Independent exact certification of machine-generated mathematics — exact arithmetic, no code shared with the claimant, refusal as a verdict. This page is the whole of it: what the machine is, what was built to make it true, and the live record underneath. Every number below was read off a record when this page was built, and every battery it reports green was executed during that build.
Published, not peer-reviewed, not independently rerun. Every claim below is rerunnable from the public repository; external reruns will be recorded here as they arrive — none has yet. Enclosures are proofs-of-object pending that independent verification.
Generation is crowded. Judgement is not. This machine decides mathematical claims — other people's and its own — in exact arithmetic, and publishes the refusals beside the verdicts. Five rules do all the work — the first is the one everything else pays for — and each is a measurement rather than a promise.
Absence of proof is never evidence of absence. A refusal here is terminal — never retried at lower rigour, never converted into a probability, never quietly dropped from the record — and that is the property a motivated party would not build.
| the idea | what it is | measured | where |
|---|---|---|---|
| The grade is the proof | A reward channel with no answer key: a model proposes an exact object, the grader re-derives it from whole numbers and returns CERTIFIED, REFUTED with the violated equation, or REFUSED. Nothing to leak, nothing to game. | 364 real-model proposals graded, 115 certified | /oracle/ |
| Refusal, counted by kind | The third verdict published as a rate with a denominator, and never merged across kinds — an instrument declining is not a claimant publishing nothing is not a budget running out. | 13.0% of 301 submitted claims refused | /reports/refusals.html |
| A certified enclosure is a canary factory | If a quantity is pinned to width w and a grader accepts anything within tol of a stored decimal, every value in the surrounding band is provably not the quantity AND passes. Adversarial submissions are minted from certificates instead of written by hand. | tolerance grader accepts 89.5%, certificate grader 0.0% | /reports/envs.html |
| An environment that rewards breaking things | The model is shown a grader and asked to break it. Ground truth is free because a certified enclosure decides both halves — and some rungs cannot be broken at all, so an auditor that always finds something fails half the ladder. | 3 of 7 rungs unbreakable by construction | /reports/envs.html |
| Evidence, not verdicts | A bare verdict scores zero however correct it is: HOLDS must ship a tiling whose every cell verifies, FAILS must ship a witness. Dressing a sampling grid as a tiling scores worse than abstaining. | the bluffing solver scores -6.00, below abstention | /reports/envs.html |
| Calibration before claim | An instrument may not state a new result until it has re-derived a published one at every run. The rule is enforced in code, not in discipline. | Mercer's λ(3) and this lab's λ(4) closed forms re-derived before any λ(5) claim | /reports/lambda5.html |
| Forgeries as measured soundness | "The grader is sound" converted from an assertion into a number: planted near-misses that must fail, in every battery, on every build. | 34 planted in the environments, 0 leaked · 21 in the audit lanes, all rejected | /reports/methods-note.html |
| Certificates that detach | A result travels without the machine that produced it: a JSON of exact numbers plus a verifier in the Python standard library, which must refute a deliberately forged value before it exits green. | 3 stdlib verifiers, no dependencies, ~1 second each | /verify/ |
| Pages born from records | No page here is written; every one is generated from the certificates it cites and re-derives its own numbers at build. A build that drifts refuses to ship. | this page, and every other | /reports/ |
| One rule, one module | A rule defined twice will diverge. The covering check that several theorems stand on lives in one module with four consumers, after it was written twice and disagreed. | instruments/covering · 4 consumers | /reports/ |
| The claims desk | Somewhere for a claim to go, with the verdict published whichever way it falls — and the submitted count published while it is still zero. | 18 decided, 0 submitted by others | /reports/claims.html |
None of these is a new species on its own. Exact arithmetic is the validated-numerics tradition; verifiable-reward environments are a category; kernel-checked proof systems have had no answer key for years. What has no neighbour is the composite — exact, independent of the claimant, re-runnable without the engine, refusal-bearing, and pointed at machine-generated claims. The one property in that list a lab cannot build for itself is independence, because it is the author of the claim.
Nothing in this section exists. It carries no numbers because there are none, and it is here so that the difference between what this machine does and what it intends is legible without asking. Each line moves into the table above on the day it has a record behind it.
A site that only ever shows finished work invites the reader to guess at the direction, and guessing is what this machine exists to replace. The fence is the price: no numbers, no screenshots, no dates, and no claim that any of it works — until it does, at which point it moves up one section and brings its record with it.
| family | what a hit asserts | generated | screened | certified → hit | stop |
|---|---|---|---|---|---|
| chowla-cosine | a finite integer set whose certified Chowla merit c = -min f_A / sqrt|A| falls below 1 | 400,000 | 1,579 | 1579 → 1508 | exhausted |
| erdos852-constants | a constant holding up the conjectured asymptotic h(x) ~ c0 log x on Erdős #852, replaced by a certified interval enclosure — and its published decimal decided exactly against that enclosure | 4 | 4 | 4 → 3 | exhausted |
| henon-census | a parameter pair (a,b) and period p for which the number of period-p points of the Hénon map is determined EXACTLY: every fixed point of H^p enclosed in a certified uniqueness box, the rest of the plane excluded by interval arithmetic | 328 | 328 | 328 → 328 | exhausted |
| henon-orbits | a parameter pair (a,b) and an explicit box in which the Hénon map provably has a periodic orbit of period p, and exactly one | 3,936 | 1,023 | 228 → 228 | exhausted |
| holmes-census | a parameter pair (d,b) and period p for which the number of period-p points of the Holmes cubic Hénon map is determined EXACTLY: every fixed point of H^p enclosed in a certified uniqueness box, the rest of the plane excluded by interval arithmetic | 124 | 124 | 124 → 124 | exhausted |
| keller-audit | an explicit polynomial map C^n -> C^n whose Jacobian determinant is proved constant by symbolic expansion over exact rationals, with certified distinct rational points sharing one image — the Jacobian conjecture refuted in dimension n, decided here and not trusted | 11 | 11 | 11 → 11 | exhausted |
| keller-fibers | a Keller map and a rational target with AT LEAST k certified preimages — pairwise-disjoint Krawczyk boxes found by blind multistart Newton on the exact map; k >= 2 re-proves non-injectivity with no witnesses consumed | 9 | 9 | 9 → 7 | exhausted |
| newman-minmod | an n-term Newman polynomial whose certified min|f| on |z|=1 exceeds every value achievable with fewer terms | 400,000 | 7 | 4 → 4 | exhausted |
| oeis-closedform | a published OEIS constant whose record — name AND fetched formula/comment fields — states no closed form, while its digits are consistent with a small closed form and every other form in the vocabulary is refuted | 14,677 | 14,593 | 14593 → 0 | exhausted |
| ramanujan-audit | a published Ramanujan Machine conjecture re-evaluated as a rigorous enclosure and decided against the claimed closed form — survival certified to the enclosure width, refutation proved (and, for their corpus, a discovery) | 52 | 52 | 52 → 51 | exhausted |
| strassen-audit | a claimed fast matrix-multiplication algorithm — a rank-r decomposition of the (n,m,p) tensor — whose defining identity (nm·mp·np exact equations) is VERIFIED over the claimed ring with r strictly below the naive nmp; the C-layout convention is detected and recorded, never assumed | 11 | 11 | 11 → 10 | exhausted |
The screen is float and may only ever prune; nothing is admitted without an exact certificate. A family plugs in by supplying six functions — enumerate, value, interesting, certify, key, statement — and inherits the loop, the scale and the dedup.
| family | object | certified enclosure | width | closed forms refuted |
|---|---|---|---|---|
| chowla-cosine | [1,2,4,6,7,8] | [0.649862827152, 0.649862827152] | 5.55e-16 | 715 / 715 |
| chowla-cosine | [1,2,3,5,6,7,8] | [0.715658879606, 0.715658879606] | 5.55e-16 | 716 / 716 |
| chowla-cosine | [1,2,3,5,7,8,9,10] | [0.715967581851, 0.715967581851] | 5.55e-16 | 716 / 716 |
| chowla-cosine | [1,2,4,6,8,9,10] | [0.725127302045, 0.725127302045] | 5.55e-16 | 718 / 718 |
| chowla-cosine | [1,2,3,5,6,8,9,10,11] | [0.725987311197, 0.725987311197] | 5.55e-16 | 718 / 718 |
| chowla-cosine | [1,3,4,5,6,9,10] | [0.742177616900, 0.742177616900] | 5.55e-16 | 718 / 718 |
| chowla-cosine | [1,4,5,6,9,10] | [0.742222719215, 0.742222719215] | 5.55e-16 | 718 / 718 |
| chowla-cosine | [1,2,4,5,6,9,10,11] | [0.750810171748, 0.750810171748] | 6.66e-16 | 718 / 718 |
| chowla-cosine | [1,2,3,4,6,8,9,10,11,12] | [0.759444373892, 0.759444373892] | 5.55e-16 | 717 / 717 |
| chowla-cosine | [1,2,3,5,8,10,11,12,13] | [0.759942805134, 0.759942805134] | 6.66e-16 | 717 / 717 |
| chowla-cosine | [1,2,3,4,8,10,11,12,13,14] | [0.763617416452, 0.763617416452] | 5.55e-16 | 717 / 717 |
| chowla-cosine | [1,3,4,7,8,10,11] | [0.763698607335, 0.763698607335] | 8.88e-16 | 717 / 717 |
| erdos852-constants | [erdos852-c0-enclosure] | [1.323228276864, 1.323228276864] | 2.22e-16 | 719 / 719 |
| erdos852-constants | [erdos852-c0-digits] | [1.323228276864, 1.323228276864] | 2.22e-16 | 719 / 719 |
| erdos852-constants | [erdos852-cstar-enclosure] | [0.075240386178, 0.075240386178] | 3.33e-16 | 703 / 703 |
| henon-census | [1.4|0.3|8] | [64.000000000000, 64.000000000000] | 0.00e+0 | 0 / 0 |
| henon-census | [1.34|0.3|8] | [48.000000000000, 48.000000000000] | 0.00e+0 | 0 / 0 |
| henon-census | [1.36|0.3|8] | [48.000000000000, 48.000000000000] | 0.00e+0 | 0 / 0 |
| henon-census | [1.38|0.3|8] | [48.000000000000, 48.000000000000] | 0.00e+0 | 0 / 0 |
| henon-census | [1.14|0.3|8] | [32.000000000000, 32.000000000000] | 0.00e+0 | 0 / 0 |
| henon-census | [1.16|0.3|8] | [32.000000000000, 32.000000000000] | 0.00e+0 | 0 / 0 |
| henon-census | [1.18|0.3|8] | [32.000000000000, 32.000000000000] | 0.00e+0 | 0 / 0 |
| henon-census | [1.2|0.3|8] | [32.000000000000, 32.000000000000] | 0.00e+0 | 0 / 0 |
| henon-census | [1.22|0.3|8] | [32.000000000000, 32.000000000000] | 0.00e+0 | 0 / 0 |
| henon-census | [1.24|0.3|8] | [32.000000000000, 32.000000000000] | 0.00e+0 | 0 / 0 |
| henon-census | [1.26|0.3|8] | [32.000000000000, 32.000000000000] | 0.00e+0 | 0 / 0 |
| henon-census | [1.28|0.3|8] | [32.000000000000, 32.000000000000] | 0.00e+0 | 0 / 0 |
| henon-orbits | [0.6|0.3|1|0.833333333] | [0.833333333333, 0.833333333333] | 2.01e-13 | 0 / 0 |
| henon-orbits | [0.62|0.3|1|0.825297414] | [0.825297414485, 0.825297414485] | 2.01e-13 | 0 / 0 |
| henon-orbits | [0.64|0.3|1|0.817519468] | [0.817519468482, 0.817519468482] | 2.01e-13 | 0 / 0 |
| henon-orbits | [0.66|0.3|1|0.809985304] | [0.809985304012, 0.809985304012] | 2.01e-13 | 0 / 0 |
| henon-orbits | [0.68|0.3|1|0.802681828] | [0.802681828468, 0.802681828468] | 2.01e-13 | 0 / 0 |
| henon-orbits | [0.7|0.3|1|0.795596939] | [0.795596939087, 0.795596939087] | 2.01e-13 | 0 / 0 |
| henon-orbits | [0.72|0.3|1|0.788719427] | [0.788719427131, 0.788719427131] | 2.01e-13 | 0 / 0 |
| henon-orbits | [0.74|0.3|1|0.782038893] | [0.782038893311, 0.782038893311] | 2.01e-13 | 0 / 0 |
| henon-orbits | [0.76|0.3|1|0.775545673] | [0.775545672898, 0.775545672899] | 2.02e-13 | 0 / 0 |
| henon-orbits | [0.78|0.3|1|0.769230769] | [0.769230769231, 0.769230769231] | 2.01e-13 | 0 / 0 |
| henon-orbits | [0.8|0.3|1|0.763085795] | [0.763085794519, 0.763085794519] | 2.02e-13 | 0 / 0 |
| henon-orbits | [0.82|0.3|1|0.757102917] | [0.757102917009, 0.757102917009] | 2.01e-13 | 0 / 0 |
| holmes-census | [2.7|0.2|4] | [49.000000000000, 49.000000000000] | 0.00e+0 | 0 / 0 |
| holmes-census | [2.775|0.2|4] | [49.000000000000, 49.000000000000] | 0.00e+0 | 0 / 0 |
| holmes-census | [2.85|0.2|4] | [49.000000000000, 49.000000000000] | 0.00e+0 | 0 / 0 |
| holmes-census | [2.55|0.2|4] | [33.000000000000, 33.000000000000] | 0.00e+0 | 0 / 0 |
| holmes-census | [2.625|0.2|4] | [33.000000000000, 33.000000000000] | 0.00e+0 | 0 / 0 |
| holmes-census | [1.95|0.2|4] | [17.000000000000, 17.000000000000] | 0.00e+0 | 0 / 0 |
| holmes-census | [2.025|0.2|4] | [17.000000000000, 17.000000000000] | 0.00e+0 | 0 / 0 |
| holmes-census | [2.1|0.2|4] | [17.000000000000, 17.000000000000] | 0.00e+0 | 0 / 0 |
| holmes-census | [2.175|0.2|4] | [17.000000000000, 17.000000000000] | 0.00e+0 | 0 / 0 |
| holmes-census | [2.25|0.2|4] | [17.000000000000, 17.000000000000] | 0.00e+0 | 0 / 0 |
| holmes-census | [2.325|0.2|4] | [17.000000000000, 17.000000000000] | 0.00e+0 | 0 / 0 |
| holmes-census | [2.4|0.2|4] | [17.000000000000, 17.000000000000] | 0.00e+0 | 0 / 0 |
| keller-audit | [Meng–Yang arXiv:2607.22198|5] | [128.000000000000, 128.000000000000] | 0.00e+0 | 0 / 0 |
| keller-audit | [Gallagher zenodo.21479195 d=2|3] | [1.000000000000, 1.000000000000] | 0.00e+0 | 0 / 0 |
| keller-audit | [Gallagher zenodo.21479195 d=3|3] | [1.000000000000, 1.000000000000] | 0.00e+0 | 0 / 0 |
| keller-audit | [Gallagher zenodo.21479195 d=4|3] | [1.000000000000, 1.000000000000] | 0.00e+0 | 0 / 0 |
| keller-audit | [Gallagher zenodo.21479195 d=5|3] | [1.000000000000, 1.000000000000] | 0.00e+0 | 0 / 0 |
| keller-audit | [Gallagher zenodo.21479195 distinct member (w-2w^3)|3] | [-1.000000000000, -1.000000000000] | 0.00e+0 | 0 / 0 |
| keller-audit | [Alpöge 2026-07-19|3] | [-2.000000000000, -2.000000000000] | 0.00e+0 | 0 / 0 |
| keller-audit | [Alpöge 2026-07-19|8] | [-2.000000000000, -2.000000000000] | 0.00e+0 | 0 / 0 |
| keller-audit | [tangent-sweep d=3 (new curve through the published mechanism)|3] | [-2.000000000000, -2.000000000000] | 0.00e+0 | 0 / 0 |
| keller-audit | [tangent-sweep d=4 (new curve through the published mechanism)|3] | [-2.000000000000, -2.000000000000] | 0.00e+0 | 0 / 0 |
| keller-audit | [tangent-sweep d=5 (new curve through the published mechanism)|3] | [-2.000000000000, -2.000000000000] | 0.00e+0 | 0 / 0 |
| keller-fibers | [fiber|alpoge] | [3.000000000000, 3.000000000000] | 0.00e+0 | 0 / 0 |
| keller-fibers | [fiber|alpoge-own-target] | [3.000000000000, 3.000000000000] | 0.00e+0 | 0 / 0 |
| keller-fibers | [fiber|sweep-d3] | [3.000000000000, 3.000000000000] | 0.00e+0 | 0 / 0 |
| keller-fibers | [fiber|gallagher-d2] | [3.000000000000, 3.000000000000] | 0.00e+0 | 0 / 0 |
| keller-fibers | [fiber|gallagher-d3] | [3.000000000000, 3.000000000000] | 0.00e+0 | 0 / 0 |
| keller-fibers | [fiber|sweep-d4] | [2.000000000000, 2.000000000000] | 0.00e+0 | 0 / 0 |
| keller-fibers | [fiber|gallagher-distinct] | [2.000000000000, 2.000000000000] | 0.00e+0 | 0 / 0 |
| newman-minmod | [0,2,4,5,8,9,10,11,12] | [1.362373178133, 1.362373178133] | 4.44e-16 | 720 / 720 |
| newman-minmod | [0,6,8,12,13,15,16,17] | [1.254933139031, 1.254933139031] | 6.66e-16 | 719 / 719 |
| newman-minmod | [0,3,8,12,13,14] | [1.013007454017, 1.013007454017] | 6.66e-16 | 694 / 694 |
| newman-minmod | [0,2,7,8,11,12] | [1.009230830699, 1.009230830699] | 4.44e-16 | 689 / 689 |
| ramanujan-audit | [rm-cat-22] | [21.310733573145, 21.310733573145] | 7.11e-15 | 719 / 719 |
| ramanujan-audit | [rm-z2-new8] | [18.190787907439, 18.190787907439] | 1.07e-14 | 719 / 719 |
| ramanujan-audit | [rm-cat-21] | [17.310693507032, 17.310693507032] | 7.11e-15 | 720 / 720 |
| ramanujan-audit | [rm-cat-18] | [13.966059953644, 13.966059953644] | 3.55e-15 | 715 / 715 |
| ramanujan-audit | [rm-z2-new7] | [13.382476033966, 13.382476033966] | 3.55e-15 | 719 / 719 |
| ramanujan-audit | [rm-zo-z5z3c] | [12.993891508980, 12.993891508980] | 3.55e-15 | 704 / 704 |
| ramanujan-audit | [rm-cat-19] | [12.735713658946, 12.735713658946] | 5.33e-15 | 719 / 719 |
| ramanujan-audit | [rm-z2-new4] | [12.382476033966, 12.382476033966] | 7.11e-15 | 720 / 720 |
| ramanujan-audit | [rm-cat-17] | [11.704592362598, 11.704592362598] | 5.33e-15 | 705 / 705 |
| ramanujan-audit | [rm-cat-20] | [11.126365014738, 11.126365014738] | 7.11e-15 | 720 / 720 |
| ramanujan-audit | [rm-z2-new9] | [9.627705192345, 9.627705192345] | 3.55e-15 | 719 / 719 |
| ramanujan-audit | [rm-z2-new5] | [8.557960170974, 8.557960170974] | 7.11e-15 | 720 / 720 |
| strassen-audit | [mm|alphatensor-f2-5x5x5] | [96.000000000000, 96.000000000000] | 0.00e+0 | 0 / 0 |
| strassen-audit | [mm|alphatensor-q-4x5x5] | [76.000000000000, 76.000000000000] | 0.00e+0 | 0 / 0 |
| strassen-audit | [mm|strassen-squared-4x4x4] | [49.000000000000, 49.000000000000] | 0.00e+0 | 0 / 0 |
| strassen-audit | [mm|alphatensor-q-4x4x4] | [49.000000000000, 49.000000000000] | 0.00e+0 | 0 / 0 |
| strassen-audit | [mm|alphaevolve-48-4x4x4] | [48.000000000000, 48.000000000000] | 0.00e+0 | 0 / 0 |
| strassen-audit | [mm|alphatensor-q-3x4x5] | [47.000000000000, 47.000000000000] | 0.00e+0 | 0 / 0 |
| strassen-audit | [mm|alphatensor-f2-4x4x4] | [47.000000000000, 47.000000000000] | 0.00e+0 | 0 / 0 |
| strassen-audit | [mm|alphatensor-q-3x3x3] | [23.000000000000, 23.000000000000] | 0.00e+0 | 0 / 0 |
| strassen-audit | [mm|strassen-1969] | [7.000000000000, 7.000000000000] | 0.00e+0 | 0 / 0 |
| strassen-audit | [mm|alphatensor-q-2x2x2] | [7.000000000000, 7.000000000000] | 0.00e+0 | 0 / 0 |
Each row is an exact enclosure, not a measurement. The last column is the engine asking whether the value has a small closed form: every candidate lying outside the enclosure is refuted exactly. The Ramanujan Machine matches truncated decimals and argues from collision probability; this decides.
54,629,173 candidate closed forms were tested against certified enclosures around 1e−15 wide, and 54,628,275 were refuted exactly — the value provably lies outside. Zero survivors means these objects have no small closed form of the shapes searched: a proved negative, not a failed search.
The count decomposes with nothing folded in: 54,629,173 tested = 54,628,275 refuted in double + 21 refuted exactly in BigInt + 877 with the form already on the OEIS record + 0 open + 0 surviving. Forms the 17-digit double screen could not separate were re-decided at the full published digit length in BigInt; forms OEIS already states are the record check working, not discoveries; the subtraction closes to zero and the engine refuses to write a ledger where it does not.
| bar | certified min|f| to beat | source |
|---|---|---|
| bar(6) | 1.000000000000000 | literature / lab |
| bar(7) | 1.065285891134415 | literature / lab |
| bar(8) | 1.101882938486186 | literature / lab |
| bar(9) | 1.311101302872326 | literature / lab |
| bar(10) | 1.378187726393090 | adopted here |
| bar(17) | 1.899892237678303 | adopted here |
| bar(18) | 1.899892237678303 | adopted here |
| bar(19) | 1.899892237678303 | adopted here |
| bar(20) | 2.018174563075912 | literature / lab |
Anchors from the literature and the source lab, plus objects this lab certified and then adopted. Both stored as witness sets and re-certified at load, so no value is transcribed. Frozen at load — the bar never moves under a running campaign, or a candidate's verdict would depend on when it was proposed.
The bars at n = 10 and n = 17 MOVED in August 2026 — bar(10) to 1.3782 past Boyd's 1986 witness, bar(17) to 1.8999 via the certified n = 13 box champion — because the exhaustive box30 sweeps behind certs/mu-table.json certified minima exceeding what the literature and the source lab held. A returning reader's remembered bar is stale because the mathematics improved, not because anything drifted; the rows tagged "adopted here" are exactly those promotions.
Staleness audit: clean — nothing certified sits above the envelope unadopted.
| battery | covers | this build |
|---|---|---|
| funnel machine | 14 items · 19 red controls | not run |
| detach | 11 checks | not run |
| interval · eqcert | falsifier-required certificates | not run |
| interval · arithmetic | 16 000 ops vs exact rationals | not run |
| interval · transcendental | sound exp/log/sin/cos | not run |
| trigmin certifier | 47 checks · 2 red controls | not run |
| newman box sweep | Goddard's 1992 box re-closed every run, cross-lab counts pinned; 100% kill audit · 7 red controls | not run |
| lambda sweep | Mercer's proved closed forms computed, never remembered; the wrong-endpoint bar refused by name and shown fatal · 4 red controls | not run |
| lambda4 campaign | Mercer's lambda(2) and lambda(3) proofs re-derived mechanically at every run — exception families DISCOVERED, thresholds DERIVED; the Section-5 generic case with its 14 exceptions matched against the hand-written list; the worklist measured at NINE families · 8 red controls | not run |
| envs (grader QA + the gyms) | the three environments over one idea — a certified enclosure is a canary factory: the fact corpus still matches the records it was read from (drift refuses), no minted canary lands inside its own enclosure, the certificate grader is sound on the whole suite while tolerance checking is broken by it, a bluffed tiling scores worse than abstaining, and the attacker ladder keeps rungs that cannot be broken · 5 red controls, including the accept-everything and reject-everything graders that must both score zero | not run |
| lambda56 campaign | the lambda(5)/(6) non-monotonicity campaign: lambda(4) generic re-derived as the calibration gate, both new generic cases certified (8 and 10 exception families DISCOVERED), the double-sum-core theorem re-proved with the Fejer-Riesz comb weight, extremizer walls counted, the record walked · 5 red controls | not run |
| sublevel (tao 179) | certified sublevel measures for root-constrained monic polynomials (Tao's #179 supremum conjecture on Erdős #1038): the 2*sqrt(2) witness enclosed, the box bound equal to the measure on thin boxes, and the degree-3 theorem plus degree-4 localization RE-PROVED at every run · 3 red controls | not run |
| mercer mu5 ladder | mu(5) <= 1 + pi/m certified m = 5..20; Mercer's Tables 5-7 reproduced, the source-lab m=6 record matched, every case point re-proved · 5 red controls | not run |
| census (henon + holmes) | closed-form calibration, two maps · 5 red controls | not run |
| keller audit + sweep | symbolic det over Q, generator calibrated on Alpöge · 4 red controls | not run |
| cf audit | all seven Ramanujan Machine sheets — 51 printed rows + the certified correction (e, pi, zeta(3), Catalan, pi^2, ln 2, mixed zeta orders) · 10 red controls | not run |
| entropy covering | certified h_top lower bounds; ln 2 calibration at the full horseshoe · 4 red controls | not run |
| strassen audit | fast-matmul tensor identities over Q and F2; Strassen 1969 calibrates · 3 red controls | not run |
| bigfloat layer | directed-rounding big-float intervals; pi/ln2/e to 50 literature digits · 5 red controls | not run |
| ivspecial (Γ + Bessel) | interval Γ (Spouge) and Bessel J_ν at fractional and NEGATIVE order — the spectral-geometry instrument: half-integer closed forms, Γ cross-derived on bigfloat against (2n)!/(4ⁿn!)·√π, J against exact-rational series brackets at dyadic points, the pinned frontier source re-hashed, fat-interval orders falsified for the band program · 6 red controls | not run |
| hotspots (ember chain) | the certified hot-spots theorem for the trapezoid outside every proven class: 8 stage records walked (inputs = upstream outputs, no hand copies), I₀ = 5/48 and C_tr and the cell partition re-decided LIVE in exact rationals/bigfloat, a collar kill re-proved live, witnesses proved to sit in tip disks · 7 red controls (mutated vertex, inflated defect, forged I₀, inflated flux sup vs the reflection layer, sign-flipped ladder, witness moved into the core, the bench's unsound tip-skip rule caught) | not run |
| erdos852 constants | certified c0 and C* enclosures; pi^2/8 product calibration · 5 red controls | not run |
| evtol energy | mission-energy feasibility verdicts cross-proved by 256-corner exact sweeps; dyadic closed-form calibration · 4 red controls | not run |
| forecast instrument | conformal coverage proved by exact rank-lemma enumeration; Winkler scores hand-computed in rationals; the ledger refuses backdating, tampering, premature and double scoring; the admission prune rule decided by exact binomial tail · 5 red controls | not run |
| covering (one module, four consumers) | the check that several theorems here quantified over a region actually stand on: do the pieces TILE it. Written once after being written twice — 1D ladders (endpoints must MATCH, relative comparison for ladders spanning decades) and 2D area accounting for adaptive box maps, with the honest limit that area proves covering almost everywhere and not everywhere. Zero-width pieces are reported and excluded, never allowed to bridge a hole · 12 red controls | not run |
| ember band (P3a, the family audit) | the hot-spots theorem on a positive-measure family c in [0.845,0.85]: an INDEPENDENT auditor re-derives the two covering ladders (17 chunks tiling the interval, 738 sigma-cells tiling [-1,0] in every stage) and every band-wide value from per-cell data, sharing no code with the producer. Eight RED CONTROLS each break the band a different realistic way — removed chunk, endpoint nudged 1e-5, one missing sigma-cell of 738, one margin at -1e-9, an escaped collar survivor, a dropped stage, tip C losing its certified sign, a ladder tiling the wrong interval · 8 red controls | not run |
| lemniscate (erdős 1038 infimum) | the #1038 infimum bracket walked and its fence enforced: five RED CONTROLS are genuine source mutations that must make a certificate refuse — the atom mass below the exact level, a displaced support endpoint, THE δ-MECHANISM (the family level defect on the wrong side, which provably kills small ε), the sliver constant below 1, and a forcing record claiming a cap its boxes do not tile · 5 red controls | not run |
| kissing ledger | D4 (24) and E8 (240) kissing witnesses re-proved from generated bytes every run — E8's 6,720 exact contacts equal the textbook 240·56/2; the AI-era dimension-11 ladder re-walked from pinned corpus bytes, one Station 604 re-certified LIVE in Z[sqrt2]; the mixed-sign sqrt2 comparator and decimal-literal exactness each guarded by a falsifier · 6 red controls | not run |
| fueleu penalty arithmetic | Regulation (EU) 2023/1805 intensity limits, Annex IV penalty and blend-flip thresholds in exact rationals — constants transcribed from pinned OJ bytes; the 1e-9 boundary forgery flips the verdict · 4 red controls | not run |
| glide band | certified engine-out reach from interval inputs; geodesy calibrated on the meridian degree and JFK-LAX, 4000-draw containment, and the point-estimate method itself run as a red · 4 red controls | not run |
| design system + charts | palette validated against the dataviz checks in BOTH modes and under three CVD simulations; the token block, the figure kit and the escaped-tag scanner · 6 red controls | not run |
| wiring | the registries nobody was checking: every report builder reachable from make reports, the two battery lists in agreement, and no built page declaring a font outside the token block · 4 red controls | not run |
| stale claims (record -> page) | every published page/row pair re-read against the record behind it, so a number that moved in a record and not on the page refuses the build. It ran in make test and NOT here until 2026-09-04, which made this list’s own green count a claim over an incomplete set — the defect check-wiring exists to catch, sitting just outside its battery-name pattern. | not run |
| measure (the layout ruler) | the geometry of all 66 built pages driven in headless Chrome at 1440 and 390: how many left edges the content sits on, how far the page scrolls sideways, what leaves the viewport unreachably, and what is clipped inside its own box — compared against design/measure-baseline.json, which only ratchets down · 7 red controls (three planted layouts, a scroll table that must NOT read as an escape, a clean page that must measure one spine, and the ratchet itself attacked three ways) | not run |
| skyaudit app | segmentation and mission calibration for the pinned ADS-B day | not run |
| bilinear certifier | bilinear identities over Q and F2 | not run |
| slp additive circuits | straight-line programs, additive cost | not run |
| mfg lab (box certifier) | the box certifier for the MFG lab | not run |
| mfg2p lab (two populations) | two-population equilibria | not run |
| mfg-cap census (EXACTLY-n) | Krawczyk exhaustion census of the even Galerkin mfg-cap system: EXACTLY 3 solutions at c=-12 re-proved live at N=2 every run, records walked for N=2..5, honest box-bounded truncation scope asserted · 3 red controls (midpoint split refuses at the constant solution’s exact coordinates, starved budget, corrupted kernel) | not run |
| erdos290 lean battery | closed forms equal enumeration exactly for l <= 12; the broken-EGF red must fire | not run |
| critcount (certified peak counts) | critical-point counts of even cosine series over an enclosure ball — count derived from CERTIFIED region signs only, outward coefficient products, ball Lipschitz folded into the cell pad; closed-form two/three-harmonic calibrations and the terra record walk · 4 red controls (mutated boundary, zeroed ball pad, degenerate curvature, two critical points in one region — each fires) | not run |
| engine + families | red controls on screen and certifier | not run |
| oracle claim library | certify() for AI math search: Strassen calibrates, the characteristic-2 pair reproduced, the sub-float forgery refuted with its exact mechanism; red controls also run at import — a broken grader refuses to exist · 6 red controls | not run |
| keller · standalone re-verifier | the detached certificate re-audited from scratch — stdlib fractions, no code from this repo; red control must fire | not run |
| strassen · standalone re-verifier | every matmul identity re-derived in stdlib Python ints; pins re-hashed; red control must fire | not run |
| erdos852 · standalone re-verifier | the C* refutation re-proved in exact stdlib ints (no tail, no rounding); the c0 window re-decided at 130 digits; 4 red controls must fire | not run |
| mfgcap · terra re-certification | the congestion-MFG peak-splitting enclosures (T1 two peaks, T6 three peaks) re-certified inside cert-machine: validate_g pinned BIT-FOR-BIT to the frozen published verifier on its embedded instance, the stdlib Gauss-Jordan approximate inverse certifying at the identical radius, the record walk, and the A2/A3 data terms — the only new lines — attacked by their own mutations · 5 red controls | not run |
| facelaw · face-dimension theorem | k = |shared| - cons + z decided against the exact Q null space — 600 live networks every run, the 572-failure origin ensemble replayed from its seed with every failing instance enumerated in the record, the exit-free-cycle family, and the origin instance whose z = 0 made the shortcut look like a law · 3 red controls | not run |
| attnflow · attention exact-Q | the decidable attention flow: the reduced flow's DOUBLE zero at c* = -1/beta (multiplicity decided by exact division), denominator SOS, no crossing at any beta > 1 — every pitchfork claim refuted exactly; cross-weights identically zero with the honest p = 1 boundary; the consensus spectrum's beta/p-freeness decided by exact dual-number expansion at n = 3..6; the phantom-bifurcation taxonomy with the locator transient re-demonstrated live · 3 red controls (one of which caught this battery's own detector being decorative) | not run |
| sos · global bound | stdlib fractions only | not run |
| sos · lyapunov | stdlib fractions only | not run |
| sos · re-verify AI result | stdlib fractions only | not run |
| llm harness — the eval's dry-run gate | a FAKE proposer gates the pipeline, not an LLM result; live model campaigns are separate, in the append-only certs/matmul-eval-ledger.jsonl and on reports/matmul-eval.html · aborts if a red control certifies | not run |
Lifted from the source lab: 130 files, 7 patched on the way in, each patch declared. Drift now: 130 unchanged · 0 source moved · 0 local edited · 0 source gone. The source lab (sin-mfg) is read-only, permanently — read anything, never write, and report an error there rather than repair it.
| certificate | what it holds | re-verify |
|---|---|---|
| erdos852-certificate.json | Both Erdős #852 constants as exact data: the c0 window re-decidable at 130 digits, the C∗ refutation as strict integer inequalities with no tail bound. | verify_erdos852.py |
| erdos852-h-records.json | Every record run of pairwise-distinct consecutive prime gaps the scan has closed — index, opening prime and length. The head reproduces OEIS A078515/A079889 term for term; the tail passes them. Each record beyond the published terms is re-proved at build by an independent Miller–Rabin verifier. | battery-gated |
| ai-claims-summary.json | The six-lane AI-claim audit as the build recorded it: every lane’s verdict, scope, check count and mutation-control count, written by the report builder from a live run of all six verifiers. Not a certificate — a record of what the verifiers said, so the shelf card and the page cannot quote different numbers. | battery-gated |
| keller-certificate.json | The Jacobian/Hessian counterexample corpus — every polynomial as explicit exact rational monomials; determinants and collisions re-derivable from the file alone. | verify_keller.py |
| erdos1038-inf.json | Erdős #1038, the INFIMUM side: the certified bracket 1.828 ≤ inf ≤ 1.8344304971959906 with both ends unconditional, Tao’s model Problem 4.1 answered affirmatively for every ε ∈ (0, 0.1] (624,275 chunks + a sliver lemma), and the thread’s three posted dual measures certified. Rebuilt live at every report build; the record carries the claim fence naming all three claimed proofs. | battery-gated |
| erdos1038-forcing-1.828.json | The forcing certificate behind inf ≥ 1.828 — every a₀-box with its comb, per-b-box frozen LP weights and certified margins. Re-checked by instruments/lemniscate/verify-forcing.js, which shares no code with the certifier and also hunts counterexamples in doubles. | battery-gated |
| certificate | what it holds | re-verify |
|---|---|---|
| ember-band.json | THE EMBER BAND: the hot-spots theorem extended from one trapezoid to every c in [0.845, 0.85] — 17 chunks tiling the interval with shared endpoints, 738 σ-cells inside them, μ₁ ≥ 11.85157 and μ₂ ≥ 13.90774 uniformly. An AUDIT record, not a re-derivation: the chain ran on the bench, and instruments/emberband/verify-band.js re-derives both covering ladders and every band-wide value from per-cell data, sharing no code with the producer. | battery-gated |
| kissing-ledger.json | The kissing ledger: every public record configuration for K(11) — AlphaEvolve’s 593, the EinsteinArena rung winner’s 594, the Station’s three exact 604s, the classical 582 shell and the D₁₂ lift — re-decided in exact Z[√2] BigInt arithmetic from sha-pinned claimant bytes, shared-nothing with every producer’s own verifier; contact counts exact, the byteless EinsteinArena 604 measured as NEEDS DATA. Rebuilt by tools/run-kissing-ledger.js at every report build. | battery-gated |
| strassen-certificate.json | 10 fast matrix-multiplication algorithms as exact tensor identities over Q and F2 — including AlphaTensor’s rank-47, decided both ways. | verify_strassen.py |
| bilinear-certificate.json | 9 bilinear algorithms for POLYNOMIAL multiplication over F2 — full, truncated and cyclic products — each found by the generation front’s free flip-graph walk and decided by instruments/bilinear, which rebuilds the target tensor from its name rather than trusting the scheme handed to it. Every entry stores its scheme in full, so any reader can re-decide it; the published bounds each one is measured against live in corpus/bilinear-bounds.json and are not results of this repository. | battery-gated the report |
| lambda4-audit.json | The adversarial audit of the lambda(4) proof: an independent clause walk sharing no code with the engine — inner products by direct trigonometric summation, family and subfamily membership by plain integer arithmetic, thresholds read from the campaign record, finite-clause sets re-certified fresh by the calibrated instrument. Every gcd-reduced 4-set in the box must be reached by an explicit clause of the proof (generic, family dot, closure, finite, delegated, or the definitional witness); a set with no clause is a hole and aborts. Also carries the full theorem sweep of the box: zero refuters. | battery-gated |
| claims-ledger.json | The claims desk ledger: every externally published mathematical claim this machine has decided, one row per claim, each DERIVED from the record that decided it — a claim with no record gets no row. Origin is tracked separately from verdict, because everything decided so far was self-initiated and the submitted count stays published even while it is zero. | battery-gated |
| grader-pilot.json | The environment run against real models, and the reference policies run on the SAME seeds so the columns are the same tasks rather than a comparable sample: 360 calls at $1.92 of a $4.00 cap, worst case reserved before every one. Every row is scored by the shipped Python package — tools/run-grader-pilot.js picks tasks and pays for calls and decides nothing. Truncations and model refusals are recorded and excluded from the rates, because a harness artifact is not a model outcome. | battery-gated |
| gym-record.json | The shipped environment, measured by RUNNING it: the difficulty dial (band size against the tolerance multiple, with the point below which no representable double fits), the corpus it draws on, the mix of a 2,000-task sample, and the two standing answers with how much of the ladder each one loses. Produced by tools/run-gym-record.js, which executes the Python package the way a buyer would rather than describing it. | battery-gated |
| envs-record.json | The environments record: the fact corpus (every entry read out of a record in certs/ and sha256-pinned to it, because a canary asserts “provably wrong” and that may not rest on a re-typed decimal), the grader QA measurement across four reference grader shapes and three tolerances, the uniformity gym’s solver table, and the attacker ladder with the rungs that cannot be broken. Written by running the environments offline — no model is called and the harness refuses the network unless explicitly switched on. | battery-gated |
| envs-ledger.jsonl | The append-only environments ledger: one row per (environment, rung, model, k) cell with pass rate, Wilson interval, wrong/refused split and the forgery-gate result for the run. Rows so far are reference and stub solvers only; a row produced by a real model is a decision, not a default. | battery-gated |
| lambda5-audit.json | The independent audit of the lambda(5) theorem: a second walk sharing no code with the symbolic engine — inner products by direct trigonometric summation, condition membership by plain integer arithmetic on the record’s own condition vectors, and every set the float screen prunes sampled and re-certified exactly, so the screen is audited rather than trusted. It decides three things and says so: the THEOREM (every gcd-reduced 5-set in the box dips strictly below L(1,2,4,5,6), except the extremizer, which attains it — a set that did not would be a refuter and aborts), the FIRST LEVEL of the reduction (the engine’s symbolic model says an inner product is its base plus the deltas of the active conditions, and direct summation must agree at every set in the box; every set the generic argument does not close must satisfy one of the eight recorded family conditions), and the OBSTRUCTION by its mechanism rather than by search (on the double-sum core the cosines cancel identically on S(e, pi), so no nonnegative weight can start, and the comb closes every core point where no positive condition is active). It does NOT walk the interior of the eight closure trees; the lambda(4) audit does that for lambda(4). | battery-gated |
| lambda4-campaign.json | The lambda(4) campaign record. Phase 0: Mercer's Section-5 strategy (INTEGERS 19 (2019) #A4) executed mechanically — lambda(2) and the whole lambda(3) proof re-derived with exception families discovered and thresholds derived, not transcribed; the lambda(4) generic case with its 14 exception conditions discovered and matched against the paper's hand-written list; the measured reduction: five of the fourteen carry strictly negative delta, so NINE families remain. Phase 1, in progress: each family closed so far carries its full derivation here — second-level dot theorem, subfamily cones, derived thresholds, decided finite parts. d = 2c is CLOSED. Re-derived symbolically at every build by the lambda4 battery; finite parts pinned here and sampled at every run. | battery-gated |
| sublevel-tao179.json | The Tao #179 sublevel campaign (Erdős #1038, supremum side): rational-weight discrete measures on [-1,1] are monic root-constrained polynomials via |q| < 1, and this record holds certified sublevel measures — the 2√2 witness, grid champions including the interior cubic champion near 2.7542 and the quintic transition peak near 2.8011 — plus per-degree branch-and-bound THEOREMS: odd degrees strictly below 2.82 < 2√2, even degrees localized to [2√2, 2.82845] with (x²−1)^{N/2} attaining the left end. Every measure an outward enclosure from BigInt Sturm isolation; the box bound calibrated to equal the measure on thin boxes. | battery-gated |
| mercer-mu5.json | The mu(5) ladder, mu(5) ≤ 1 + π/m rung by rung to m = 20 — every exceptional tuple closed by one exact rational evaluation. | battery-gated |
| mu-table.json | The Newman min-modulus table: every set in the named boxes exhausted, champions certified, orbits classified, conservation per row. | battery-gated |
| mu-table-40.json | The wider-box extension of the mu table — billions of sets exhausted, the narrow-box crowding artifacts corrected. | battery-gated |
| lambda-table.json | The lambda table: the source lab’s rows reproduced exactly, plus rows that lab’s record does not hold and no table this lab has read holds — a claim about the certificates, not about priority. | battery-gated |
| census-high-periods.json | The Hénon high-period census, p = 13..16: every period point of the classical map found and counted exactly, the plane exhausted box by box, each row re-checked with zero unmatched — the proving ground the interval instruments were calibrated on. | battery-gated |
| entropy-henon.json | The certified entropy lower bound for the classical Hénon map: h-sets, covering relations, and the exact spectral argument. | battery-gated |
| erdos290-tail-ext.json | The Erdős #290 sweep extension: degrees closed beyond the cited page, the constant’s enclosure tightened degree by degree. | battery-gated |
| mfg2p-regime-map.json | The TWO-population regime map: every cell of the coupling plane with its verdict and its exact witness — the symmetric cross-coupling s against the attack-defense asymmetry d, two disjoint enclosures where uniqueness provably fails, the Lasry-Lions monotone strip where it does not, and the refusal reason everywhere else. There is no single-file certifier for this map: it is decided by labs/mfg2p/box2p.js and re-gated by that lab’s battery at every build of its report. | battery-gated the report |
| mfg-regime-map.json | The mean-field-game regime map: every cell of the coupling–potential plane with its verdict and its exact witness — two disjoint enclosures where uniqueness provably fails, the monotone enclosure where it does not, and the refusal reason everywhere else. | mfg-certify.js |
| matmul-eval-ledger.jsonl | The matmul eval’s append-only ledger — every campaign row, every verdict, every tag; the leaderboard is built from this file. | battery-gated |
| chowla-records.json | Certified Chowla merits c(A) = -min_x sum cos(ax) / sqrt|A|, one row per set size n, each an exact UPPER bound proved by instruments/trigmin for the set stored beside it. THIS FILE MAKES NO CLAIM ABOUT CHOWLA’S COSINE PROBLEM. Chowla’s question is asymptotic — whether c can be driven to 0 as n grows — and a low c at one n is a fact about that n. The measured trend here RISES with n (0.6558 at n=10 to 0.8205 at n=30), which is consistent with Chowla’s conjecture that the order is sharp, i.e. evidence against the direction, not for it. Explicit sets with small c are occupied literature (Mercer, INTEGERS 2019; Bedert, arXiv:2509.05260); these rows beat only the classical families recomputed here beside them. | battery-gated |
| matmul-eval-corrections.json | Corrections to rows already written in the matmul eval ledger. The ledger is append-only and is never rewritten, so a row that turns out to be MISLABELLED is corrected here and the correction is applied when the report displays it — currently one: 90 rows tagged v4-effort-low ran at the API’s DEFAULT effort, because the harness dropped its --effort argument in campaign mode. A correction naming a tag no row carries, or claiming a row count the ledger does not hold, refuses the report build. | battery-gated |
| matmul-loop-ledger.jsonl | The verifier-in-the-loop ledger — every trajectory round with its verdict and the exact feedback sent; the loop report is built from this file. | battery-gated |
| skyaudit-forecast-ledger.jsonl | The prediction ledger — interval FORECASTS committed before their target day (sha-pinned, append-only) and scored after in exact rationals; coverage claims are conformal counting theorems, never model faith. Wrong forecasts stay forever. | battery-gated |
| forecast-gym-ledger.jsonl | The Forecast Gym’s append-only ledger — every proposer’s forecast sha-committed before its outcome exists, every score an exact Winkler rational; the gym report and its admission board are built from this file. | battery-gated |
| lambda56-campaign.json | The lambda(5)/lambda(6) campaign record (Chowla’s cosine dip, the non-monotonicity program): lambda(5) = −L(1,2,4,5,6) closed in full — the generic case certified, all eight exception families closed with derived thresholds, 1725 finite sets decided, the extremizer walled — and lambda(6) in progress with nine of its ten families closed in this record. The double-sum-core obstruction theorem (no classical weight works on b+c = a+d = e) and the Fejér–Riesz comb weight that beats it are re-proved by the battery at every run. | battery-gated |
| terra-sigmastar.json | The exact crossover of the MFG splitting program: σ* = 1/(8π²), discount-free, DECIDED IN EXACT RATIONALS — the crossover polynomial factors, the γ-coefficient is identically zero (checked k = 2..12), the band-pass identity and both harmonic windows are exact, π enters only as a Machin bracket of width 1.3e-44. | battery-gated the report |
| terra-bracket-table.json | The bracket table under the splitting theorems: seven certified rows straddling both predicted thresholds — negatives below the amplitude threshold and past the crossover, replications, and the threshold pin r_c ∈ [0.13, 0.14] with the exact-rational prediction landing inside. Honest counting lives here: two theorems plus a table, never eight. | battery-gated the report |
| mfg-cap-multiplicity.json | Certified multiplicity for the ergodic quadratic MFG past its pitchfork: at each of six couplings, at least THREE distinct exact solutions enclosed in pairwise disjoint uniqueness balls with certified positive density — exactly where Lasry–Lions monotonicity is silent. The c = −9.5 monotone-regime boundary, where the branch collapses and no claim is made, is recorded too. | battery-gated the report |
| attnflow-theorems.json | The attention-wing theorems: a rational-kernel token flow chosen so equilibrium and stability are DECIDABLE in exact ℚ — the consensus spectrum proved β- and p-free by exact dual-number expansion, the two-cluster cross-weights identically zero with the honest p = 1 boundary, the reduced flow’s double zero decided by exact division (every pitchfork claim refuted), and the phantom-bifurcation taxonomy with its live artifact. | battery-gated the report |
| facelaw-theorem.json | The face-dimension law k = |shared| − cons + z, decided against the exact ℚ null space on two seeded 4,000-network ensembles; every instance where the natural shortcut fails (precisely the z > 0 cases) is ENUMERATED here so any reader can re-run any one. | battery-gated the report |
| terra-recert-t1.json | The T1 enclosure of the congestion-MFG splitting program: a radii-polynomial certificate around the stored candidate — exact instance parameters, certified ℓ¹_ν ball radius, contraction bound Z₁, closure margin, and density floor — re-certified here against the frozen published verifier extended with the data terms; nine falsifiers fire per instance at every battery run. | battery-gated the report |
| terra-peakcount-t1.json | The certified peak count for T1: the exact number of strict maxima of EVERY density in the T1 enclosure ball, derived from CERTIFIED region signs only (instruments/critcount) — outward coefficient products, ball Lipschitz folded into the cell pad, never a float sign. | battery-gated the report |
| terra-recert-t2.json | The T2 enclosure of the congestion-MFG splitting program: a radii-polynomial certificate around the stored candidate — exact instance parameters, certified ℓ¹_ν ball radius, contraction bound Z₁, closure margin, and density floor — re-certified here against the frozen published verifier extended with the data terms; nine falsifiers fire per instance at every battery run. | battery-gated the report |
| terra-peakcount-t2.json | The certified peak count for T2: the exact number of strict maxima of EVERY density in the T2 enclosure ball, derived from CERTIFIED region signs only (instruments/critcount) — outward coefficient products, ball Lipschitz folded into the cell pad, never a float sign. | battery-gated the report |
| terra-recert-t3.json | The T3 enclosure of the congestion-MFG splitting program: a radii-polynomial certificate around the stored candidate — exact instance parameters, certified ℓ¹_ν ball radius, contraction bound Z₁, closure margin, and density floor — re-certified here against the frozen published verifier extended with the data terms; nine falsifiers fire per instance at every battery run. | battery-gated the report |
| terra-peakcount-t3.json | The certified peak count for T3: the exact number of strict maxima of EVERY density in the T3 enclosure ball, derived from CERTIFIED region signs only (instruments/critcount) — outward coefficient products, ball Lipschitz folded into the cell pad, never a float sign. | battery-gated the report |
| terra-recert-t4.json | The T4 enclosure of the congestion-MFG splitting program: a radii-polynomial certificate around the stored candidate — exact instance parameters, certified ℓ¹_ν ball radius, contraction bound Z₁, closure margin, and density floor — re-certified here against the frozen published verifier extended with the data terms; nine falsifiers fire per instance at every battery run. | battery-gated the report |
| terra-peakcount-t4.json | The certified peak count for T4: the exact number of strict maxima of EVERY density in the T4 enclosure ball, derived from CERTIFIED region signs only (instruments/critcount) — outward coefficient products, ball Lipschitz folded into the cell pad, never a float sign. | battery-gated the report |
| terra-recert-t5.json | The T5 enclosure of the congestion-MFG splitting program: a radii-polynomial certificate around the stored candidate — exact instance parameters, certified ℓ¹_ν ball radius, contraction bound Z₁, closure margin, and density floor — re-certified here against the frozen published verifier extended with the data terms; nine falsifiers fire per instance at every battery run. | battery-gated the report |
| terra-peakcount-t5.json | The certified peak count for T5: the exact number of strict maxima of EVERY density in the T5 enclosure ball, derived from CERTIFIED region signs only (instruments/critcount) — outward coefficient products, ball Lipschitz folded into the cell pad, never a float sign. | battery-gated the report |
| terra-recert-t6.json | The T6 enclosure of the congestion-MFG splitting program: a radii-polynomial certificate around the stored candidate — exact instance parameters, certified ℓ¹_ν ball radius, contraction bound Z₁, closure margin, and density floor — re-certified here against the frozen published verifier extended with the data terms; nine falsifiers fire per instance at every battery run. | battery-gated the report |
| terra-peakcount-t6.json | The certified peak count for T6: the exact number of strict maxima of EVERY density in the T6 enclosure ball, derived from CERTIFIED region signs only (instruments/critcount) — outward coefficient products, ball Lipschitz folded into the cell pad, never a float sign. | battery-gated the report |
| terra-recert-t7.json | The T7 enclosure of the congestion-MFG splitting program: a radii-polynomial certificate around the stored candidate — exact instance parameters, certified ℓ¹_ν ball radius, contraction bound Z₁, closure margin, and density floor — re-certified here against the frozen published verifier extended with the data terms; nine falsifiers fire per instance at every battery run. | battery-gated the report |
| terra-peakcount-t7.json | The certified peak count for T7: the exact number of strict maxima of EVERY density in the T7 enclosure ball, derived from CERTIFIED region signs only (instruments/critcount) — outward coefficient products, ball Lipschitz folded into the cell pad, never a float sign. | battery-gated the report |
| terra-recert-t8.json | The T8 enclosure of the congestion-MFG splitting program: a radii-polynomial certificate around the stored candidate — exact instance parameters, certified ℓ¹_ν ball radius, contraction bound Z₁, closure margin, and density floor — re-certified here against the frozen published verifier extended with the data terms; nine falsifiers fire per instance at every battery run. | battery-gated the report |
| terra-peakcount-t8.json | The certified peak count for T8: the exact number of strict maxima of EVERY density in the T8 enclosure ball, derived from CERTIFIED region signs only (instruments/critcount) — outward coefficient products, ball Lipschitz folded into the cell pad, never a float sign. | battery-gated the report |
| mfg-cap-census-N2-c-12.json | The Krawczyk exhaustion census at Galerkin level N = 2: every subbox of the printed box eliminated by interval residual or Krawczyk exclusion, each survivor isolated by K(X) ⊂ int(X) — EXACTLY three solutions of the truncated even system at c = −12, matched one-to-one to the candidates. Box-bounded, truncation-level; the PDE-level count is the stated open problem. | battery-gated the report |
| mfg-cap-census-N3-c-12.json | The Krawczyk exhaustion census at Galerkin level N = 3: every subbox of the printed box eliminated by interval residual or Krawczyk exclusion, each survivor isolated by K(X) ⊂ int(X) — EXACTLY three solutions of the truncated even system at c = −12, matched one-to-one to the candidates. Box-bounded, truncation-level; the PDE-level count is the stated open problem. | battery-gated the report |
| mfg-cap-census-N4-c-12.json | The Krawczyk exhaustion census at Galerkin level N = 4: every subbox of the printed box eliminated by interval residual or Krawczyk exclusion, each survivor isolated by K(X) ⊂ int(X) — EXACTLY three solutions of the truncated even system at c = −12, matched one-to-one to the candidates. Box-bounded, truncation-level; the PDE-level count is the stated open problem. | battery-gated the report |
| mfg-cap-census-N5-c-12.json | The Krawczyk exhaustion census at Galerkin level N = 5: every subbox of the printed box eliminated by interval residual or Krawczyk exclusion, each survivor isolated by K(X) ⊂ int(X) — EXACTLY three solutions of the truncated even system at c = −12, matched one-to-one to the candidates. Box-bounded, truncation-level; the PDE-level count is the stated open problem. | battery-gated the report |
| ember-spectrum.json | Stage 1 of the hot-spots chain: two-sided spectrum localization for the trapezoid — interval Galerkin Rayleigh–Ritz uppers (floats only pick the subspace), exact-rational Crouzeix–Raviart assembly with interval LDLᵀ inertia counts and Liu’s framework for the lowers; the certified gap makes μ₁ SIMPLE. The rectangle regression encloses π² at every run. | battery-gated the report |
| ember-defect.json | Stage 2: the frozen Helmholtz trial’s boundary defect — order-2 interval Taylor jets along every edge with Bessel-ODE closure, a certified midpoint-Taylor cell rule, value AND derivative bridges against an independent float evaluation, and the exact trial coefficients frozen for every downstream stage. | battery-gated the report |
| ember-eigenpair.json | Stage 3: the eigenpair certificate — the boundary-residual identity with the rational star-shaped trace constant and the CR localization tightens μ₁ by a factor of ~105 and encloses the eigenfunction in L²; also the H¹ error and the δλ bound the pointwise machinery consumes. | battery-gated the report |
| ember-pointwise.json | Stage 4: the solid-mean pointwise machinery — kernel norm I₀ = 5/48 DERIVED in exact rationals, witness balls decided inside Ω in exact rationals, and every CORE cell of the 1/100 grid (corner-min depth ≥ 3/40, exact by concavity) killed on both sides with zero survivors. | battery-gated the report |
| ember-collar.json | Stage 5: the collar sweep — every sub-core cell killed by the value argument with REFLECTED pointwise bounds across its nearest open edge (the single layer bounded by the certified per-edge flux sups), kill-or-refine to 1/800, ZERO residual cells. | battery-gated the report |
| ember-corner.json | Stage 6: the four corner-tip certificates — Bessel–Fourier coefficients certified by annulus L² extraction AND re-extracted at a second annulus (the enclosures must intersect — a condition of entry), value kills at B/C/D, radial monotonicity at A, and the ladder-identity wedge bound at C where the boundary minimum lives. | battery-gated the report |
| ember-cross.json | The independent cross-derivations: I₀ = 5/48 in exact rationals, the trace constant re-derived on directed dyadic big-floats from the exact-rational star geometry, μ₁ bounded above on an independent conforming P1 basis, and the two-annulus corner condition re-asserted. | battery-gated the report |
| ember-theorem.json | The assembled hot-spots theorem: cross-record chain consistency (every stage’s inputs equal the upstream outputs), the interior partition RE-DECIDED IN EXACT RATIONALS, every sweep and tip verdict re-checked, the honest framing with its fence list, and the sha256 of every input record. | battery-gated the report |
Three of these carry a detached verifier — standard library Python, nothing to install, zero code shared with the engine, and each one must refute a deliberately forged value before it will exit green. The rest are gated by a battery instead: re-derived at every build, with planted forgeries that must fire. A record with neither is not on this list, because the build refuses when certs/ holds a file this table cannot describe.
Everything above decides. Nothing above shows — and a chart is an assertion too. Charts and graders both assert things, and neither distinguishes what the data forces from what the renderer or the tolerance chose. The same sentence covers both halves of this machine, which is why the second half is not a side project.
Nine instruments carry it. Two of them draw a certificate from the shelf in §8 at its own resolution — the covering that forced an Erdős lower bound, shaded so the bright seams are where the argument nearly ran out, and the weights the linear programme chose inside those boxes. Five decide their headline number in exact integer or rational arithmetic. One is floats throughout and leads with that.
The grammar they are drawn in is four standings on a lattice — refused < chosen < computed < decided — combining by minimum, so a mark’s standing is the weakest standing on its path to the pixel. Three consequences follow and none is a judgement call: the arithmetic demotes and never the author, so a float step takes decided to computed automatically; refused is absorbing; and an argmax over a non-unique optimum yields chosen. Stroke carries it, never colour, so it survives greyscale — and where a figure’s marks all share a standing the channel is free, which is a rule found by porting four pages rather than by reasoning.
This is not a confidence encoding and must not be read as one: a value can be known to fourteen decimals and still be undecided, and certified with a wide bracket. It is narrowed against uncertainty visualization (Padilla, Kay & Hullman), provenance visualization (Ragan et al.), imputed-value encoding (Song & Szafir), verifiable visualization (Kirby & Silva) and the lineup protocol (Buja et al. 2009) — all scouted, all recorded in corpus/targets.json, and the last of them is a method one of those pages reinvented before it knew the name. There is no user study, which is stated up front rather than left for a reviewer to find.