderator Sam Tripoli (the fifth). The bracket’s contrast case: the matched pair to Guptill/Tejada — same question, another degree word — but here the term is pinned mid-round and the debate immediately acquires shape. Neg concedes the data, the mechanism, and “environment makes the man,” then severs culture from circumstances. Result pending |
| ✅ | Is Therapy Culture Making People Weaker? | Josh Smith vs. Haji Ouedrago | → review; 1:10, moderator Kyla Turner (her fifth). The bracket’s best definitional work — both load-bearing terms pinned, and the Neg catches a question-begging definition in real time (refusing “therapy culture := the overuse of therapy” as smuggling the conclusion), which supplies a third manipulation type for The Load-Bearing Word. Aff wins ground by severance (CBT ≠ therapy culture) and then concedes the causal claim — “I’m not saying it’s the direct cause” — which the Neg fails to collect. Result pending |
| ✅ | Does Western Therapy Work for Men? | Spencer Summerhays vs. Michael Gilkison | → review; 1:14. Two defective definitions of works (provider-carries-everything vs. any-learning-counts); the Aff’s market argument has a hole neither man finds. Moderated by Ryan Mullally — answers finding 1’s open successor. Result pending |
| ✅ | Are We Too Sensitive as a Society? | Forrest Guptill vs. Jose Tejada | → review; 1:20, longest round so far. The Aff argues the Neg’s case for ~20 minutes before the opponent corrects him — and after the correction they go on agreeing. The bracket’s first verbal dispute |
14 of 16 first-round matchups visible on the channel as of 2026-08-05; two not yet listed.
Production note — 2026-08-17, and it is a genuine anomaly. The same two matchups are still unlisted twelve days later, while round 2 is already being filmed — reported to Chris directly by contenders, not visible from the channel. Two rounds of Thunder 32 have therefore not aired, not been queued, and not been announced, yet the tournament has advanced past them.
Sourcing: contender word-of-mouth via Chris, uncorroborated against the site or channel. Recorded because the coverage policy already flags this information channel as part of the series’ evidence base, and because it is the kind of fact that never appears in a transcript.
The release cadence, which is what makes it odd. Rounds have been posted on a 48-hour cycle, and on that schedule the two missing matchups were due today. So this is a schedule slip against an otherwise regular pattern, not a gap of unknown length.
Why it matters to the findings, as hypotheses rather than conclusions:
Coverage status as of 2026-08-17: all 14 aired rounds reviewed ✅ — round 1 coverage is complete up to the two matchups that have never been listed. Only those 2 remain.
Ana Kasparian vs. Pearl Davis · Michael Rectenwald vs. Shabbos Kestenbaum · Ryan Mullally vs. Pisco.
Update (2026-08-11): WW1 has happened. Moderating the therapy round, Mullally introduces himself as “the winner of the first ever Word War Debate in Atlantic City” — so the event this hub logged as upcoming is in the books, and Mullally won. Stated by him on air; the site’s /ww1-debate-results page would confirm the full card and is not yet pulled.
Chris: “let’s wait for more results and then we can break apart why people won or lost.”
The bracket is complete as of 2026-09-01 — fourteen winners, twelve margins. Run the cross-round causal analysis now, rather than another per-round review. The materials are assembled and this is the payoff pass:
Harvest to run in the same pass — three things the intros and outros give up cheaply. Chris’s addition: “note if we can catch the real names vs internet handles from the rereads.” All three are read straight off the moderator’s scripted segments, so they cost nothing extra once a transcript is open:
@_vigilante_tv; Octavius Thunder → Eli Halpern). Harvesting all 28 makes the reach analysis runnable for the first time, because follower counts become addressable. Build it as a table: billed name · handle · platform · channel, with unknowns marked rather than guessed.Do all three in the same read; they come from the first ninety seconds and the last two minutes of each tape.
Status 2026-09-01: 14 of 14 winners in hand, 12 of 14 margins. Therapy culture and the feminism rerun came from the round-of-16 bracket; their cards were never published and never will be, so those two enter the autopsy as winner-known, margin-unknown — codable for every predictor except the margin-based ones. The earlier advice stands with its reason changed: run it on all fourteen now, and run it before the round-2 topics drop, because the two-stage protocol needs finding 17’s model coded and registered the moment they do.
Chris: “this should help me make better predictions for the 16 and up brackets. I have not been able to understand the audience.”
That reframes the pass. A retrospective why did each side win produces fourteen stories, each plausible and none testable. What is wanted is a model that pays out on round 2, which means the fourteen rounds are a training set, not a set of anecdotes — and the discipline that follows is: score candidate predictors against all fourteen outcomes at once, keep the ones that beat the base rate, and discard the ones that only explain in hindsight.
The leading candidate is now finding 17’s two-factor model — default to the audience’s prior on the resolution; override only on a large execution gap; rhetoric moves the margin, not the winner — which fits 11–12 of 12 retrospectively and must now be tested prospectively rather than refined further. The autopsy’s job is to try to break it, not to decorate it. Remaining candidate predictors, each cheaply codable per round:
| Predictor | Coding | Currently supported by |
|---|---|---|
| Argument quality (vault verdict) | Aff / Neg / neither | Charsky +58, Brunet +57 |
| Delivery & legibility | which side was easier to follow | Owlish +48, Garcia’s 25.8 |
| Uncollected concessions | did the winner leave admissions on the table? | Bourdeau +47, Ouedrago |
| Audience prior on the resolution | which conclusion flatters this pool | Tejada +89, Aftermath +44 |
| Affiliation / turnout | personal network size, not follower count | finding 13’s ballot counts |
| Structural edge | last word, opener, moderator’s rule | finding 9’s corrections |
| Conceded the room | did a debater signal he expected to lose? | David S., pending |
The one calibration already in hand — and the honest starting point for “I don’t understand the audience” — is finding 15: the misses are one-directional. Every wrong ballot call but Hamm came from expecting rhetoric, sympathy, affiliation, or a flattering conclusion to beat the better argument, and this pool keeps declining. So the first correction is not a new variable, it is a weighting change: stop discounting argument quality. The audience has been more legible than the model applied to it — the difficulty was never that they are unpredictable, it is that the prior was cynical.
Success test, set in advance: the model earns its keep if it beats 3-of-7 on round 2, and it should be registered before the round-2 tapes are watched, or it is not a prediction.
Chris: “this new format needs lots of polish!”
Consolidated because the observations have accumulated across a dozen reviews and are more useful as a list than scattered through findings. Ordered roughly by how cheaply each could be fixed.
| # | Defect | Evidence | Cheap fix |
|---|---|---|---|
| 1 | ✅ FIXED at round 2 — side assignments are now published in the video description, and have been for all eight round-2 and all four round-3 matches | the round-2 and round-3 descriptions name the Aff and Neg explicitly | done |
| 1b | Speaking order still contradicts the intro order. The Neg is introduced first in 12 of 12; the Aff usually speaks first. Two signals, opposite directions, neither flagged | Tejada argued his opponent’s case for a fifth of a round | Say which side is which before the clock starts — the bios already state the position |
| 2 | Closing order follows THREE different rules — two of them deterministic and opposite. Kyla: opener closes first, 5/5. Perez: opener closes last, 4/4. Mullally: side-based, and self-inconsistent | finding 9’s correction table plus its third correction | Pick one and put it in the format sheet. The last word is a real edge, and the bracket’s three narrowest rounds went by 12.4, 16.8 and 9.0 |
| 3 | Load-bearing terms are not negotiated pre-round — only the topic noun is | The feminism rerun was decided by this before it started | Require an agreed definition of the resolution’s operative term, or announce on air that none was reached |
| 4 | No live bracket. Staggered windows plus image-only results mean nobody outside the organizers can see the standings. Still true at round 3 — the results page has round headings and no tree | The reason this hub maintains one, and why the projection had to be derived rather than read | A single standings page — the data already exists, and an outsider reconstructed the whole tree from one list |
| 5 | ⭐⭐ The results page has no memory — it is a current-state display, and on 2026-09-08 it was wiped. Every one of the ~19 percentage cards, round 1 included, was replaced by a single Final Four bracket graphic. The QF B and QF C margins were never published and now cannot be, and the entire margin distribution — the instrument behind findings 11, 13, 17 and 18 — survives only because this hub transcribed each card on the day it appeared. Original defect, still true: results posted late, out of order, and four never at all | the whole card archive, deleted; one false negative earlier caused by checking too early | Publish on close, and keep an archive. Failing that, anyone tracking this must transcribe every card the day it appears |
| 6 | Matchups unlisted while later rounds film — and two slots were filled with no ballot at all. The block structure of the round-2 pairings shows the tree was fixed at seeding, so nothing prevented publishing it | two of sixteen never announced; Medeiros and Horseradish advanced unopposed | Publish the full bracket at seeding |
| 9 | The published prompt and the spoken prompt differ — silently reallocating the burden of proof. (Chris reports having seen title/prompt mismatches across the bracket, so this is a recurring production habit rather than a one-off — watch round 3.) The amnesty round is billed as a pairing (“Amnesty vs. Mass Deportation”, burden shared) and read on air as a proposition (“we should give amnesty, not deport”) with an assigned affirmative. The two debaters argued different versions and nobody reconciled them | one debater carried a burden the other did not know existed | Read the description’s prompt verbatim, or publish the one that will be read |
| 8 | Topics are assigned without regard to the pairing. The suffrage round put a resolution in front of two contenders who share a worldview; the affirmative opened by disclosing he was a devil’s advocate, and neither man defended the proposition | “the lesson here is a more contentious topic” — both debaters engaged in good faith and the round still had no contest in it | Before assigning, ask the pair whether they actually disagree with the proposition |
| 7 | Moderators are drawn from the competitor pool | Pisco moderates and competes; Mullally moderates after winning WW1 | Not obviously wrong — but it makes disclosure load-bearing |
| 10 | Contenders get roughly 12 hours’ notice of the resolution, and it shows. In the abortion round both men failed their five-minute opening in opposite directions — the Aff mis-clocked the format (“I have five minute. I thought I had one minute”) and improvised four minutes; the Neg used two and yielded, then said in his closing “there’s some things I probably should have studied a little bit more.” Neither brought a single medical fact to an abortion debate | round of 16, on the most-litigated proposition in American politics | Give the resolution at seeding. The bracket was fixed on day one; the topics could have been too |
| 16 | ⭐ A six-hour ballot is not a smaller sample of the same electorate — it is a different one. The window ran 48 hours in round 1, 24 in round 2, and 6 for the Final Four and the final. Chris: “the 6 hour voting window might mean anything goes.” Against the confirmed dilution result, a shorter window does not merely shrink the pool — it shifts who is in it, from people who watched the debate to people who were online when it dropped, so the ballot increasingly measures list speed. The format has compressed by 8× heading into the rounds that decide a $5,000 championship | SF1 and SF2 both on 6-hour ballots, both over a compressed schedule | Restore 48 hours for the decisive rounds, or accept that the last rounds are the noisiest |
| 15 | ⭐ The last word is free, and contenders have worked that out. Closings are unanswerable by construction, so substantial new argument in a closing costs nothing and gains everything. Now at three consecutive rounds — QF C, QF D, and SF2, where the Aff’s entire positive case arrives in his closing after he had answered “no, I don’t have it” when asked for it on the floor. Compounded by defect 2: which speaker gets that free shot is decided by whichever closing-order convention the moderator happens to run | three consecutive rounds; the round-2 Iran round makes four | Bar new arguments in closings, or give the first closer a short reply |
| 14 | ⭐ A prompt can present non-exclusive alternatives as exclusive — “Nationalism vs. Socialism.” Chris, before the tape: “I am not sure nationalism and socialism are opposites!!” He is right, and it is structural rather than pedantic. They answer different questions: nationalism answers what is the relevant political community?; socialism answers who should own productive property? Those are orthogonal axes with occupants in all four quadrants — the actual opposite of nationalism is internationalism, and of socialism, capitalism. So the format assigned each contender a position on a different axis and told them to collide. Distinct from defect 9 (published vs. spoken prompt) and from the unquantified comparative: here the pairing itself is neither exhaustive nor exclusive | QF D, the narrowest round in the tournament (+9.0) | Pair opposites on one axis, or state a proposition |
| 12 | ⭐ The 24-hour ballot is too short, and round 3 ran it over a holiday weekend. Round 1 gave voters 48 hours; round 2 halved it, and round 3 kept the shorter window while airing on Sunday 6th and Monday 7th September — Labor Day weekend, with the live final on the 12th. Chris: “these 24-hour windows hurt the format, not enough time for fans to get in, esp on a holiday weekend.” It compounds with finding 11’s now-confirmed dilution result: a shorter window means a smaller pool, and a smaller pool means a wider margin — so the format shortened the ballot in exactly the rounds where the stakes were highest | round 3’s four ballots all ran ≤24h across a holiday; QF A resolved on ~550 views, QF B is pacing at 20/hr | Go back to 48 hours, and never run a decisive round against a holiday |
| 13 | ⭐ The calendar was under-scheduled early and then compressed against a fixed live date. Round 1 aired across early-to-mid August; round 2’s first tapes did not appear until 31 August — roughly a two-to-three-week gap — after which rounds 2, 3 and 4 plus the final were squeezed into twelve days. Chris’s read: “my guess is this was rushed after the long gap after round 1.” This is the root cause behind several defects already logged separately — the 24-hour windows (12), cards posting late and out of order (5), round 3 airing while round 2’s coverage was in progress, and two of four quarterfinals landing on near-identical Iran resolutions because four topics had to be produced at once | ~3 weeks idle, then 8 matches in 12 days | Fix the live date first and schedule backwards from it |
| 11 | Byes are unannounced, unexplained, and have now decided four advancements. Three matchups across two rounds simply did not happen; no announcement, no forfeit notice, no card. Round 2’s took Chris Brunet — the contender whose round this page called argument quality’s best specimen — out of the tournament without a contested loss | round-2 match 3; round-1 matches 15 and 16 | Announce a withdrawal when it happens. A bracket with a silent 12% no-show rate is not measuring what it says it is |
⭐⭐ A hypothesis that may reclassify defects 3, 8 and 14 at once — Chris, 2026-09-09: “I bet the debaters get to negotiate the prompt before the debate. Common in these types of events is to rattle off topics until they find disagreement and go with that — it would also explain Summerhays’ nationalism-vs-socialism topic, as this one was also very odd.”
A search that terminates at the first genuine disagreement produces exactly the artefacts in this table. The resolution then records where two particular men happen to split, not a clean opposition — so odd pairings (“Nationalism vs. Socialism”) and over-wide wordings are the expected output rather than drafting errors. SF1 is the clearest specimen: “all drugs should be legal” is stretched around a disagreement that is really only about opioids, and both debaters say on the record that they agree about psychedelics. The evidence for the hypothesis is Thunder’s own “I did take a risk by choosing the prompt.”
⚠ If it holds, the cheap fix changes shape. Defects 8 and 14 stop being “assign topics more carefully” and become “require a narrowing step after the disagreement is located” — the negotiation has already done the hard part (finding real disagreement) and then fails to write it down accurately. ⭐ It also sharpens defect 3 into something odder than it looked: the contenders apparently negotiate the topic and then do not negotiate the operative term inside it, which is the half that actually decides rounds. Unconfirmed — it needs a contender or an organiser to describe the process; nothing in the published material does.
📌 Defect 17 — ⭐⭐ results are now withheld rather than published, and the prior season’s archive is winner-name-only (2026-09-09). The results page moved (/voting-results → /view-results), lost every percentage card, and shows a Final Four bracket with neither semifinal winner advanced; /vote-now-1 reads “check back soon.” Meanwhile /word-war-i-results shows what the archive looks like once a season ends: a name, no number, and broken images. Cheap fix: publish the count and the split at ballot close, in text, on a permanent per-round URL — the reveal can still be a reveal without destroying the record, and text costs nothing to keep. Detail: above.
Chris, closing the season:
“WWD tries to market themselves as something like ‘the prizefight of debate’ — basically they want the showmanship of boxing or MMA, but for verbal combat. This tends to promote more of a bloodsports format… I think we saw this happen in real time with Jay vs. Stefan. Jay’s combative style seemed to have more audience appeal as he promoted that style.”
Read the list again with the prizefight frame and it stops being a list of mistakes:
| Defect | The prizefight reason for it |
|---|---|
| 8, 14 — prompts chosen badly, non-exclusive alternatives | a card needs clash, and clash is selected for over clarity |
| 13, 12, 16 — compressed calendar, shrinking ballots | a card needs a fixed live date, and everything upstream compresses toward it |
| 5, 17 — results wiped, then withheld, then deleted | the reveal is the product; an archive competes with it |
| 18 — the coin toss | that is the boxing convention, imported without its compensating rules |
| (new) — a commentary desk between argument and ballot | every prizefight has one |
⭐⭐ So the memo changes. “Publish your margins on permanent URLs” asks a promoter to damage his own product. The fixes that survive are the ones a prizefight already wants — and that turns out to be most of them.
Boxing and MMA are not improvised. They have unified rules, weight classes, published judges’ scorecards, an official record for every fighter, and a sanctioning body. The showmanship sits on top of an unusually rigid administrative substrate — that is what makes the drama legible.
Word War Debate has adopted the showmanship and almost none of the infrastructure. No written rulebook (defects 1b, 2, 3, 15), no published scorecards (5, 17), no fighter records, no schedule (13).
⭐ So the bloodsport framing is the strongest argument for the fixes, not an excuse for their absence. “If you want to be the UFC of debate, notice that the UFC publishes every judge’s scorecard.” That is a better memo than anything else on this page, and it is entirely in the promoter’s own language.
The tape documents it: a super-chat / Rumble-rant segment mid-debate, with questions read out on air for anyone who paid. ⭐ So the questions that reach the stage are selected by willingness to pay, which selects for engagement rather than relevance — and it sits alongside the commentary desk as a second uncontrolled input between argument and ballot.
And Dyer vs. Molyneux is the specimen. Molyneux says out loud that he is playing to the room — “which I see by the smiles” — and loses in a runaway to the more combative man. ⚠ Chris’ balance is the right one and this page keeps it: he would have preferred the civil version, and Molyneux’s wandering monologues were a genuine failure to engage. Both are true, and Dyer’s procedural objections were largely founded.
⭐⭐ Which means combat was correct play. If an audience ballot rewards combat, a contender who brings it is not lowering the tone — he is reading the rules accurately. Style selection is not a flaw in this format; it is the format’s output, and finding 6 is what it looks like from the inside.
Chris: “WWD prefers the more bloodsport style while I tend to review the more substantial ones.”
📌 Worth filing against every cross-debate finding here. The debates collection is not a random sample — it is selected for substance by the reviewer. So findings about what wins, what loses, and what audiences reward are drawn from a corpus chosen on a correlate of the thing being measured. ⚠ This compounds the steelman asymmetry from finding 28: the reviewer’s prior shapes not only how a round is scored but which rounds enter the record at all.
Chris: “The championship round was the best of the evening as it was a good mix of substance and thrill on a rather interesting topic too.”
That is the important closing fact. Jones vs. Thunder ran the Nuremberg Code, two Supreme Court patent cases, a topicality frame and a consent trap — and it was also the most entertaining thing on the card, in a room that had paid to be entertained.
⭐⭐⭐ So substance and showmanship are not the trade-off the format’s branding implies. The best round had both. Which means the defects were never the price of the style — they were just defects, and the format’s own championship is the proof.
Chris, 2026-09-09: “The defect page got long, as this is their first attempt. They needed a real schedule and sticking to it. It means debaters must be on the ball if they want to participate, and it also gives the time required for each round to be fair. They literally jammed 13 debates within a week IIRC; the 6-hour voting windows for the SF seems bad form… but tbh, aside from the time constraints, the later rounds were much clearer and cleaner.”
Taken from the airing dates on the round pages:
| Block | Debates | Aired |
|---|---|---|
| Round of 16 | 8 | 3 – 10 August |
| — idle — | 0 | 10 – 31 August (21 days) |
| Round 2 | 7 (one bye) | 31 Aug (3), 2 Sep (3), 3 Sep (1) |
| Quarterfinals | 4 | 6 Sep (3), 7 Sep (1) |
| Semifinals | 2 | 8 Sep (2) |
| The crush | 13 | 31 August – 8 September (9 days) |
⭐⭐ So: 36 days from first tape to last, and 21 of them idle in a single block. The schedule did not lack time — 21 debates over 36 days is one every 1.7 days, which is slower than round 1 actually ran (8 in 8 days) and leaves room for 48-hour ballots on every round with slack to spare. ⚠ The three-week gap and the six-hour ballot are the same decision. The idle fortnight was the resource the compressed rounds needed, and it was spent before the crunch arrived, against a live date that was fixed the whole time.
This is why Chris’s fix is the right one and is not merely administrative. A published schedule does two things no amount of good intent replaces: it filters for commitment ex ante (“debaters must be on the ball if they want to participate”), and it equalises prep time across rounds, which is a fairness property, not a convenience. The current arrangement gave round 2 contenders ~12 hours’ notice of their resolution and semifinal voters six hours to cast a ballot — and both numbers were set by a calendar nobody had written down.
Seventeen items reads as a catastrophe. It is not; it is five problems, and one of them causes a third of the list:
| Root | Defects | Fixable by |
|---|---|---|
| ⭐ The calendar | 6, 10, 12, 13, 16, 17 | a published schedule — Chris’s fix, and it retires six items at once |
| Rules that were never written down | 1b, 2, 3, 15 | ⭐ a one-page document. Speaking order, closing order, definition negotiation, and whether new argument is allowed in a closing. Costs nothing, depends on nothing else, and would survive any schedule |
| Record-keeping | 4, 5, 11, 17 | publishing counts as text on permanent URLs |
| Prompt design | 8, 9, 14 | a narrowing step after the disagreement is located — if the negotiation hypothesis holds, this is a procedure gap rather than a drafting one |
| Conflict of interest | 7 | disclosure; arguably not a defect at all |
⭐ The second row is the free win. It is the only cluster with no dependency on money, calendar, or venue — four of the seventeen defects are a page of rules nobody has typed, and two of them (closing order, definitions) have decided rounds.
The record supports it. Round 1 produced layer-0 epistemic non-overlap and verbal disputes with no definitions. By round 3 the bracket had its first topicality argument and its first grant-and-outweigh, and SF1 is the best-argued round of the tournament and the only one where both sides brought a case.
⚠ But the confound has to be named: that is what a tournament is for. Half the weak debaters are gone after round 1 and three-quarters after round 2. Later rounds being better is the bracket working, not the format improving — and the two are easy to confuse when they arrive together. Separating them:
| Improved? | Evidence | |
|---|---|---|
| The debating | ✅ markedly | topicality, grant-and-outweigh, real clash — but selection explains most of it |
| Production | ✅ one real fix | defect 1 — side assignments published from round 2, held for all 12 later matches |
| Moderation | ❌ no convergence | finding 21 — two founders moderated round 3 and behave nothing alike; closing order still ran on three different rules |
| Ballot integrity | 🔻 degraded | 48h → 24h → 6h |
| Transparency | 🔻 degraded | cards wiped, then results withheld entirely |
| Prompts | 🔻 arguably worse | “Nationalism vs Socialism”; “all drugs” stretched over a disagreement about opioids |
⭐ So the honest season verdict: the debating got better and the production did not — and in three measurable respects it went backwards. That is a real finding rather than a grumble, because it separates what a second season should keep (the bracket, the crossfire round, the contender pool) from what it must fix (the calendar, and the page of rules nobody wrote).
And the encouraging half of Chris’s read stands: the format’s problems are almost entirely procedural, which is the cheap kind. Nothing on this list requires more money, a bigger audience, or better debaters. It requires a calendar and a rulebook — and one of those is a single page.
📌 Defect 18 — ⭐⭐ a coin toss allocated both the first and the last word, and nobody noticed (2026-09-14). In the championship the toss was announced as deciding who opens. Under this moderator’s closing rule (the opener closes last) it also decided who closes — so the Negative opened the debate, defined the term, assigned both burdens, pre-butted his opponent, and then had the unanswerable final word. Chris: “they still did the first-last / last-first ordering as if the first speaker was the affirmative… there was more to it than just order.” ⭐⭐ Confirmed 4-for-4 across the whole WWD II card (2026-09-14): the toss winner opened and closed in every debate — Jones, Lynn, Dyer and Mockler — and three of the four won, with the closest round of the night going to the man holding both ends. ⚠ And the two non-propositional rounds are the worst case: “Angels vs. Demons” and “God! vs. God?” have no affirmative and no burden of proof, and the burden is the entire justification for the first-and-last privilege — so there it compensates for nothing. Even where a burden exists, the toss ignores which side carries it. Cheap fix — and it is Chris’s, and it is already in the format’s own history: “keep the order for openings and closings.” That is Kyla’s round-1 rule, applied 5-for-5 before Perez inverted it: whoever opens also closes first, so the second speaker gets the last word as compensation for hearing the case first. Symmetric, costless, and abandoned. This compounds defects 1b, 2 and 15 instead of sitting beside them.
Chris, closing the measurement thread:
“There is another debate format where they poll the audience at the start and then at the end, and measure by % of people moved — so this is your test done in another format. Though % moved might not be a perfect record: someone who has 95% of the vote has a hard time getting another 1%… measured in absolute % points or relative %, might have troubles at the extremes.”
That is the Intelligence Squared design, and the connection he draws is right: a pre-poll is the prior, collected before the verdict, which is exactly the finding 28 protocol run on an audience instead of a judge panel.
Every complaint this page has filed about ballot quality is a complaint about a level ballot.
Finding 5 — the better-argued side loses. Finding 10’s correction — the pool’s axis decides rounds. The SF1 result — the best debater in the bracket lost on a topic his opponent’s axis owned. All of these say the same thing: a level ballot measures where the room already was, and the room was fixed before anyone spoke.
⭐⭐⭐ A movement ballot cancels the axis by construction, because the axis is baked into the pre-poll. Whatever the room believed walking in is subtracted out, and what remains is what the debating did. That is a direct measurement of execution — the thing this bracket spent fourteen rounds failing to measure.
⚠ And it is robust to the pool-selection problem that broke the reach instrument twice: a self-selected audience makes a level ballot nearly uninformative, but movement within a biased pool is still movement.
| Metric | Fails where | Why |
|---|---|---|
| Absolute points gained (what IQ2 used) | ⚠ the top | start at 95% and only 5 points exist to win; the metric structurally rewards whoever starts behind |
| Relative gain (% of own starting share) | ⚠ the bottom | 1% → 2% is a 100% gain; the metric rewards whoever starts tiny |
Two fixes, and the second is the standard answer to exactly this problem:
📌 Filed as the cheap fix for defects 16 and 18 and the only proposal on this page that would test finding 5 rather than merely restate it. ⚠ It costs one extra poll and nothing else.
In fairness, three things the format does well, and they should survive any polish: the moderated crossfire round is the most productive segment in almost every round (it is where definitions actually get pinned); the ballot window lets voters watch the whole tape rather than reacting live — though round 2 halved it to 24 hours, which cuts against that virtue; and disclosure norms are emerging on their own — Mullally opens the party-loyalty round with “candidly, I know both these guys.”