Track record · The Ledger

What we got right, and what we got wrong.Every claim, its status, and the evidence that moved it.The claim registry, both columns.Claim registry. Machine copy: /s-claims.json.

There's no stamp that makes something true — only evidence, and it can point either way. So we keep a list with both sides: the wins and the mistakes, with dates. A crossed-out claim stays visible on purpose. There is no mark of truth. There is only evidence, and it points both ways — so this is kept as a ledger, with entries on both sides: what we proved, what the evidence turned against, what we had to correct. A REFUTED claim is a claim with evidence against it, not a claim that has stopped being a claim. Typed registry of staked claims. Gate counts come from campaign logs; survival and reversal proportions are being computed from those same logs and will be published rather than asserted — until then this page shows entries, not rates. Corrections to the site itself are logged separately in the errata. Entries below; rates pending computation from campaign logs — do not infer base rates from an excerpt. Site corrections: /errata.html. Overturn a staked claim and the entry carries your name and the date.

Open1,241 claims gated to date · survival and reversal rates pending — computed from the campaign logs, published here when they exist, not before.

Both columns

Proved
Transversal Theorem
Every valid base with an unsaturated coordinate is capped-infeasible. Machine-checked, exact arithmetic. Union-closed sets campaign.
Refuted
The Dichotomy conjecture
Evidence against: a counterexample. The anchored cube becomes capped-feasible at d = 8. The Coverage conjecture went with it. Entered the day it was found.
Corrected
~~D*(p) ≤ f(p) for all primes~~ → D*(p) > f(p) at finite p
A prior compaction claim disproved; a new elementary bound proved and numerically verified. The original stays visible, struck through. That is the point.
Refuted
Five cross-domain bridges in the Atlas
Evidence against: the edge list. 154 typed edges, exactly one crosses a domain. Overturned by counting, which is the cheapest verification there is.
Verified
One person can generate research at institutional scale
Evidence for: the corpus itself — 657 base records, 235 expansion, one human directing the pipeline.
Open
Verification, not generation, is the binding constraint
Our founding bet. Standing. Still a bet. If the evidence turns, it gets an entry on the other side of this page — and we will not move it quietly.

Why keep the mistakes up? Because a page with only wins tells you how we advertise, not how we think. What you should take from this page isn't that we're always right — it's that our claims are the kind of thing that could be shown wrong, and sometimes are, in public, with the date attached. Why a ledger and not a list of results. A page of only successes is a page about our publishing habits, not our epistemics. Both columns are evidence; neither is a verdict about the world. What a reader should take from this page is not that we are right, but that our claims are the kind of thing that could be shown wrong — and sometimes are, in public, with the date attached. Why a ledger. Success-only pages measure publishing habits, not epistemics. Both columns are evidence; neither is a verdict about the world. Falsifiability is enacted here rather than asserted: staked claims carry standing challenges, and an overturning entry is credited and dated. The verdict vocabulary on this page adds REFUTED — evidence against — which is distinct from the report-scope UNSUPPORTED, evidence not found; the full schema is defined in the methods. Schema note: REFUTED (evidence-against) is Ledger-scope; UNSUPPORTED (evidence-for not located) is report-scope. Definitions: /methods.html. Registry: /s-claims.json.