Enforcement cost is what has kept liability defaults apart from conduct commands. Automation makes patrolling free and silently promotes the first into the second — changing what a rule is without changing a word of what it says.
Links: Registration and Scope Creep, Mens Rea in Libertarian Law, Government Formation, The Technology-Governance Gap, Technology → Culture → Politics · Specimen: Flock Cameras — Summerhays vs. Rex Jones · Drafting-stage twin: Lazy Law — the cheap-measurement gradient applied when the rule is written, not when it is enforced
A legal system has two independent parameters — how much the code forbids, and how expensive it is to catch you — and every society in history has been safe because only one of them was ever turned up at a time.
Chris: “It has gotten to the point where just walking outside your house is a violation of a half-dozen laws, but these are not enforced because of the understanding that the laws are either obsolete or only there to resolve differences… I would prefer a world where our laws were minimal, simple and clean, and it was ok to automate them — but this is a big ask. Minimal, simple, clean is next door to complete totalitarianism.”
The gap between prohibited and punished is the slack. It is not sloppiness. It is doing structural work: it is where discretion, mercy, context, proportion and ordinary life fit. And it exists almost entirely because enforcement costs money — an officer’s hours, a prosecutor’s docket, a jury’s patience, a filing cabinet’s finite size.
Automation is the first technology that lets a society hold the code thick and drive enforcement cost toward zero at the same time. That combination has no historical precedent, which means there is no accumulated experience to reason from and no worked example of it going well.
| Expensive enforcement | Cheap enforcement | |
|---|---|---|
| Thin code | pre-modern; under-governed, disputes unresolved | Chris’s preferred world — and see below |
| Thick code | where we actually live. Tolerable because the slack absorbs the excess | ⚠ the empty cell. Never occupied at scale, now reachable |
The bottom-right cell is what a surveillance layer over an existing legal code produces. Nothing about the law changes; everything about your exposure to it does. A rule that has been on the books unenforced for forty years does not become a new rule when a camera starts detecting it — but it becomes, for the first time, a real one.
This is why the intuition Chris reports is correct rather than reflexive: society is enthusiastic about automation everywhere else and nervous about it here, and the nervousness is tracking something true. Elsewhere automation lowers the cost of producing something we want more of. Here it lowers the cost of punishing, and we have been quietly consuming the high price of punishment as a liberty.
Chris’s own escape route is to thin the code until automating it is safe — the appeal of a Misesian / first-principles legal order, grounding law in a small set of derivable rules and nothing else — and he immediately doubts it. The doubt is well placed, for three reasons:
So the honest position is Chris’s: “we want law to create good behavior, but we don’t want it so restrictive and efficient that it prevents behavior.”
Chris: “We could build cars that don’t speed, but is that really reasonable? I think not.”
The intuition is right and the reason is worth naming: law has historically operated by deterrence, and automation enables prevention, and these are different things morally as well as practically.
| Deterrence | Prevention | |
|---|---|---|
| What it does | you may do it and face consequences | you cannot do it |
| Agency | preserved — the choice is real | removed |
| Moral status | compliance is a choice, therefore creditable | compliance is a fact, therefore empty |
| Hard cases | speeding to a hospital is possible, then defensible | impossible; no exception handling |
| Failure mode | under-enforcement | no override, at scale, forever |
A speed limit backed by tickets leaves you a governed agent. A car that physically cannot exceed the limit leaves you a passenger. The second eliminates the offence and the virtue together, and it removes the emergency case with no mechanism for reinstating it. Chris’s “I think not” is a judgement that agency is worth some quantity of violations — which is exactly the trade the slack has been making silently all along.
The most underrated consequence, and the one that makes “just enforce the law as written” incoherent.
Automated enforcement is not applied evenly across the code. It arrives wherever detection is cheapest, and detection cost has nothing to do with severity. A licence plate, a speed, a red light and a parking meter are trivially machine-legible. Fraud, coercion, negligence, assault and abuse of office are not.
So a surveillance layer does not deliver the law, enforced. It delivers a re-weighted code in which the offences a sensor can see become the offences that reliably carry consequences, while everything requiring inference, testimony or intent stays as under-enforced as it ever was. The effective legal system drifts toward whatever is easy to measure — which is the same failure mode as any metric that becomes a target, applied to the criminal law.
⚠ And it compounds the first problem: the machine-legible offences are disproportionately the trivial ones, so the slack is removed precisely where its removal is least justified.
⚠ Refined again below — Chris subsequently corrected this reading too, and the liability-default framing supersedes it. Both passes are kept because the progression is the argument: slack-as-protection → slack-as-repair-of-bad-law → the law is fine and its *mode* is being changed.
Chris’s correction, and it changes what this page is about.
“I have often done a rolling stop in a right turn lane and turned right under red because there was no traffic going in that direction. Technically illegal, absolutely no harm done — and this is a failure of law, not enforcement.”
The page above treats the slack as protective — something being consumed as a liberty and about to be taken away. That is not wrong, but it is the wrong level of description. If a rolling right-on-red into an empty intersection is harmless, then the rule prohibiting it is simply incorrect at the margin, and non-enforcement is the system silently repairing a drafting error.
So: slack is not a good thing we are about to lose. It is a repair mechanism for defective law — and its protective character and its discriminatory character are both side effects of repairing by discretion rather than by amendment. The danger of automation is therefore sharper than “it removes something valuable”:
Automation removes the correction while leaving the error in place.
⭐ This partially dissolves the open question this page registered against Registration and Scope Creep. Slack is neither primarily a protection nor primarily a discrimination — those are the two faces of unofficial repair. The reason the same phenomenon looks like liberty from one angle and arbitrariness from the other is that a repair administered by discretion is unevenly distributed by construction. Both descriptions are downstream of the same defect: a rule that is wrong, and a system correcting it informally instead of fixing it.
May we automate this rule? — Would you be willing to see it enforced perfectly, every time, against everyone? If the honest answer is no, the rule is defective and automation does not create the problem; it exposes one that was already there.
⭐ And that is the optimistic reading, which this page owes as the counterweight to everything above. A law that can only survive by not being enforced has never actually been tested. Automation is a brutal but genuine law-quality audit — it converts drafting defects from invisible to intolerable, and a code that has accumulated forty years of dead letters has never had to justify a single one of them. The failure mode is not that automation arrives; it is that it arrives without the repeal that should accompany it.
Chris: “The system will gravitate towards the cheapest options, so we live in a world where every traffic violation is reported, but the big crimes are never touched. Sounds a lot like where we are.”
Correct, and it upgrades the machine-legibility argument from a prediction to an explanation. The re-weighting is not a future risk of surveillance technology — it is a description of the present, and the technology only accelerates a gradient the system was already sliding down. Enforcement resources flow toward whatever is cheapest to detect and prove, so the offences that reliably carry consequences are the ones a sensor or a form can establish, while fraud, corruption, regulatory capture and abuse of office — which require inference, testimony and intent — remain effectively optional.
That is the same structural failure as any metric that becomes a target, running on the criminal law: the measurable subset becomes the operative code, and it does so without anybody legislating the change.
Chris offers the strongest objection to his own position:
“We probably should still try to make enforcement cheaper so we free up resources to go after bigger crimes.”
It is the right instinct and the mechanism does not work, for a reason worth stating: automation does not free capacity, it creates demand. Detecting a thousand times more violations does not reduce anyone’s workload — each detection generates a citation, a contest, a hearing, a collection — and all of that new load lands in the trivial-offence lane, which is the only lane the technology can see. To convert cheap detection into freed capacity you would have to decline to act on most of what you now detect, which is precisely the discretion automation was adopted to eliminate.
The route that actually frees resources is repeal, not automation. If the aim is to stop spending enforcement on rolling right-on-reds, the instrument is deleting the rule, not detecting the violation more efficiently.
Chris: “I think the option to break the law is necessary. This is the reason juries exist. We have countless examples showing that no legal system can account for everything… so the jury is there to discern the edge cases.”
This is the institutional half the page was missing. Not all slack is the accidental residue of enforcement cost — some of it is built, and the jury is where. A jury’s acquittal is unreviewable: double jeopardy bars retrial, courts decline to instruct jurors that they may refuse to convict, and no appellate mechanism can correct them. That is not an oversight. It is the one place where a legal system has institutionalised the proposition the rules got this case wrong, and then deliberately made the override uninspectable.
Two consequences:
Until now this page has measured slack by size — how wide the gap between what the law says and what enforcement does. That is the wrong axis, and Chris’s objection to drug decriminalization is what exposes it:
“Ugg. Something is either legal or not legal. Just because your local body doesn’t enforce something doesn’t make it legal. The federal government would be perfectly valid in enforcing drug laws in the ‘legal’ states — but I think they want to see how it all turns out first.”
He is legally correct, and the case is Gonzales v. Raich (2005): the Controlled Substances Act reaches purely intrastate, non-commercial cultivation, so it applies in “legal” states and always has. What keeps federal enforcement out is not law but discretion — the Cole Memo (2013), rescinded in 2018, plus an annual appropriations rider covering only medical use. ⚠ And as of 28 April 2026 the incoherence is written into the schedule itself: DOJ/DEA moved FDA-approved marijuana products and products under a qualifying state medical licence to Schedule III, while everything else — including the recreational market that is most of the industry — stays Schedule I. Federal legality now turns on which piece of paper your state issued you. That is not “legal,” not “illegal,” but a third thing.
⚠ Here is the tension, and it is a real one. “Something is either legal or not legal” is a demand that text and enforcement match — and perfect matching is exactly what automated enforcement delivers. The objection to decriminalization and the objection to flock cameras appear to point in opposite directions.
⭐⭐ They don’t, and the jury section above already contains the resolution — as a criterion this page should have stated outright:
Slack is legitimate when it has a designated holder and a public procedure. It is illegitimate when it belongs to whoever happens to hold the office.
| Who holds it | How exercised | Verdict | |
|---|---|---|---|
| Jury nullification | jurors, appointed for the case | in public, case by case, defendant present, unreviewable by design | ✅ designed slack |
| Prosecutorial non-enforcement / decriminalization | whoever holds the office | categorically, in advance, revocable at will — and revoked in 2018 | ❌ undesignated |
| Automated enforcement | nobody | the gap is deleted for everyone | ❌ the other failure |
So the objection to decriminalization is a rule-of-law objection, not a pro-enforcement one, and it is fully compatible with hating the cameras. What is wrong with declared non-enforcement is not that the gap exists — the jury’s gap is the same size — but that nobody was appointed to hold it, it operates categorically rather than case by case, and it leaves a standing class of people who are guilty at someone’s discretion. Automation is the same defect run the other way: it takes the gap away from everyone, including the people the system deliberately gave it to.
⭐ This also repairs the page’s own framing. The reframe held that slack is a symptom rather than a feature. The corrected statement is narrower and survives better: accidental slack is a symptom; designed slack is a feature; and undesignated slack is a third thing that is neither, and is what most policy arguments are actually fighting about.
I answered pigeonhole. Chris’s own version is stronger and I was reaching for the weaker argument:
“Action is infinite, laws are finite, so no law can account for all actions. It felt the same as the math showing a simulation of reality is greater energy than just reality, and that a computer cannot solve the ECC as it would take more information to recreate it than just live it. I guess I saw law as a simulation that must live in reality.”
That is not a cardinality argument, it is an information-theoretic one, and it is the right shape. Pigeonhole says a finite code must group together cases that differ — true, but it is only about coarseness, and it invites the reply “then write more rules.” Chris’s version says something much harder to escape: the information needed to specify the right answer is generated by the cases themselves and does not exist in advance to be encoded. More rules do not help, because the shortfall is not resolution — it is that you are trying to compress a process that is still running.
⭐ And the ECC connection is exact, not merely evocative. The Mises–Hayek calculation problem is precisely this argument about prices: the information a central planner would need is dispersed, tacit, and continuously created by the very activity being planned, so it cannot be assembled ex ante — an in principle limit, not an engineering one. A legal code is a central plan for conduct, and it faces the identical objection: no code written in advance can encode the correct disposition of every future case, for the same reason no plan can price every future good.
Which produces the interesting implication, and it is not one this page went looking for:
The common law is the market solution to the calculation problem in law. — spun out to its own page at Chris’s request: Common Law as a Discovery Procedure, which carries Hayek’s nomos/thesis distinction, four disanalogies stated up front, and the sharper form of the automation worry: an enforcement layer does not repeal the discovery procedure, it starves it, because disputes stop reaching adjudication once the machine has disposed of them.
Case-by-case adjudication is a discovery procedure. It does not demand the information in advance; it generates it locally, at the point of the dispute, from the parties who actually hold it — and the jury is the local-knowledge holder. Statute is central planning; precedent is price discovery. That is the same axis as Aggregation vs. Sorting — voting aggregates, markets sort — with adjudication in the sorting column.
⚠ Where the analogy strains, stated rather than skipped. Prices are a continuous signal with a tight feedback loop; legal outcomes are discrete, sparse, and feed back slowly. And law has a virtue prices do not need in the same way: you want to know the rule before you act. Predictability is an argument for codification and against pure case-by-case discovery, so the rule-of-law virtue and the calculation argument pull in opposite directions — which may be the real reason every mature system runs both, statute over common law, rather than choosing.
(Chris’s verdict — *“not worth the math effort” — is right about chasing a theorem. It is not right about the argument: this is a genuine claim with a rigorous cousin, not a hunch. The jurisprudential statement remains H. L. A. Hart’s open texture; the tempting-but-wrong tool remains Gödel, and the vault’s own Gödel Governance Problem is the page that warns about reaching for it.)*
Chris corrects his own rolling-stop example, and the correction is sharper than either of our first passes:
“I think a rule ‘no turn on red’ is a good one — but the purpose is to place blame if my judgement is wrong and I do create an accident. The error here is enforcing this as wrong at every instance. I could also use jaywalking as the example — fine if it is across an empty road.”
So the rule is not defective. Its mode of operation is being changed. “No turn on red” is an excellent rule for deciding who pays when there is a collision, and a bad rule for issuing a citation to someone who caused no collision. Same text, two entirely different instruments.
This page already had the distinction — a rule can be an adjudication default rather than a conduct command — and then mis-classified Chris’s own example under the wrong heading, reading the rolling stop as evidence that the rule was wrong. It is not. The rule is right and it is being run in the wrong mode.
That gives the thesis its tightest statement, and it supersedes the framing above:
A liability default is a rule that answers “who is at fault?” once harm has occurred. A conduct command is a rule that answers “may I do this?” before anything happens. Enforcement cost is what has kept the two apart — you cannot patrol what you cannot afford to watch. Automation makes patrolling free, and so silently promotes every liability default into a conduct command, changing what a rule is without changing a word of what it says.
A red-light camera is not a more efficient traffic officer. It converts a fault-allocation rule into a prohibition, and nobody legislated that conversion.
The earlier test — would you accept this enforced perfectly? — is serviceable but blunt, because it collapses two different failures. The sharper question is prior to it:
Is this rule a conduct command or a liability default? Only conduct commands are candidates for automated enforcement. Liability defaults should be invoked when harm occurs and never patrolled.
Applied to the specimens: don’t defraud people is a conduct command and automating its detection would be unambiguously good (and is exactly what is hardest to automate). No turn on red and don’t jaywalk are liability defaults — worth having on the books precisely so that the person who misjudges an empty intersection and causes a crash is the one who pays, and worth never enforcing against the person who judged correctly.
⚠ And this re-scopes the “defective law” reading rather than replacing it. Both failures are real and they need separating:
| What automation reveals | Example | Right response |
|---|---|---|
| A rule that is genuinely wrong at the margin | a dead-letter provision nobody would re-enact | repeal |
| A good rule being run in the wrong mode | no turn on red; jaywalking | classify it — and exempt it from automated enforcement |
The first failure calls for legislative courage. The second calls only for the statute to say which kind of rule it is — which is cheap, and which no code currently does.
Chris’s objection, and it lands on this page’s own proposal:
“This is also a threshold argument, which makes it tough for law. We know the case where right turn on red is wrong and we know where it is right — but what about edge cases? You don’t cause an accident, but it causes the next five cars to slam on their brakes. Legal? This is what makes law tough, and why people would rather take clear lines and deal with the ‘injustices’.”
The five-braking-cars case breaks the test I proposed two sections up. “Invoke liability defaults only when harm occurs” sounds decisive until you ask what counts as harm. No collision, no injury, no property damage — and five drivers who had to react to a hazard that should not have been there. Near-miss? Externality? Nothing at all? The trigger is as vague as the rule was, so the proposal does not eliminate the indeterminacy; it relocates it from the prohibition to the predicate.
That is worth conceding plainly rather than patching, because it is the same limit the rest of the page keeps running into: whether this rolling stop was reasonable is exactly the local, tacit, case-generated information the calculation argument says cannot be encoded in advance. The threshold problem and the calculation problem are one problem seen from two directions.
The threshold problem has a twin that arrives from the other direction, and Chris raised it about drug legalization: does an overdose have to affect everyone else?
“From a grand perspective yes — the universe is a closed system, everything interacts with everything else. Does this mean anything for a society? No. I think there has to be a different measure.”
That is the collapse of but-for causation, stated exactly. Every act has a causal path to every other, so “it affects others” licenses regulating anything — the non-uniqueness objection in its general form. A principle that proves everything decides nothing.
⭐ The law hit this wall long ago and built a second test to cut the chain: proximate cause. Palsgraf v. Long Island Railroad (1928) is the canonical case — the fireworks really did cause the scales to fall, in a but-for sense, and the court held that is not the kind of causation liability runs on. The doctrine exists for no other reason than that but-for causation proves too much.
And it is a standard, not a rule. There is no formula: it asks whether this was the kind of harm, to the kind of person, that makes attribution fair — decided case by case, after the fact, usually by a jury. ⭐⭐ So the answer to “what is the different measure?” is: the law has one, it works, and it is precisely the sort of thing that cannot be encoded. The threshold problem and the causation problem are the same problem, and the legal system solved both the same way — by refusing to write a rule and appointing someone to decide.
⚠ Which is the sharpest form of this page’s thesis. An enforcement layer that can only apply rules cannot apply proximate cause at all. Automating enforcement does not merely favour machine-legible offences — it is structurally incapable of the doctrine the system uses to stop every prohibition from swallowing everything.
Three tests that do sharpen “affects others” beyond the closed-system triviality, all of them standards rather than rules:
| Test | Question | Applied to the overdose |
|---|---|---|
| Severability (Coase; Chris’s “1. don’t do that”) | can the cost be routed back to the actor short of banning the act? | yes — pricing or coverage, so the cost does not justify the ban |
| Chain of free choices | how many independent human decisions sit between act and harm? | the medical bill routes through a policy choice to pool costs; drunk driving’s harm routes through nobody’s choice but the driver’s |
| ⭐ Reciprocity of risk (Fletcher, 1972) | is the risk mutual, with both parties taking the benefit? | no — and this is the strongest form of the case against, because traffic risk is reciprocal and addiction’s socialised cost is not |
⭐ The reciprocity test is the one that bites the driving analogy, and the answer is instructive: the fix for non-reciprocity is to restore reciprocity, not to prohibit. Mandatory auto insurance is exactly the instrument that makes traffic risk reciprocal in cost terms — you pre-pay your expected externality — and a Pigouvian excise is its analogue for a consumed good. Which gives the whole regulatory design space as one inequality:
expected social cost ≤ tax ≤ cost of evasion
Chris supplies the limit case from the right-hand side: “in California when they became legal, it actually caused a spike in drug crime, as the taxes were so high people circumvented the legal routes.” Set the price above the cost of evasion and the illicit channel reappears — legal supply, illegal supply, and a tax base funding neither. California tested that boundary and repealed the cultivation tax in 2022. ⚠ And when the left-hand side exceeds the right, no pricing regime exists, and prohibition-style arguments start to be defensible on their own terms. That is a measurable version of is the juice worth the squeeze — and a far better question than the one the debate actually had.
| Rule (no turn on red) | Standard (drive with reasonable care) | |
|---|---|---|
| Fit to the case | poor at the edges — over- and under-inclusive by construction | good |
| Predictable before you act | yes | no |
| Cost to administer | low | high |
| Who decides | the drafter, in advance | the adjudicator, after |
| Failure mode | known injustices at the margin | discretion, unpredictability, capture |
A rule is a compression that accepts a known error rate to buy predictability. A standard refuses the compression and pays in uncertainty and discretion. Neither is correct in general; the choice per domain is a judgement about which cost is worse there.
⭐ And Chris’s point is that the choice is deliberate, which this page had not credited. “People would rather take clear lines and deal with the injustices” is a revealed preference for predictability over case-fit — not inherited sloppiness, not a drafting failure, but a considered trade repeatedly re-made. The bright line is chosen knowing it will be wrong at the edges, because knowing where you stand is worth more than being right in every case.
Chris, on the licensure/graduated-privilege model:
“Licensure, graduated privilege and the like are good solutions to what we have. People will argue them (‘who controls the licensing!’) — but it gets to the point that there are other options. FWIW the whole class system was an attempt at this, so maybe the argument is to revisit and maybe even subclass to allow for finer control.”
⭐⭐ This breaks the table above, and in the right direction. Rule and standard are not two options; they are the two ends of a continuum, and the free parameter is how many classes the code partitions the conduct space into.
| Classes | What it is | Fit | Predictability |
|---|---|---|---|
| 1 | a bright-line rule — no turn on red | worst | perfect |
| a few | a schedule — licence classes, drug schedules, tax brackets, tariff codes | better | good |
| many | a subclassed schedule with endorsements | better still | falling |
| one per case | a standard — reasonable care | perfect | none |
A rule is a schedule with one class. A standard is a schedule with a class per case. Everything real is in between, and “subclass for finer control” is a proposal to turn the dial — which is why it feels like a different kind of answer than ban it or allow it. It is: it is a move along an axis the ban/allow argument never mentions.
Chris is right that turning it reduces error. Doubling the classes roughly halves the distance between a case’s actual risk and the class it is assigned to. You get more boundaries (N classes have N−1 of them, each with its own threshold problem) but each misclassification is smaller. On error terms alone, finer is better and the limit is a standard.
⚠ So what stops it? Two costs, and they are the ones that set the optimum:
⭐⭐ And the CSA schedules are the specimen of getting this wrong in a way finer subclassing would not fix. Five classes for thousands of substances is coarse, but the deeper defect is that the schedules are a scalar: one ordinal ranking for a risk profile that varies on at least four independent dimensions — dependence liability, lethal-dose margin, behavioural toxicity, and medical utility. Marijuana sitting in Schedule I with heroin while fentanyl sits in Schedule II is not an error in the ranking; it is evidence that no single ranking exists. Adding classes to a line that is the wrong shape buys nothing.
⭐ Driving already solved this, and the solution is the one Chris is reaching for. A licence is class + endorsements — Class C plus motorcycle plus hazmat plus passenger. Endorsements are dimensions; schedules are a scalar. The right reading of “subclass to allow for finer control” is therefore not more rungs on the ladder but more ladders: a substance carries a vector of restrictions (prescription-only, quantity-limited, supervised-administration, no-take-home) rather than a rank, and a person carries a vector of privileges. That is a schedule that can be finer without becoming longer.
Chris anticipates the objection and it deserves better than dismissal, because it is the criterion from the jury section arriving upstream. A permission regime is a prohibition with a gate; building the gate relocates the discretion from the enforcement layer to the licensing layer. And a licensing board holds slack categorically, in advance, in private, with no defendant present — which is the profile this page just marked ❌ for prosecutorial non-enforcement.
So what makes driver licensure benign and taxi medallions a scandal? Not the fact of licensing. Four structural properties:
| Property | Filter (✅ driver’s licence) | Quota (❌ medallion, certificate-of-need, liquor licence) |
|---|---|---|
| Criteria | objective, cheaply verified, near-zero discretion at issue | discretionary, or a fixed number |
| Issuance | uncapped — ~90% of adults hold one | capped, so the licence becomes an asset |
| What it filters | competence | competitors |
| Revocation | procedural, appealable, for cause | rarely at issue — the value is in the scarcity |
⭐ The test: is the licence a filter or a quota? A filter with objective criteria and unlimited issuance is a competence check. A capped licence is a cartel with a government stamp, and it will be captured, because the rent is the point.
That is a real answer to “who controls the licensing?” — the controller matters much less when there is nothing scarce to control. And it connects straight to Registration and Scope Creep: a filter that is later capped converts into a quota without any change to the statute, which is this page’s whole thesis about mode-change-without-text-change, arriving in a third form.
Everything above concerned which offences get enforced. The threshold problem shows automation also re-weights the form of law itself:
A machine can apply a rule and cannot apply a standard. No turn on red is trivially automatable; drive with reasonable care is not automatable at all.
So an enforcement layer does not merely favour the machine-legible offences — it favours the machine-legible form, pushing every domain it touches toward bright lines and away from standards. And bright lines are precisely the form whose failure mode is the known-injustices-at-the-margin that discretion used to absorb.
That is the whole thesis in one sentence: automation increases the rate at which law produces edge-case injustice, while removing the mechanism that was quietly cleaning it up. Not one effect but two, working the same direction — which is why “just enforce the law as written” is not a neutral proposal even when the law is good.
⚠ This also sharpens the classification proposal rather than rescuing it. Marking a provision as a liability default keeps automation off it — a real gain — but does nothing about the threshold vagueness in whatever triggers it. The proposal buys the first problem and not the second.
The affirmative case in the flock-camera round rests on it: an officer may lawfully observe you in public, so why not a device? Four disanalogies, none of which is about privacy in the naive sense:
Red-light cameras are the rehearsal for all of this — the same technology, the same public-space justification, the same objections, twenty years earlier and at smaller scale. The strongest form of the traffic-camera critics’ position is the one Chris endorses: law enforcement does not automatically get the latest tool; it has to establish that using the tool is constitutional first.
Not a solution — the tension is real and Chris is right that it does not dissolve. But the framing generates design constraints that are more tractable than “how thin should law be?”:
This page argues automation makes enforcement too cheap, converting liability defaults into conduct commands. Semifinal 1 supplies the same thesis from the opposite end, arrived at independently by a debater who has never heard of it.
Medeiros, on the Florida pill-mill epidemic: “So what stopped it? Not the market. Not consumer choice. Enforcement.… And here’s what I want you holding: that off switch only existed because the whole thing was still technically illegal. There was a prescription to falsify, a licence you could revoke, and a statute to charge. My opponent’s resolution deletes all three.”
Legalization removes the instrument; automation makes the instrument free. Both are arguments that the enforcement apparatus — not the rule text — is the load-bearing variable, and that changing it silently changes what the law is while leaving what it says untouched.
⭐ The two semifinals of that bracket ran the two halves against the same audience inside twenty-four hours: the cameras round was won by arguing enforcement capability is dangerous, so never build it (“you have to say your side is going to be in charge forever”); the drugs round was argued on its being essential, so never delete it. Neither debater noticed they were making the same claim with the sign flipped — which is the clearest available evidence that the variable is real and that nobody has a name for it.
The criterion above held that undesignated slack is illegitimate because nobody was appointed to hold it. Chris supplies a stronger reason, from the commercial side:
“We often see companies start ‘borrowing’ policies when it turns out another company gets away with it.”
The enforcement gap is itself a public signal. One firm adopts a risky practice, nothing happens, and the rest copy — no agreement, no communication, just a free experiment whose result everybody can read. So discretionary non-enforcement does not produce a quiet exception; it produces an industry norm, by the mechanism worked out in Coordination Without a Coordinator.
⭐ Which means the choice is never between enforce and quietly don’t enforce. It is between enforce and publish a loophole.
⚠ This also sharpens the automation argument from an unexpected direction. If unenforced rules decay into collective practice, then the pre-automation world was not as slack-rich as this page implies — some of the gap was being continuously consumed by whoever noticed it first. The people who lost the most when enforcement became cheap were the ones already exploiting the gap at scale, not the driver doing a rolling stop. That is worth holding against the page’s own sympathies.