Vault work lives at two altitudes: a specimen (a bounded, dated, source-anchored artifact — a debate review, a game project, a transcript study) and a thesis (a portable, source-independent idea the specimen feeds or tests). Keep them in separate pages — the separation is correct, not a leak. Connect them with a two-way ledger. The downward half of that ledger (thesis → its specimens) is the half that rots, and three failure-clusters prove it.
Links: The Anchor Method (this is the developed form of its “seam discipline” bullet), Force Doctrine — The Theory (the specimen that does it right), Online Debates · failing clusters: Monopoly frontier theory, MoO optimal strategy, MoM tier system
Discovered, not designed — surfaced by auditing why a “lost” debate discussion turned out to be correctly promoted to a separate theory page, then stress-tested against three game clusters that all failed the same way.
| Specimen | Thesis | |
|---|---|---|
| Answers | “what is this particular thing?” | “what’s true in general?” |
| Anchored to | a source (video, repo, transcript), dated | nothing external; portable |
| Lifecycle | frozen once reviewed | accretes across many specimens |
| Lives in | debates/, projects/ |
research/, philosophy/, method/ |
| Example | wilson-rights-dont-exist-only-force.md |
force-doctrine.md |
The instinct to fold everything into the specimen (“the discussion went missing!”) is wrong — the discussion’s portable content was promoted to its proper altitude, which is a feature. A specimen is evidence; a thesis cites evidence. Different jobs, different pages.
This is the same shape as atomic skills: one idea per page, ledgered to its evidence, pointed at by everything that needs it — easy reference, no duplication. A thesis page is to its specimens what a skill is to its callers.
Separation is automatic (different folder, different altitude). Connection is a discipline — a ledger maintained in both directions:
force-doctrine.md is the existence proof: it cites all five Wilson debates, each with a contribution line, and every debate back-links up. Fully reciprocal.
The audit’s central finding: upward links are mostly healthy; the downward ledger is mostly absent. The cause is structural, not laziness:
A hub that predates its evidence never gets the back-link. When a thesis page is created as the harvest target (a “doctrinal spin-off,” like
force-doctrine.md), its author writes the downward ledger by reflex. But when specimens cite a pre-existing thesis (risk-and-entrepreneurship.md,battle-value.md,bilateral-trade-valuation.md), nobody goes back to the older page to add the reciprocal entry. The loop is left open at the top.
So the rule: when a specimen cites a thesis, the same edit adds the specimen to that thesis’s downward ledger. Both ends, one move.
frontier-trade-theory.md does exactly this).feedback_*) feels like promotion and isn’t — the audit found four game-theses that reached memory but never a vault page. Memory is a recall hint, not the durable artifact. If it’s worth a memory note, it’s worth a thesis line too — do both, and link them.Edge case — external-repo specimens (a seam-breaker, discovered 2026-06-08). When a project moves to a standalone repo (three did on 2026-06-05), its in-vault page becomes a pointer-stub and the specimen physically leaves the vault. Preserve both ledger ends anyway: the stub links up, and the thesis lists the repo as its specimen down. The 2026-06-05 extraction silently severed pygone’s only link to its grounding theses (principled-llm-code, repairing-llm-code) — so extraction doesn’t just risk the rot, it causes it by default. Re-audit every seam after any extraction.
(This page eats its own dogfood: here is its downward ledger.)
llm-agents-across-games.md) is linked one-way only, and ≥4 explicitly-“generalizable” theses sit trapped in one page. Contributes: proof that a great upward ledger does not save you — promotion and the downward half fail independently.risk-and-entrepreneurship.md) has an empty downward ledger; the generalized MIRR / 4X-as-capital-allocation thesis is trapped in optimal-strategy.md even though a sibling game already depends on it. Contributes: the “hub predates its evidence” mechanism in its purest form.feedback_* memory but never a vault page. Contributes: the memory-as-release-valve discovery, and the external-repo-specimen edge case.Run it as a lens on the vault. It found the gaps (the Monopoly orphan, the MoO empty hub, the MoM memory-valve). The test it must pass: filling those gaps makes each affected thesis queryably two-way, and a re-audit finds nothing newly trapped.
First closed case (2026-06-08). The MoO/MoM MIRR harvest was promoted to MIRR — 4X Strategy as Capital Allocation, wired down to its specimens (MoO optimal-strategy + mirr-analysis + the optimizer project, MoM tier-system) and up to its parent risk-and-entrepreneurship — whose empty downward ledger was filled in the same pass — and an independent re-audit returned PASS on all four conditions. The method closed its first case. Remaining backlog: the shared capability-without-leverage / hard-counter-to-zero thesis (Monopoly + MoM, still memory-only); the Monopoly live-LLM orphan (llm-agents-across-games); the MoM external-repo specimen needing an in-vault pointer.
A method that names a gap it can’t close is just prose — same bar the Anchor Method sets for itself; this one closed its first.