The record

Prediction registry and scoring record

Numerical predictions are SHA-256 hashed and published before the deciding computation, with explicit falsification bands and stated confidence. A seal the run cannot falsify is refused by the apparatus, not filed anyway. The scoring itself has twice been wrong, and both corrections are recorded here rather than repaired quietly.

Registry status

Zero independent verifications

82 kill tests declared, 7 claims refuted internally, and 0 independently verified. — **Eighty-two declared kill tests, seven claims killed, and not one killed by anybody but us.** The third number is the institute's most important open problem. /api/v1/ledger.json

Those figures carry over from universesolver.com, SIMARC's predecessor registry, and they are displayed exactly as they stand. Zero independent verifications is not a footnote — it is the diagnosis. Authority in research consists of other researchers staking their name on a result. Everything in this institute's design, especially the agent interface, exists to make that number climb.

Verify something →

P-01 · Substrate Physics

Seam scorecard

Six necessities any real computer must have, six seams it would leave, six verdicts. This is the only program-level ledger with entries decided.

seam	test	evidence	verdict
grid in space	ℓ=4 cubic harmonic power	756k quasars	NULL
grid in time	v_g = c·cos(ka/2)	GRB 090510	NULL
memory — entanglement	the 2ᴺ wall	live frontier	NO SIGNAL
memory — finite box	S₁/₂ · quadrupole · parity	WMAP + Planck, 4 cleanings	ROBUST ANOMALY
an RNG	PRNG periodicity / structure	quantum bit streams	NULL
arithmetic	round-off · irreversibility	precision bounds	NULL

# the one entry that refuses to die — and every attempt made to kill it the anomaly ....... missing large-angle CMB power, ~0.6% level, cut-sky S₁/₂ > 3σ reproduced in ..... 4 independent sky-cleanings · WMAP AND Planck periodic box ...... DEAD — killed by circle searches walled box ........ DEAD — no parity signature corner observer ... DEAD — the one geometry that DOES write an odd sky (~5× ΛCDM) fails its own anti-mirror kill-shot on real data the "axis" ........ fragile · T-E polarization inconclusive verdict ........... a real pattern with no working physical explanation. Reported as an anomaly. Not as a discovery.

P-02 · P-03 · P-04

Programs with no sealed predictions

In each of the three younger programs the sealing apparatus is built, tested, and — in two cases — has been demonstrated refusing a real prediction the run could not have falsified. Nothing has been committed to any of them.

QUANTUM COMPUTER · P-02 — prediction ledger: empty. Sealing apparatus built and tested, with a pre-registered statistical power gate. 16/16 claims certified (7 PROOF · 6 EXACT · 3 STATISTICAL) — but a certificate is a re-verification, not a sealed prediction, and the two are not interchangeable.

CANCER · P-03 — prediction ledger: empty. The seal gate was demonstrated refusing a real underpowered prediction: detecting HR 1.60 requires 143 deaths; the pilot's key subgroup had 17. One target (T1) was killed by its own gates before it consumed compute — its premise, that robust undrugged dependencies are under-studied, was refuted: they are roughly twice as well studied as random genes.

DIMENTAI · P-04 — prediction ledger: empty. Two experiments recorded, both calibration: the trial graveyard (classifier passing 44 controls in both directions) and the causal rung (positive and negative controls run before use). Two of three targets are blocked on data access and written down as blocked.

An empty ledger reported as empty is worth more than a full one that quietly excludes its misses. This page will get less clean over time, and that is the point.

Corrections

Corrections are first-class ledger entries. They are never edits: the record is append-only, and a repaired mistake with no trace is indistinguishable from a mistake never made.

what	correction
prediction scoring	Wrong twice. Both corrections recorded in the ledger rather than repaired quietly.
error rates, P-02	Every error rate in the program was a factor of two too small. Caught by the apparatus before any result existed.
a validation test, P-02	Had exactly zero statistical power against the failure it was written to catch.
"−8", P-02	A substrate result nearly reported as a number, before a guard caught that two ranks were being subtracted across different vector spaces.
graveyard classifier, P-04	A negation bug had reclassified the largest anti-amyloid failure in the corpus as a business decision. Caught by a control list that existed only because a CANCER lesson demanded one.
the APOE gate, P-04	p = 5.75e-659 underflows a double to 0.0, and the gate asserting p < 1e-50 was satisfied by it. The control passed on the bug.
the single-slit pattern, P-01	A razor-thin barrier produced a fake double-peaked pattern — the wave was scattering off the barrier's own under-resolved corners. The discretization invented physics.

Machine-readable

Machine-readable registry

Every entry on this page is available as JSON, and will be published as an Atom feed in Phase 1 so agents and humans can subscribe to seals and scores as they land.

GET/api/v1/ledger.jsonseals, verdicts, corrections, registry carry-over

GET/api/v1/claims.jsonthe claims registry, with kill-test pointers