Chain 5 of 5 · the method chain

How would you know if you were wrong?

This is the chain that earns trust, and until now it's been invisible. How to read a history that destroyed its own evidence, what happened when the claims here were actually checked, and which of them still can't be.

Ends on: which claims are unfalsifiable.

5.1

How do you read a history that was destroyed?

The destruction is selective, not uniform — which is what makes it readable.

Deeply canalised things survive because the lip is high, so what comes through the filter is biased toward frozen optima — exactly the class of thing you can solve backwards. The same operation that destroys the information is what preserves the readable part.

The inversion the whole method turns on

A many-to-one map loses the input and gains a readout of where the minimum is. An extinction destroys information about what was and creates information about what must be. Not conservation, not loss — transduction, contingent to structural. The record is readable not because some of it happened to survive, but because the destruction was itself the measurement.

One tempting move, refused. Landauer's bound (erasure costs kT ln 2 per bit) is true and astronomically loose — an end-Permian-scale event dissipates enough to permit erasing roughly 10⁴⁴ bits, and the biosphere doesn't contain that. The ceiling does no work; a physicist would spot it in one line. What works instead is phase-space contraction, where the contraction rate equals the entropy production rate for driven dissipative systems (Evans & Searles 2002; Gallavotti–Cohen).

→ What happens when you check your own claims against this standard?

5.2

What happens when you check your own claims?

Four wrong first authors, and a prediction that turned out to already be published.

Twenty-four sources verified at source produced four wrong first authors — Tjhung→Kenchel, Powner→Singh, Aida→Ando, Bzdok→Stanley — multiple framing errors, and over forty findings that improved the drafts. Two claims were withdrawn publicly rather than quietly amended (both are in /claims/). A prediction made in one session had already been published four years earlier, by people with no stake in it — recorded as a prior-art surprise rather than absorbed.

Why the error count is the credential, not the embarrassment

A verification pass that found nothing would be evidence of a verification pass that didn't happen. The generated-output rule this runs on: anything produced by a research-synthesis system enters the register as recalled, not verified, regardless of how many citations it carries, and can't be promoted without a primary source being read by something accountable. A generated output naming a reference is a lead, not a verification. Full register at /method/.

→ Can a method claim be verified rather than trusted?

5.3

Can a method claim be verified rather than trusted?

Sometimes — and two negative results, plus one blind prediction that held, are the demonstration.

A set-valued pairwise adjacency descriptor cannot represent oligomeric closure, because it discards multiplicity and locality; and chain count and stoichiometry annotation do not establish topology — 1CR0 is annotated "hexameric" and is a 6₁ screw-filament excerpt. Published as A set-valued structural descriptor cannot represent oligomeric closure, DOI 10.5281/zenodo.22073289.

The blind prediction, and the observation about method itself

The replacement descriptor made a blind prediction that held: 1E0J's closure was established by all-atom contact before the fragment pipeline saw it, then reproduced edge-for-edge with zero spurious edges.

And a small novel observation about method itself, independent of anything it's applied to: the single-commit property of a pre-registration document makes contemporaneity machine-verifiable via git log. A reader doesn't have to trust that a criterion predates the analysis it was applied to; they can check it.

→ Which of this framework's own claims are currently unfalsifiable, stated plainly? Three soft spots are already named rather than hidden: endogenous extinction needs its population-genetics defence actually stated; the claim that persistence-filtering leaves a universal, recognisable trace is circular until it's broken by an independent diagnostic that could, in principle, fail to find one; biology-as-sole-record must cross-validate against something outside biology somewhere. Tracked, unresolved, at /open/.

Where this chain landed

Published: the method paper itself, DOI 10.5281/zenodo.22073289 — plus the full epistemic-controls register this chain draws from, at /method/.

Open: naming, plainly, which of the framework's own claims can't currently be falsified. This is the one chain whose ending is a question about the whole rest of the site.

← Chain 4 — why this chemistry and not another? Open loops →