---
slug: the-number-nobody-is-holding
passed: 2026-10-18
by: the assay line (research/assay-line/claude-1018b/), Claude fixer (claude-opus-5-5) directed by claude-assaying-boyle-1014; source record by a Claude agent (claude-sonnet-5-5), checked by a Claude agent (claude-opus-5-5)
checked: 133
confirmed: 108
wrong: 11              # 1 WRONG + 10 MINOR
unverifiable: 0
observed: 14
stale: 0
---
# Claims pass: The Number Nobody Is Holding

This pass was the Assay Office's assay line, batch claude-1018b, re-reading the published layer against its sources, source first: one Claude agent (claude-sonnet-5-5) built the source record (research/assay-line/claude-1018b/the-number-nobody-is-holding/N1-sources.json), a second (claude-opus-5-5) checked every entry against the opened sources and the repository's own artefacts, recreating the census counts, the specification curve, the curveball bisection and the verifier's reach (N2-claude.md), and a third (claude-opus-5-5) decided each finding (triage.json) and made the one fix. The checker raised two WRONG and nine MINOR lines. One WRONG was upheld and fixed: the page said both unsettled figures are named on their own pages, and only the keyboard page names its figure. The other, that the page's verifier re-derives every figure the page prints, was re-run and holds as an overclaim (sentence figures are not held), but it is an overclaim about the page's own checking, which the office sets at MINOR; with the other nine MINOR lines it is listed, not edited, at tier B. The checker's reader-side note (the page names its verifier without a link to the published copy under /checks/) is not a claim and was not acted on.

## Claims
- CONFIRMED | This page published 742 orphans among 12,886 claim-shaped figures, a 5.76 percent rate. | research/unasserted-figures/README.md | [h-rate] "That cell is 742/12,886, or 5.758187179885146%." Recreated: the verifier's S1 recount of census.json passed (230/230).
- CONFIRMED | This page published 742 orphans among 12,886 claim-shaped figures, a 5.76 percent rate. | public/strata/the-number-nobody-is-holding/data.json | [h-counts] figures: figuresTotal 12886, orphan 742; the verifier recounts both from research/unasserted-figures/census.json and passes.
- CONFIRMED | the 428 specifications with at least 75 figures span 3.0 to 12.4 percent. | research/unasserted-figures/README.md | [h-range] Recreated with public/_kit/multiverse.js over data.json multiverseRows: nOk 428, min 2.9702970297, max 12.4087591241.
- CONFIRMED | The published point sits at the 25th percentile; 74.5 percent report more exposure. | research/unasserted-figures/README.md | [h-percentile] rank 109, 25.350467289719625th percentile, 319/428 above; the verifier asserts all three from census.json and passes.
- CONFIRMED | The fact-check found three distinct wrong claims, enough to prove the problem is real and too few to distinguish the sampled strata. | public/strata/the-number-nobody-is-holding/data.json | [h-three-wrong] auditDistinctWrong 3; Wilson intervals 1.10-9.06% (orphan 3/93) and 0.37-10.90% (corroborated 1/48) overlap, asserted by the verifier.
- OBSERVED | On 2 August 2026 somebody here was fact-checking a card and noticed that | memory/log.d/2026-08-02T1900Z-relabelled-blind.md | [fire-accident-date] "The adversarial fact-check of the 27 new card texts (26 PASS, 1 FIX) turned up two live errors", the first being what-is-fire's ~15,000 K.
- CONFIRMED | said nitrogen does not cross one percent thermal ionisation until about 15,000 K, while the same page's own printed Saha table showed 1.75 percent at 10,000 K. | public/strata/what-is-fire/index.html | [fire-said-15000] "Corrected 2026-08-02: this line read "~15,000 K"..."; the verifier's printed table row is "10000 9.81e-1 1.75e-2".
- OBSERVED | The real crossing is near 9,450 K. | recreated: node research/what-is-fire/verify.mjs (scratch copy at 9ddc1e9045) | [fire-9450-page] prints "nitrogen crosses 1% thermal ionisation at ~9,450 K (NOT 15,000 K) crossing = 9449 K", 21 passed.
- MINOR | The real crossing is near 9,450 K. | https://physics.nist.gov/cgi-bin/ASD/ie.pl?spectra=N+I&units=1&format=0&order=0&at_num_out=on&sp_name_out=on&ion_charge_out=on&el_name_out=on&seq_out=on&shells_out=on&conf_out=on&level_out=on&ion_conf_out=on&e_out=0&unc_out=on&biblio=on&submit=Retrieve+Data | [fire-9450-nist] NIST gives only the input (N I 14.53413 eV). 9,450 K is the crossing of what-is-fire's simplified model (research/what-is-fire/verify.mjs line 172: sahaFraction(m, 14.534, numberDensity(m), 1), i.e. g_i/g_0 = 1). With the ground-term weights (N II 3P, g = 9; N I 4S, g = 4) the same 1 atm Saha bisection crosses near 9,090 K. "real" claims more than the model; "the page's own Saha model crosses near 9,450 K" is what is shown. Tier B: listed, not edited.
- OBSERVED | The layer's check had been green the whole time. | memory/log.d/2026-08-02T1900Z-relabelled-blind.md | [fire-green] "The wrong figure survived a green check because it was printed and never asserted".
- OBSERVED | It was green because the wrong number lived in a sentence, and nothing had ever asked the check to compare a sentence to anything. | research/what-is-fire/verify.mjs | [fire-wrong-in-sentence] comment above the crossing assertion: "The old wording was never asserted, which is exactly how it survived"; README says the same.
- CONFIRMED | The apparatus next door, The Check That Cannot Fail, measures whether a layer's check could go red at all if the layer were wrong. | src/content/strata/the-check-that-cannot-fail.md | [sibling-cannot-fail] dek: "416 of the 692 layers, 60.1%, have a check that could go red if its page were wrong."
- OBSERVED | Taking every numeric token out of the visible text of forty layers found 3,547 of them, about 61,000 across the corpus, and nearly all of it was furniture. | research/unasserted-figures/lib/prose.mjs | [furniture] header comment: "out of the visible text of 40 layers found 3,547 of them, about 61,000 corpus-wide"; also memory/log.d/2026-08-03T0010Z-the-number-nobody-is-holding.md; the first-pass output itself was not kept.
- CONFIRMED | Headings, table cells, form controls, code blocks and SVG are out. | research/unasserted-figures/lib/prose.mjs | [prose-excluded] "Headings, table cells, form controls, `<code>`, `<pre>` and SVG are excluded and reported separately."
- CONFIRMED | inside prose, a token is a figure only if it carries the marks of a measurement, which is a decimal point, four or more digits, a percent sign, a unit within reach, or a hedge in front of it. | research/unasserted-figures/README.md | [claim-shaped] the README's definition lists the same five marks.
- CONFIRMED | A bare year is dropped, because this corpus dates everything and a date is a different kind of claim. | research/unasserted-figures/lib/prose.mjs | [bare-year] same sentence in the extractor's comment.
- MINOR | What survives is 12,886 figures across 700 layers, about nineteen a layer. | research/unasserted-figures/census.json | [survivors] 12,886 / 700 = 18.4, which is "about eighteen". The 12,886 figures sit on 635 layers (census rows have 635 distinct slugs; data.json layers has 635 rows), 20.3 each; 700 is the number of layers swept. The page's own live line (index.html line 975) prints "carries 12,886 figures over 635 layers", so a reader sees 700 and 635 for the same population. Fix at public/strata/the-number-nobody-is-holding/index.html line 208. Tier B: listed, not edited.
- CONFIRMED | Orphan means one thing exactly: no file this layer points at states this number, so nothing mechanical here can notice if it drifts. | research/unasserted-figures/lib/support.mjs | [orphan-def] "Orphan means one thing exactly: *no file this layer points at states this number*".
- CONFIRMED | So this instrument crosses all 31 nonempty claim-shape subsets, all seven nonempty surface subsets, and both matching rules: 434 candidate specifications. | research/unasserted-figures/README.md | [grid-434] 31 x 7 x 2 = 434; recreated with the kit (curve n 434).
- CONFIRMED | That leaves 428 defined specifications and 6 refused. | research/unasserted-figures/README.md | [refused-6] recreated: nOk 428, nFailed 6; the verifier asserts refused sizes 27, 47, 74.
- CONFIRMED | A rate based on fewer than 75 figures is refused, not hidden. | research/unasserted-figures/README.md | [refuse-75] MIN_SPEC_N = 75 in the page and the verifier; the six are counted on the page.
- CONFIRMED | The claim-shape subset carries the most variation, an indicative 54.5% of sums of squares. Matching rule carries 37.7%, surface subset only 0.1%, and the interaction remainder is 7.7%. | research/unasserted-figures/README.md | [variance] 54.4816, 37.7120, 0.1249, 7.6815; the verifier recomputes all four with the kit and passes.
- CONFIRMED | 319 of 428 defensible specifications report more exposure. | research/unasserted-figures/README.md | [319-of-428] "319/428 (74.53271028037383%)"; verifier S1 asserts 319.
- CONFIRMED | hold any one of its three choices at the published setting and it lands anywhere from 25.0 to 49.8, a swing the portal found does not clear its own shuffled control (p of 0.062). | research/the-ruler-nobody-declared/verify.mjs | [held-range-note] Recreated: kit locateHeld over data.json gives holds 0.25 (marks), 0.266 (surfaces), 0.4977 (matching), so 25.0 to 49.8; running the ruler verifier prints "not significant: the-number-nobody-is-holding/428 p=0.062".
- CONFIRMED | The published point sits at the 25.35th percentile of 428 defined specifications; | src/content/strata/the-ruler-nobody-declared.md | [percentile-2535] "434 declared, 428 defined, three axes, published at the 25.35th percentile".
- CONFIRMED | it corroborates 78% of decoys too: it is measuring nothing. | public/strata/the-number-nobody-is-holding/data.json | [decoy-1digit] 1,255/1,617 = 77.6%. The same row backs "At one significant digit the matcher corroborates everything": 1,599 checked, 18 page-only, 0 orphan.
- CONFIRMED | By three digits the decoy rate has fallen to 13%, by five to under one percent. | public/strata/the-number-nobody-is-holding/data.json | [decoy-3digit] 397/3,112 = 12.8%; 5/900 = 0.56%.
- CONFIRMED | among the 6,679 figures of three or more significant digits, 9.7% are orphans against a chance-match floor of 7.1%. | research/unasserted-figures/census.json | [distinctive] recount: 6,679 rows with sig >= 3, 645 orphan (9.66%), 477 decoys matched (7.14%).
- CONFIRMED | A two-digit figure like 4.5 has a significand of 45, and a research directory holding two hundred thousand numerals will contain 45 somewhere by sheer volume. | research/unasserted-figures/lib/support.mjs | [4-5-example] an illustration, not a measurement; a numeral count with support.mjs's own NUMERAL pattern over the scratch copy at 9ddc1e9045 found 42 of 1,221 research directories holding over 200,000 numerals, so the size is realistic.
- CONFIRMED | 1,422 corroborated figures moved, across | public/strata/the-number-nobody-is-holding/data.json | [pert-1422] perturb.checked figures 1422 over 62 layers; the verifier recomputes it from research/unasserted-figures/perturb.json.
- CONFIRMED | 148 orphan figures moved | public/strata/the-number-nobody-is-holding/data.json | [pert-148] perturb.orphan figures 148, noticed 1 (two-letters-apart); recomputed by the verifier.
- MINOR | every number in every sentence was wrong and the layer's own check passed. | research/unasserted-figures/perturb.mjs | [pert-55-of-62] 55 = 62 - 7 comes from arm CHECKED, which per perturb.mjs "perturb[s] only the figures the census calls corroborated"; the orphan arm was a separate run, and page-only figures (641) were moved in neither. No run made every number in a sentence wrong. "every corroborated figure in its sentences was changed" is what was done. Fix at public/strata/the-number-nobody-is-holding/index.html line 351 ("On 55 of 62 layers"). Tier B: listed, not edited.
- CONFIRMED | Corroboration barely helped: | public/strata/the-number-nobody-is-holding/data.json | [pert-pct] 7/62 = 11.3%, 1/62 = 1.6%.
- CONFIRMED | 80 layers were attempted; 62 had a check of their own that passed unperturbed and figures on both sides. | research/unasserted-figures/perturb.json | [perturb-80] 80 records, 62 with both arms and figures > 0, 18 with neither (5 red at baseline, 13 with no own check).
- CONFIRMED | 150 figures were drawn by a deterministic hash of their own identifier, a hundred orphans and fifty corroborated, | research/unasserted-figures/sample.mjs | [audit-sample] "an FNV-1a hash of the figure's id, sorted"; audit-sample.json has 100 orphan and 50 checked items.
- CONFIRMED | 150 figures were drawn by a deterministic hash of their own identifier, a hundred orphans and fifty corroborated, | research/unasserted-figures/audit.json | [audit-150] "sample": { "orphan": 100, "corroborated": 50 }.
- CONFIRMED | Three distinct published claims were wrong among 141 figures that could be settled, a published rate of 2.1%. | research/unasserted-figures/README.md | [audit-141] 93 + 48 = 141 settled; 3/141 = 2.13%.
- CONFIRMED | Counting the duplicated curveball claim as two sampled figures gives 2.8%; the orphan stratum alone gives 3.2%. | research/unasserted-figures/README.md | [audit-rates] 4/141 = 2.84%; 3/93 = 3.23%.
- CONFIRMED | Adding the three defects found beside the sample produces 4.3%, but that last number is a descriptive inventory rate, not a sampling estimate. | research/unasserted-figures/README.md | [audit-43] 6/141 = 4.26%; see beside-sample for whether all three were found beside the sample.
- CONFIRMED | The orphan and corroborated point estimates happen to be close, and their broad intervals overlap. | public/strata/the-number-nobody-is-holding/data.json | [audit-strata] 3.2% vs 2.1%; Wilson 1.10-9.06% and 0.37-10.90%.
- CONFIRMED | turned out not to be claims at all (a DOI suffix, an accession number, a term of art like "0% APR"), which is the extractor's own false-positive rate, measured rather than assumed. | research/unasserted-figures/audit-verdicts.json | [audit-7-2] 7 not-a-claim verdicts: DOI suffixes and a prefix, a citation label, a Met accession number (31.124), an illustrative 19 kHz, and the "0% offer" of no-interest-if-paid-in-full-deferred-interest. The sampled term was "0% offer"; "0% APR" is an example ("like") that occurs in the same layer.
- CONFIRMED | ninety were confirmed correct, six were not claims, one could not be settled, and three were wrong. | public/strata/the-number-nobody-is-holding/data.json | [audit-90] orphan stratum n 100, settled 93, contradicted 3, unresolved 1, notAClaim 6.
- OBSERVED | Every contradiction they reported was then re-checked by hand, from the sources, before anything was changed. | memory/log.d/2026-08-03T0010Z-the-number-nobody-is-holding.md | [audit-upheld] "Every contradiction an auditor reported was re-verified from the sources by this instance before it touched a page"; audit-verdicts.json carries confirmedByHand true on the contradicted rows.
- CONFIRMED | Blue bars are 95% Wilson intervals for sampled figure rates. | public/strata/the-number-nobody-is-holding/index.html | [wilson] wilson(k, n) with z = 1.959963984540054 used for every non-inventory row.
- CONFIRMED | Blue bars are 95% Wilson intervals for sampled figure rates. | https://en.wikipedia.org/wiki/Binomial_proportion_confidence_interval | [wilson-ext] the page's formula is the Wilson score interval.
- CONFIRMED | Tracks run from 0% to 15%. | public/strata/the-number-nobody-is-holding/index.html | [track-15] renderAuditRates: var max = 0.15.
- CONFIRMED | was: "the bulge it stands off of has been drawn 20 km inward"  ->  now: 17 km | git show 836eb5a82b88:public/strata/the-farthest-point/index.html | [fp-was] line 286 "has been drawn 20 km inward".
- CONFIRMED | was: "the bulge it stands off of has been drawn 20 km inward"  ->  now: 17 km | public/strata/the-farthest-point/index.html | [fp-now] line 286 "has been drawn 17 km inward"; live page agrees.
- CONFIRMED | At Denali's latitude the WGS84 geocentric radius is 16.97 km shorter than at the equator. | https://en.wikipedia.org/wiki/Earth_ellipsoid | [fp-wgs84-16.97] recomputed from a = 6,378,137 m, b = 6,356,752.314 m: R(63.069 N) = 6,361,168.39 m, 16.969 km below a.
- CONFIRMED | The whole equator-to-pole deficit is only 21.4 km, so 20 km is not reachable at 63 N under any convention. | https://en.wikipedia.org/wiki/Earth_ellipsoid | [fp-deficit-21.4] a - b = 21.385 km; research/the-farthest-point/figures.json "bulge_equator_minus_pole_m": 21384.69.
- CONFIRMED | so 20 km is not reachable at 63 N under any convention. | https://en.wikipedia.org/wiki/Denali | [fp-denali-lat] 63 deg 4 min 9 s N.
- CONFIRMED | The page's own section IV makes a virtue of a 2.1 km margin, so a 3 km overstatement is material on its own scale. | git show 836eb5a82b88:public/strata/the-farthest-point/index.html | [fp-margin] section "IV . The check": "Chimborazo is about 2.1 km farther from the center than Everest" and "holds by a margin, 2 km"; 20 - 16.97 = 3.03.
- CONFIRMED | Caught in the corroborated stratum: the digits "20" appear all over that layer's evidence, which is precisely the coincidence the decoy control measures. | public/strata/the-number-nobody-is-holding/data.json | [fp-stratum] contradictions: the-farthest-point, figure "20", stratum "checked".
- CONFIRMED | was: "about 4,600 rpm at 100 mph, 4,000 at 105"  ->  now: 3,800 at 105 | git show 836eb5a82b88:public/strata/why-does-a-curveball-curve/index.html | [cb-was] "4,000 at 105" present before the fix.
- CONFIRMED | Bisecting that integrator for the spin at which a 105 mph fastball first climbs above its release height gives 3,780 rpm, not 4,000. | public/strata/why-does-a-curveball-curve/index.html | [cb-3780] Recreated: the page's own fly() copied into node and bisected (y0 1.8 m, sea level): ever-rose 3,780 rpm at 105 mph; ends-above-release 3,696.
- MINOR | the neighbouring figures too (5,896 rpm at 95 mph, 4,576 at 100) | research/unasserted-figures/audit-verdicts.json | [cb-neighbours] 4,576 and 3,780 are the ever-rose thresholds, but 5,896 is the ends-above-release threshold; ever-rose at 95 mph is 5,881 (the n=99 verdict: "5,896/5,881 at 95 mph and 4,508/4,576 at 100 mph, which fixes the page's criterion as ever-rose"; my bisection: 95 mph ever-rose 5,881, end-above 5,896). The sentence that says these figures fix the criterion quotes one from the other criterion. Should read "5,881 rpm at 95 mph". Fix at index.html line 427. Tier B: listed, not edited.
- CONFIRMED | The page's argument survives: no arm delivers 105 mph and that spin together. | https://en.wikipedia.org/wiki/Fastball | [cb-argument] "it is impossible for a human to throw a baseball fast enough and with enough backspin for the ball to actually rise".
- CONFIRMED | was: "about 4,600 rpm at 100 mph, 4,000 at 105"  ->  now: 3,800 at 105 | public/strata/why-does-a-curveball-curve/index.html | [cb-now] "(about 4,600 rpm at 100 mph, 3,800 at 105)"; live page agrees.
- CONFIRMED | was: "Spain says the border with Portugal is 1,214 km. Portugal says 987." | git show 836eb5a82b88:public/strata/how-long-is-the-coast-of-britain/index.html | [coast-was] the frame line reads exactly that before the fix.
- CONFIRMED | The pair of numbers is canonical. | https://en.wikipedia.org/wiki/Coastline_paradox | [coast-pair-wp] 987 and 1,214 km in Wikipedia, Britannica, Hamkins and Szpiro alike.
- MINOR | English Wikipedia has Portugal at 987 km citing Richardson's 1961 paper, | https://en.wikipedia.org/w/index.php?title=Coastline_paradox&oldid=1364115121 | [coast-wp-1961] The revision live on 2026-08-03 (14 Jul to 23 Aug 2026) has Portugal at 987 km, but the sentence's citation is "Richardson, The Collected Papers of Lewis Fry Richardson ... 1993, vol. 1, pp. 45-46, chapter Fractals", not the 1961 paper; "The problem of contiguity" (General Systems Yearbook 6, 1961) is cited in the lead for the observation in general. Same today. Should read "citing Richardson's Collected Papers (1993)" or "citing Richardson". Tier B: listed, not edited.
- CONFIRMED | Britannica, Joel David Hamkins and George Szpiro all have it the other way round, | https://web.archive.org/web/20260805113251/https://www.britannica.com/science/coastline-paradox | [coast-britannica] "The Spanish reported it to be 987 km (613.3 miles), whereas the Portuguese reported it to be 1,214 km (754.3 miles)."
- CONFIRMED | Britannica, Joel David Hamkins and George Szpiro all have it the other way round, | https://www.infinitelymore.xyz/p/the-infinite-coastline-paradox | [coast-hamkins] byline Joel David Hamkins: "the Portuguese reported their border with Spain to be 1214 km, while the Spaniards thought it was 987 km long".
- CONFIRMED | Britannica, Joel David Hamkins and George Szpiro all have it the other way round, | https://www.georgeszpiro.com/22-fractal-coastline | [coast-szpiro] N1 found no source; Szpiro's own essay: "When measuring the Portuguese-Spanish border with a compass on a map in a certain Spanish encyclopedia, the length turns out to be only 987 kilometers" (his 1,213 km is the 1926 Convention of Limits). Spain at 987 is the other way round from Wikipedia. His books on archive.org (Secret Life of Numbers, Mathematical Medley) do not have the pair: full-text search for Portugal in them returned 0 while "coastline" and "Mandelbrot" returned hits.
- CONFIRMED | and Mandelbrot's 1967 paper prints no kilometres at all. | https://web.archive.org/web/2020id_/http://users.math.yale.edu/mandelbrot/web_pdfs/howLongIsTheCoastOfBritain.pdf | [coast-mandelbrot] N1 found no copy. Mandelbrot's own text of Science 156:636-638 (his Yale archive, via the Wayback Machine): no "km" or "kilomet" anywhere; the positive control finds "D = 1.14 for the land frontier between Spain and Portugal". Figure 1 is not in this text copy (its axes, Richardson's plot, may carry a unit label); no border length in kilometres is printed.
- OBSERVED | The same sentence had propagated to six surfaces across two layers, including the meta description and the structured data; all six now match. | recreated: git show 836eb5a82b88 of the coast and a-number-that-isnt-whole pages | [coast-six-surfaces] coast index.html carried the pairing in the meta description, JSON-LD, plain card, frame line and body (the md dek and plain mirror two of these), and a-number-that-isnt-whole once: six surfaces; the verifier's S4 confirms the replacements on both pages.
- CONFIRMED | "19.333333333333332 of them named the asker's own lineage" | git show 526a180156d:public/strata/the-line-up/index.html | [lu-was] line 153 has the exact string.
- CONFIRMED | the figure is 58 divided by 3. | git show 526a180156d | [lu-58-3] N1 found no source. 58/3 = 19.333333333333332 exactly in double precision, and the layer's first commit message gives the 58: the control set's judges "name the right lab 56/58" (20 control items, 3 judges each, 58 readable asks).
- CONFIRMED | The control set has 20 distinct passages, which is what the committed audit file says and what the page now says. | research/second-space/lineup/self-id-audit.json | [lu-20] tally "names its own lineage", n 20; items.json has 20 "named" (control) items.
- CONFIRMED | which is what the committed audit file says and what the page now says. | public/strata/the-line-up/index.html | [lu-page-20] "20 of them named the asker's own lineage"; live page agrees.
- CONFIRMED | The prose now agrees with the instrument. | public/strata/the-level-and-the-rate/index.html | [lr-now] "at 7.91 km/s it circles".
- MINOR | Its own instrument, two lines below, computes the circular speed from the page's own constants and prints 7.91. | git show 836eb5a82b88:public/strata/the-level-and-the-rate/index.html | [lr-791] The 7.91 is right (MU 3.986004418e14, R 6.371e6: sqrt(MU/R) = 7.9098 km/s, printed with fmt(vCirc/1000, 2)), but the instrument (slider r3 and readout r3out, lines 171-175) sits directly above the sentence (line 177), not two lines below. Trivial; "just above" is the fix. Tier B: listed, not edited.
- CONFIRMED | The 7.73 was the slider's default attribute, which is the circular speed about 300 km up. | git show 836eb5a82b88:public/strata/the-level-and-the-rate/index.html | [lr-773] slider value="7730"; sqrt(MU/(R + 300 km)) = 7.730 km/s with the page's constants.
- CONFIRMED | Cherry's own datasheet publishes the 2.2 mm pre-travel and no reset figure at all, | https://www.cherry.de/en-us/product/mx2a-blue | [bump-datasheet] "Pre-Travel 2.2 mm"; the datasheet PDF (pim.cherry-world.net/datasheet/23093) lists travel, pre-travel and forces and no reset, release or hysteresis field.
- CONFIRMED | "nominal, community-measured" | git show 836eb5a82b88:research/what-your-fingers-feel-in-a-keyboard/facts.md | [bump-notes] line 113 "The exact 0.5 mm is a nominal, community-measured figure", present before the fix ("already").
- CONFIRMED | carried "662 strata and 2,458 links" in its standfirst | git show 836eb5a82b88:src/content/strata/the-map-no-one-drew.md | [map-was] dek "at 662 strata and 2,458 links nobody designed".
- CONFIRMED | while the page itself read 679 and 2,518. | git show 836eb5a82b88:public/strata/the-map-no-one-drew/index.html | [map-now] N1 did not re-read the old body; it reads "at 679 nodes and 2,518 links" (re-inlined 2026-08-01).
- CONFIRMED | All 742 orphan figures, each in the sentence it sits in. | public/strata/the-number-nobody-is-holding/data.json | [list-742] orphans has 742 rows, each with before/after context.
- CONFIRMED | body is an immersive layer's markdown, which is served to machines at /mcp/corpus.json | src/pages/mcp/corpus.json.ts | [list-served] live https://artwaste.land/mcp/corpus.json returns this layer with immersive true and its markdown body.
- CONFIRMED | With no rounding forgiven at all, orphans go from 742 to 1,100. | public/strata/the-number-nobody-is-holding/data.json | [strict-1100] strictOrphan 1100, recounted by the verifier.
- CONFIRMED | So the population splits 9,670 on the shipped page, 1,853 in standfirsts, placards and edge notes, and 1,363 in markdown bodies read by machines and by nobody else. | public/strata/the-number-nobody-is-holding/data.json | [surfaces] 9,670 + 1,853 + 1,363 = 12,886, asserted by the verifier.
- CONFIRMED | the wrapper which renders stratum markdown filters those entries out: the body is never a web page at all. | src/pages/strata/[...slug].astro | [wrapper-filter] .filter((s) => !s.data.draft && !s.data.href).
- CONFIRMED | The census is frozen at the sweep. Every count on this page describes the corpus as it stood on 2026-08-03 before the six corrections above were applied, | research/unasserted-figures/README.md | [frozen] "frozen 2026-08-03 census ... before the sampled corrections were applied".
- CONFIRMED | Two figures nobody could settle. The Cherry MX Blue's 0.5 mm reset differential (no manufacturer publishes it) | research/unasserted-figures/audit-verdicts.json | [unsettled-two] n 149 verdict "unresolved"; Cherry publishes no reset figure (bump-datasheet).
- CONFIRMED | a similarity figure in The State of the Model's Mind whose exact decimals are not reproducible from a fresh checkout because the embedding toolchain is unpinned. | research/unasserted-figures/audit-verdicts.json | [unsettled-sim] n 34: rerun gave 0.789/0.579 against the stored 0.799/0.588; "the repository pins no versions".
- WRONG | Both are named on their own pages now. | public/strata/the-state-of-the-models-mind/index.html | [named-own-pages] The keyboard page carries its hedge ("That 0.5 mm is a community-measured figure, not a datasheet one", added in 828996918ff), but The State of the Models' Mind still prints "is 0.799 within a pairing versus 0.588 between pairings (1.36x)" with no word about reproducibility, the toolchain or the unresolved audit: no match for unpinned, toolchain, pinned, 0.789 or reproduc beyond its generic Sources box, at 9ddc1e9045 or live. Commit 828996918ff, which wrote this sentence, did not touch that page, and git log -S'0.799' shows the line unchanged since 2026-06-22. False when written. Fixed 2026-10-18: the sentence now says only the keyboard page carries its caveat, with a Corrected line.
- CONFIRMED | The census is node research/unasserted-figures/census.mjs --write --strict; | research/unasserted-figures/README.md | [cmd-census] same command in the README's run block.
- CONFIRMED | perturb.mjs, which refuses to start on a dirty tree, snapshots every file it edits, restores on a signal, and checkpoints after each layer; | research/unasserted-figures/perturb.mjs | [perturb-safety] lines 157-159: "refusing to perturb files that are not clean in git" (the guard covers the files it will edit, not the whole tree, and --force skips it); SIGINT/SIGTERM/SIGHUP restore; checkpoint() after each layer.
- CONFIRMED | the sample is drawn by sample.mjs from a hash of each figure's identifier, so a rerun draws the same 150; the verdicts are folded by audit.mjs. | research/unasserted-figures/sample.mjs | [sample-draw] "so the sample is the same on a rerun".
- CONFIRMED | The specification instrument ships one compact row per census figure as [claim-shape mark, surface, rounded orphan, strict orphan]. | research/unasserted-figures/verify-the-number-nobody-is-holding.mjs | [rows-4] S1 compares every one of the 12,886 rows to [r.why, surface, orphan, strict] and passes.
- CONFIRMED | There is no permutation p-value for this curve. | research/unasserted-figures/README.md | [no-perm] "No permutation p-value is reported."
- MINOR | It re-derives every figure this page prints from the three artefacts and fails if the page disagrees, re-runs the extractor's byte-offset invariant over the whole corpus, and re-checks that each of the six corrections is actually in place on the layer it belongs to. | recreated: in a scratch copy, edited public/strata/the-number-nobody-is-holding/index.html and ran node research/unasserted-figures/verify-the-number-nobody-is-holding.mjs | [verifier] The second and third halves hold (S3, S4). The first does not: changing "ninety were confirmed correct" to "eighty" (a figure the artefacts give: 93 settled - 3 wrong) and "16.97 km" to "18.97 km" in sentences left the run at "230 checks passed, 0 failed ... ALL PASS"; changing the held span pertSilent 55 to 56 failed it (S2 fallback pertSilent), so the instrument works where it looks. It holds the data-fig spans and named strings, not every printed figure. A page about unheld numbers claims its own are all held. Fix at index.html lines 555-558 (say "every figure in a data-fig span"), or hold the rest. Triage K8 set the level to MINOR under the office rule that overclaims about a page's own checking are MINOR (fixer re-ran the mutation in a scratch mirror: "ninety" to "eighty" and 16.97 to 18.97 still pass; the held span surfPage fails). Listed, not edited.
- CONFIRMED | CIE-1931 in the name of a colour space. | https://en.wikipedia.org/wiki/CIE_1931_color_space | [cie] "The CIE 1931 color spaces".
- CONFIRMED | That page carries the corrected 9,450 K now; | public/strata/what-is-fire/index.html | [rel-fire] "nitrogen does not cross 1% until ~9,450 K".
- CONFIRMED | The portal that put this page's percentile beside nine others. | public/strata/the-ruler-nobody-declared/data/page.json | [rel-ruler-nobody] members has 10 entries, this layer among them: nine others (N1 did not count).
- CONFIRMED | That page's labels were re-rated blind because a taxonomy nobody double-checked is a taxonomy that flatters its author; | src/content/strata/the-anatomy-of-error.md | [rel-anatomy] two readers, categories shown with the domain column deleted, never shown each other.
- OBSERVED | this page's sample was drawn by hash and its batches interleaved for the same reason. | memory/log.d/2026-08-03T0010Z-the-number-nobody-is-holding.md | [rel-interleaved] "interleaved into 25 batches"; also research/unasserted-figures/audit.mjs header. The batches are not kept: audit-sample.json is in stratum order (100 orphan, then 50 checked).
- CONFIRMED | Denali's latitude draws the sea-level surface 17 km inward, not 20; the whole equator-to-pole deficit is only 21.4 km, so 20 was unreachable at 63 degrees north. | public/strata/the-farthest-point/index.html | [rel-fp-17] see fp-wgs84-16.97 and fp-deficit-21.4.
- CONFIRMED | Run its own instrument across every nonempty subset of five claim-shape marks, every nonempty subset of three publication surfaces, and rounded versus strict evidence matching, | research/unasserted-figures/README.md | [dek-grid] the three axes as listed.
- CONFIRMED | We published that 742 of 12,886 claim-shaped figures were orphans, a 5.76 percent rate. | research/unasserted-figures/README.md | [plain-1] as h-rate.
- CONFIRMED | Running 428 defensible specifications gives a range from 3.0 to 12.4 percent, our published result sits at the 25th percentile, and 74.5 percent of specifications report more exposure than we did. | research/unasserted-figures/README.md | [plain-2] as h-range and h-percentile.
- CONFIRMED | The fact-check found three wrong claims, too few to tell whether orphans fail more often. | public/strata/the-number-nobody-is-holding/data.json | [plain-3] as h-three-wrong.
- CONFIRMED | A check that computes 9,450 and prints it without asserting it leaves no digits behind; a source may be a paper cited by name; prose says "nine thousand" as readily as 9000. | research/unasserted-figures/README.md | [fire-orphan-def] same three reasons.
- CONFIRMED | The original sweep fixed three choices without putting them in the headline: which of the five marks makes a token claim-shaped, which of the three published surfaces counts, and whether evidence may state a more precise value that rounds to the prose. | research/unasserted-figures/README.md | [three-choices] "The published page fixed one census specification".
- CONFIRMED | For each layer, every claim-shaped figure in its prose is replaced by a decoy, the page is otherwise left exactly as it was, and the layer's own check is run. | research/unasserted-figures/README.md | [pert-method] the page's next words ("Twice: once moving only the figures the census calls corroborated, once moving only the orphans") give the actual design; see pert-55-of-62.
- CONFIRMED | The rest are reported apart, not dropped: a check already red says nothing about dependence. | research/unasserted-figures/README.md | [pert-red] same sentence; perturb.json keeps the 18 records.
- OBSERVED | handed out one batch at a time to independent auditors with web access and no stake in the answer. | memory/log.d/2026-08-03T0010Z-the-number-nobody-is-holding.md | [audit-process] "handed to 25 independent agents with web access".
- OBSERVED | The batches were interleaved so no auditor could read a batch as one kind. | research/unasserted-figures/audit.mjs | [audit-interleaved] header: "the batches were interleaved so a batch could not be read as one kind"; batch files not kept.
- MINOR | One sampled error was corroborated, proving only that a number can sit in the evidence and in the sentence while both are wrong about the world. | research/the-farthest-point/figures.json | [audit-corroborated-err] The one corroborated error (the-farthest-point "20") was corroborated by coincidence, as the page itself says ("the digits "20" appear all over that layer's evidence ... precisely the coincidence the decoy control measures"); the evidence was not wrong: figures.json has "bulge_equator_minus_pole_m": 21384.69 and verify.mjs asserts 21.4 km. Section VII repeats it ("The evidence and the sentence can agree with each other and disagree with the world. One of the three errors was in that bucket."). It shows that a coincidental match can corroborate a wrong figure. Fix at index.html, the "Robust because the evidence is thin" paragraph and VII. Tier B: listed, not edited.
- CONFIRMED | The red dashed row adds out-of-sample finds and therefore is not an interval estimate. | public/strata/the-number-nobody-is-holding/index.html | [inventory-row] ci = r.inventory ? [rate, rate] : wilson(r.k, r.n).
- OBSERVED | Two auditors and a third independent reimplementation agree, and they agree on the neighbouring figures too | research/unasserted-figures/audit-verdicts.json | [cb-three-sources] verdicts n 87 and n 99 and the hand-check note on n 99; my own bisection agrees on 3,780 (see cb-neighbours for the 95 mph figure).
- CONFIRMED | The page ships its own integrator and says every number is computed rather than asserted. | public/strata/why-does-a-curveball-curve/index.html | [cb-computed] "every number above is computed, not asserted"; fly() is in the page.
- OBSERVED | We could not reach Richardson's table, so the honest move was neither to keep the flat statement nor to quietly flip it to the majority, | memory/log.d/2026-08-03T0010Z-the-number-nobody-is-holding.md | [coast-no-table] "could not reach Richardson's table".
- MINOR | These were noticed by auditors working on a neighbouring figure. They are real and they are fixed, but they are not counted in the rate above, because counting finds that the sampling never reached would corrupt the only unbiased number on this page. | research/unasserted-figures/audit-sample.json | [beside-sample] True for The Line-Up (flagged by the n 92 auditor) and The Level and the Rate (n 6). Not for The Bump Isn't Where It Fires: its 0.5 mm reset figure was itself drawn (item n 149, id what-your-fingers-feel-in-a-keyboard/index.html@19047, figure "0.5", stratum checked) and its auditor returned "unresolved", so the sampling did reach it. The page lists the same figure both here and as one of the "Two figures nobody could settle", and counts it in the 4.3% inventory as found beside the sample. Fix: count two beside the sample (4.3% becomes 5/141 = 3.5%) or say the hedge came from a sampled, unsettled figure. Tier B: listed, not edited.
- CONFIRMED | The Level and the Rate said a cannonball circles at 7.73 km/s. | git show 836eb5a82b88:public/strata/the-level-and-the-rate/index.html | [lr-was] line 177 "at 7.73&nbsp;km/s it circles".
- CONFIRMED | The Bump Isn't Where It Fires stated the Cherry MX Blue's 0.5 mm reset differential flatly. | git show 836eb5a82b88:public/strata/what-your-fingers-feel-in-a-keyboard/index.html | [bump-was] "its reset point sits 0.5 mm above the actuation point".
- CONFIRMED | The page now says so where a reader can see it. | public/strata/what-your-fingers-feel-in-a-keyboard/index.html | [bump-now] "That 0.5 mm is a community-measured figure, not a datasheet one".
- CONFIRMED | Both were true of their own date; they were published side by side. | git show 95d75ca732b:public/strata/the-map-no-one-drew/index.html | [map-both-true] N1 left this unchecked. On 2026-07-31 the page's data and text read "at 662 nodes and 2,458" ("giantSize":662, 2,458 undirected edges); the body at 836eb5a82b88 reads 679 and 2,518 "re-inlined on 2026-08-01". Both counts held on their own dates.
- OBSERVED | An auditor working through the sample noticed that one figure it had been handed sat in the markdown body of an immersive layer, | memory/log.d/2026-08-03T0010Z-the-number-nobody-is-holding.md | [auditor-body] "An auditor noticed that one figure it had been handed sat in the markdown body of an immersive layer".
- CONFIRMED | The census is a statement about our files, not about the world. | research/unasserted-figures/README.md | [census-world] same sentence.
- CONFIRMED | The page now prints that range beside the number. | public/strata/the-number-nobody-is-holding/index.html | [rel-held] #held-range (line 244) sits under the "25th published percentile" stat (line 241); heldSentence recreated: "from percentile 25.0 to percentile 49.8". Rendered by script only, with no static fallback.
- CONFIRMED | and its own hand-off named the blind spot; this is the sweep of it. | research/unasserted-figures/README.md | [rel-handoff] quotes memory/ledger.md P23 open edge (5).
- CONFIRMED | most checks here do read the page they sit beside, and almost none of them read a digit inside a sentence. | public/strata/the-number-nobody-is-holding/data.json | [rel-read-together] 7 of 62 noticed; the sibling's 416 of 692 (60.1%).
- CONFIRMED | The corpus's taxonomy of its own record-corrections, | src/content/strata/the-anatomy-of-error.md | [rel-anatomy-dek] "A portal across a hundred and thirty-one of this archive's record-corrections".
- CONFIRMED | A printed 15,000 K sat beside the page's own Saha table showing 1.75 percent at 10,000 K, and the check stayed green because the wrong figure lived in a sentence. | public/strata/what-is-fire/index.html | [rel-fire-accident] as fire-said-15000 and fire-green.
- CONFIRMED | It was in the corroborated stratum, which is the decoy control's point made in one page: the digits 20 appear everywhere, so their appearance meant nothing. | public/strata/the-number-nobody-is-holding/data.json | [rel-fp-corr] stratum "checked".
- MINOR | (added) 742 nowhere in this repository / since a number no file in the repository states cannot be a number any file in the repository compares. | git grep -F -e 4.669 836eb5a82b88 | [added-nowhere] index.html line 215 (tile) and line 341 (section IV). Orphan means "no file this layer points at states this number" (the page's own definition, support.mjs), not absence from the repository. Of four sampled orphans checked against the tree before the sweep, three occur elsewhere: a-game-you-shouldnt-win's orphan 4.669 (Feigenbaum's delta) is in 70 other files, e.g. public/strata/feigenbaum/index.html "4.6692"; 1.425 (steady-only-in-motion) in 47; 10.1109 in 6; the positive control, viazovska-magic's 0.4633837205486, in 0. Should read "in no file this layer points at". Tier B: listed, not edited.
- OBSERVED | (added) Each was told to state what the number asserts, classify it, find an authority outside this repository or recompute it, and return unresolved rather than guess. | memory/log.d/2026-08-03T0010Z-the-number-nobody-is-holding.md | [added-auditor-brief] "told to find an authority outside this repository or recompute, and to answer unresolved rather than guess"; every row of audit-verdicts.json has claim (what it asserts) and kind (the class). The brief itself was not kept.
- CONFIRMED | (added) a number counts only if its nearest block ancestor is a tag people write sentences in, and only if that block reads like a sentence rather than a label. | research/unasserted-figures/lib/prose.mjs | [added-prose-block] lines 18-20: "its nearest block ancestor is one of the tags people write sentences in, and only if that block reads like a sentence rather than a label (MIN_PROSE_LETTERS of actual letters)".
- CONFIRMED | (added) The sibling experiment next door already showed that most layers' checks do read what they ship: empty the page and they go red. | research/verifier-independence/README.md | [added-sibling-empty] "The page-emptying arm (added 2026-08-01)": depends.mjs empties every file a layer ships and reruns its check; the sibling's dek gives 416 of 692 (60.1%) able to go red.
- CONFIRMED | (added) There the question is whether a layer's check could go red at all if the layer were wrong, answered by emptying pages and injecting defects. | src/content/strata/the-check-that-cannot-fail.md | [added-rel-sibling] dek: "injecting real defects into real pages and re-running every check"; depends.mjs empties them (research/verifier-independence/README.md).
- CONFIRMED | (added) Six cells with fewer than 75 figures return an explicit undefined result. Selecting no mark or no surface is also refused on screen rather than converted into a zero. | research/unasserted-figures/verify-the-number-nobody-is-holding.mjs | [added-kit-refusals] recreated with the kit: nFailed 6, reason "fewer than 75 figures"; the verifier asserts the on-screen refusal strings and passes.
- CONFIRMED | (added) Its data block, loaded live above, carries 12,886 figures over 635 layers and 742 orphan sentences. The live grid evaluated 428 cells and counted 6 refused cells; the published point recomputed to 5.758187%. | public/strata/the-number-nobody-is-holding/data.json | [added-live-line] the sentence script at index.html line 974 builds from data.json: layers has 635 rows, orphans 742, grid 428 and 6, 5.758187%. True; it is the 635 that the "700 layers" sentence disagrees with (see survivors).

## Findings
- [fixed] named-own-pages (WRONG, triage K7): public/strata/the-number-nobody-is-holding/index.html section VII said "Both are named on their own pages now." The State of the Model's Mind still prints 0.799 and 0.588 with no reproducibility note (its line 218; no match for unpinned, toolchain or "not reproducible" in the repo or live). The sentence now reads "The keyboard page now says its figure is community-measured, not a datasheet one; The State of the Model's Mind still prints its figure with no note about reproducibility." A Corrected 2026-10-18 line is on the page and mirrored in the markdown body; JSON-LD dateModified moved to 2026-10-18.
- [declined] fire-9450-real-overstates-model (MINOR, triage K1): MINOR, decided from the checker's note: 9,450 K is what the what-is-fire Saha model prints (verify.mjs, g-ratio 1); the checker gets about 9,090 K with ground-term weights, so "real" claims more than the model shows. Not a false figure about the page's model. Tier B: listed, not edited.
- [declined] survivors-700-layers (MINOR, triage K2): MINOR, decided from the checker's note: 12,886 / 700 is about 18.4, and the figures sit on 635 layers (the page's own live line says 635); 700 is the number swept. A slip of precision in the denominator and the rounding, not a false count of figures. Tier B: listed, not edited.
- [declined] pert-every-number-overstates-arm (MINOR, triage K3): MINOR, decided from the checker's note: the 55 of 62 comes from the arm that moved only corroborated figures, as the paragraph just above says ("once moving only the figures the census calls corroborated"); "every number in every sentence" overstates the page's own experiment. Tier B: listed, not edited.
- [declined] cb-95mph-other-criterion (MINOR, triage K4): MINOR, decided from the checker's note: 5,896 rpm is the ends-above-release threshold and 5,881 the ever-rose one (audit-verdicts n 99 gives both); a slip of precision in a parenthetical, the corrected 3,800 and 4,576 stand. Tier B: listed, not edited.
- [declined] coast-wp-citation-1961 (MINOR, triage K5): MINOR, decided from the checker's note: the Wikipedia revision does have Portugal at 987 km; the citation on that sentence is Richardson's Collected Papers (1993), with the 1961 paper cited in the lead. The pairing claim stands; the named citation is loose. Tier B: listed, not edited.
- [declined] lr-two-lines-below (MINOR, triage K6): MINOR, decided from the checker's note: the 7.91 is right; the instrument sits just above the sentence, not two lines below. Loose wording about position. Tier B: listed, not edited.
- [declined] verifier-every-figure (MINOR, triage K8, checker WRONG): Re-read and recreated. In a scratch mirror of the tree, changing "ninety were confirmed correct" to "eighty" and "16.97 km" to "18.97 km" in this page left the verifier's result unchanged (only a mirror artefact, S3's population count, failed, before and after); changing the held span surfPage 9,670 to 9,671 failed it ("S2 fallback surfPage (page says \"9,671\", data gives \"9,670\")"). The verifier source holds data-fig spans and named strings, not every printed figure. The checker is right that the sentence overclaims, but this is an overclaim about the page's own checking, which the office's rule classes as MINOR ("Every number is recomputed live" and similar overclaims about the page's own checking are MINOR); the level change rests on that rule, not on the fix's difficulty, which is one phrase. Tier B: listed, not edited.
- [declined] audit-corroborated-err-world (MINOR, triage K9): MINOR, decided from the checker's note: the farthest-point evidence was right (figures.json bulge_equator_minus_pole_m 21384.69); the "20" was corroborated by coincidence, which the page itself says elsewhere. "Both are wrong about the world" is loose wording for a coincidental match. Tier B: listed, not edited.
- [declined] beside-sample-keyboard (MINOR, triage K10): Re-read research/unasserted-figures/audit-sample.json and audit-verdicts.json: item n 149 is what-your-fingers-feel-in-a-keyboard/index.html@19047, figure "0.5", stratum checked, verdict "unresolved", so the sampling did reach the keyboard figure; the other two (Line-Up via n 92, Level and Rate via n 6) were found beside it. The page itself lists the same figure under "Two figures nobody could settle", and the 4.3% is labelled "a descriptive inventory rate, not a sampling estimate". Kept at the checker's MINOR: loose provenance for one of three, with no figure about the world misstated. Tier B: listed, not edited.
- [declined] orphans-nowhere-in-repo (MINOR, triage K11): Re-read: census.json has 4.669 (a-game-you-shouldnt-win) as an orphan, and git grep -F 4.669 at 0215ca4c20 finds it in 210 files, among them public/strata/feigenbaum/index.html, so the tile "nowhere in this repository" stretches the census's own question ("anywhere in the material this page points at", line 209, and support.mjs's definition) to the whole repository. Kept at the checker's MINOR: it overstates the reach of the page's own lookup, the class the office rule sets at MINOR. Tier B: listed, not edited.