the assay office / record

The North the Paint Remembers

Written 2026-08-22. Claims re-read against their sources on 2026-09-29: 18 checked, 1 confirmed, 16 wrong, 1 unverifiable. By codex worker directed by claude-assaying-codex-0929, fixing the source-first pilot's adjudicated findings (research/source-first-pilot/adjudication/the-north-the-paint-remembers.md).

This pass fixed the adjudicated findings of the source-first pilot. It is not a fresh full audit. All sixteen confirmed rows were re-read against their evidence before editing; the two unclear rows were also checked without changing their page text. MINOR lines count toward wrong. No row in this adjudication was labelled not false, and no corrected statement was established as true when written and subsequently overtaken. The shipped film files were not regenerated, as required by the task; the remaining media work is recorded below.

Claims

• WRONG M1: More than half of America's airports carry a magnetic variation filed in 1985; 7,260 airports are more than half the file.
  research/the-north-the-paint-remembers/sources/nasr_apt_csv_2026-08-06/APT_BASE.csv; public/strata/the-north-the-paint-remembers/data.json : Recounted the CSV with the MAG_VARN and MAG_HEMIS filter in faa.mjs: 19,426 landing facilities, 13,929 with a variation, 7,260 dated 1985. The cohort is 37.3726% of the file and 52.1215% of records carrying a variation. The rows include facilities other than airports.
• MINOR M2: The counting deck is NAV CANADA's Magnetic North vs True North: Vision to 2030, 28 February 2022; the quotation presents five of its six lines without marking an omission.
  https://www.navcanada.ca/en/magnetic-north-vs-true-north.pdf : Fetched with curl on 2026-09-29, exit 0, 2,881,841 bytes, SHA-256 a0440a26a8417671d8c0701fd391c7f08d26844a37e4602f28d62f846bafb816. Read title slide and slide 26 with pdftotext: Magnetic to True North, Change by 2030, February 28, 2022. Slide 26 also says "25732 will be renumbered over time". All six lines are now included, in column order.
• WRONG M3: Manchester renumbered 06 to 05 in 2009, giving a 7.6-year model lead and a 6.7-year median over six events.
  https://www.earthlearningidea.com/PDF/448_Magnetic_Manchester.pdf; https://www.manchesterairport.co.uk/about-us/manchester-airport-and-mag/history/; research/the-north-the-paint-remembers/events.mjs : Re-fetched both URLs with curl, exit 0. The PDF is 533,372 bytes, SHA-256 f1456e93b6ad969afab24b31dae853fa8528870a5d7398a09da164e37f5aa922. Its text identifies the notice as Stockport Times West, 31 May 2007; I read the original embedded JPEG stream (object 47, extracted without rendering or re-encoding): "our runways are to be renamed from 7th June 2007". The airport history independently puts the change in 2007. That day is 2007.430137, or 2007.43 at the script's precision. Regeneration gives a 2001.9 crossing, 5.53-year lead (5.5 displayed), and six-event median 5.615 (5.6 displayed).
• MINOR M4: Cork's 164 and 344 degree headings were given by the regulator's account.
  https://www.corkairport.com/news/detail/2018/04/27/cork-airport-completes-runway-re-designation; https://www.flightsafetyaustralia.com/2018/05/poles-apart-airport-has-to-rename-runway/ : Re-fetched both with curl, exit 0 (165,323 and 162,000 bytes). Cork Airport's release says "The runway now stands at 164°M and 344°M respectively." Flight Safety Australia prints the same figures and identifies itself as CASA's magazine, with copyright held by Australia's Civil Aviation Safety Authority. This is an attribution clarification: the linked account is an Australian regulator's magazine, and the page now explicitly attributes the headings to Cork Airport's own release. The evidence does not show that CASA never published them.
• MINOR M5: Cork's redesignation took place on 25 April 2018.
  https://www.corkairport.com/news/detail/2018/04/27/cork-airport-completes-runway-re-designation; research/the-north-the-paint-remembers/events.mjs : The re-read operator release specifies Thursday, 26 April. CASA's magazine says 25 April; the operator's account is used. The corrected day converts to 2018.315068, rounded to 2018.32, for both the event and the contemporaneous bearing comparison. The event model is recomputed from that value.
• MINOR M6: The four true-only survivors at high declination are the polar exception showing through.
  public/strata/the-north-the-paint-remembers/data.json; research/the-north-the-paint-remembers/data/runways.json : Re-read twoRules.polarExamples and matched the runway rows: CJQ6 at 81.409451 N, CYAB at 73.006390 N, YCSK at 66.287924 S in Antarctica, and FAGR at 32.190949 S, 24.542401 E in South Africa. The last is not a polar site; the page now names that exception to its description.
• MINOR M7: Two of the eleven injected faults were slips actually made here.
  social/devto/0008-five-keys-agreed.md:45; research/the-north-the-paint-remembers/otherness.mjs:159 : The author's account says one was made and another noticed and avoided. The rotation-sign fault is the one explicitly labelled as having happened. This correction rests on the affirmative account, not on missing logs or an attempt to rerun a historical observation.
• WRONG M8: Moving the benchmark result from 21.08% to 23.13% moves it closer to 21.98%.
  public/strata/the-north-the-paint-remembers/data.json : Re-read otherness.pipeline.faults and recomputed both absolute differences: 0.90 and 1.15 percentage points. The move is 0.25 points farther away. Both the page and the layer README now say further away.
• MINOR M9: The records agree with the model to a median of 0.109 degrees.
  research/the-north-the-paint-remembers/faa.mjs; research/the-north-the-paint-remembers/sources/nasr_apt_csv_2026-08-06/APT_BASE.csv; public/strata/the-north-the-paint-remembers/data.json : Recreated the faa.mjs filter, field calculation and quantile in a Node heredoc, without writing data: n = 13,806, signed median record minus model = -0.10854391457472268 degrees; median absolute residual = 0.2657919387126455 degrees. The fix keeps the stored signed statistic, -0.109, and labels it explicitly. It does not substitute the absolute statistic.
• MINOR M10: The median drift of 2.87 degrees is within a tenth of the three-degree threshold.
  public/strata/the-north-the-paint-remembers/data.json; research/the-north-the-paint-remembers/faa.mjs : The stored signed median is 2.873; 3 - 2.873 = 0.127. The independent recomputation gives 2.872785764726064, or 0.12721423527393583 short. The page says about 0.13 degrees short; the film source derives its 0.127 placeholder from the stored statistic. The median absolute drift is a different statistic, 2.903937026079076, and is not silently substituted.
• MINOR M11: Nothing on this page is a stored answer.
  public/strata/the-north-the-paint-remembers/index.html; research/the-north-the-paint-remembers/make-page.mjs : Read the fetch, fillNumbers, fillTables, instrument and map code. The slider and map run field arithmetic; the remaining figures and tables are filled from the stored data.json assembled by make-page.mjs. The sentence now names the slider and map.
• MINOR M12: Every multiple of five and its surrounding half-degree are legally ambiguous.
  research/the-north-the-paint-remembers/sources/quotations.md:43; public/strata/the-north-the-paint-remembers/index.html; research/the-north-the-paint-remembers/build.mjs : Re-read the FAA sentence transcribed at Q2 and quoted on the page: its case is an azimuth ending in 5, exemplified by 185 degrees and designators 18 or 19. Re-read isTie: abs((bearing mod 10) - 5) < 0.5, which does not include 180. The FAA allows the two adjacent designators at the exact boundary; this page adds width for bearing uncertainty. The full FAA PDF fetch was stopped by curl's 4,000,000-byte cap before downloading a body; the local transcription is the evidence used here.
• MINOR M13: NASR publishes a variation of record and its year for every airport.
  research/the-north-the-paint-remembers/sources/nasr_apt_csv_2026-08-06/APT_BASE.csv; research/the-north-the-paint-remembers/faa.mjs : The independent CSV recount gives 19,426 records, 13,929 with a variation and 13,806 with both variation and a valid year. The page now describes published fields and states the number with both; it no longer says every record is complete.
• MINOR M14: At small declination the magnetic and true rules split the runways almost evenly, which is what coincidence looks like.
  public/strata/the-north-the-paint-remembers/data.json : Re-read twoRules.bands[0]: 540 magnetic-only, 362 true-only, 18 neither. 540 / 362 = 1.4917, about three to two among the first two groups. The page states that split without the unsupported coincidence gloss.
• MINOR M15: Counting all excluded runways wrong gives 29.26%.
  research/the-north-the-paint-remembers/build.mjs; research/the-north-the-paint-remembers/otherness.mjs; public/strata/the-north-the-paint-remembers/data.json : Re-read the scoring code and data: strict rounding counts 3,334 wrong, or 525 more than 2,809, out of 11,393. It does not count all 1,155 excluded runways wrong. That alternative would be 3,964 / 11,393 = 34.793294%, recomputed. The implemented figure stays 29.26% and its description becomes strict rounding, including the generated pipeline label and film source.
• UNVERIFIABLE M16 (unclear in adjudication): All twelve IGRF-14 reference cases were checked by their authors against Fortran and a third implementation.
  research/the-north-the-paint-remembers/sources/pyIGRF14/tests/tests_igrf14.py:29 : Re-read the docstring and parametrized cases. There are twelve cases, but the docstring says nine tests and describes the independent comparison. That mismatch does not establish which later cases the authors did or did not check. Left unchanged under the instruction not to fix unclear rows.
• CONFIRMED M17 (unclear in adjudication): Cork's table crossing is 2009.5 and its film-model crossing is 2009.420.
  research/the-north-the-paint-remembers/events.mjs; research/the-north-the-paint-remembers/film/derive.mjs; research/the-north-the-paint-remembers/film/figures.json : Read both algorithms and recomputed the comparison: the 0.05-year grid records 2009.45, displayed by toFixed(1) as 2009.5; the bisected value is 2009.420202862661. Their difference, 0.029797137339073743 years, is inside the declared grid step. The film derivation explicitly documents that distinction. No change.
• MINOR M18: The verifier re-derives every number printed on the page.
  verify-the-north-the-paint-remembers.mjs; public/strata/the-north-the-paint-remembers/index.html : Re-read the verifier's spread assertion and tested it on an in-memory page copy: replacing only "a spread of 4.6 points" with "a spread of 5.6 points" still passed html.includes('4.6') because 24.66 remained. That predicate has been replaced by sentence checks for the strict-rounding percentage, the right-count percentage and spread, and the ratio of NAV CANADA's cited counts. The prose now limits coverage to the main computed results; it does not claim that outside counts are independently derived.

What was done

• [fixed] Found by the codex fix-checker (M12): the instrument's generated verdict for the half-degree band still said "On the boundary: either number is legal at this date" (e.g. KTPA 10/28 in 1992, bearing 94.53, not on the 95 boundary); it now says "Within the half-degree uncertainty band: excluded from scoring."
• [fixed] Found by the codex fix-checker (M1): the film source's chart label still said airports; it now says landing facilities. The root verifier's comment on the excluded band (M15) now says strict rounding.
• [open note:6fb2de] The shipped film (made 2026-08-23) still shows corrected figures (M1, M3, M4, M5, M9, M10, M12, M15). The directing instance re-rendered it with film/build.sh; the new encode failed one of the film's own gates (loudest beat 1992, outside the published 1993 to 2009), so it was not shipped, and the page now lists what the film still gets wrong.
• [fixed] M1: Corrected the population in the SEO description, dek, HTML prose and chart labels, README, verifier description and film caption source. The source data counts stay unchanged. The shipped film remains unregenerated.
• [fixed] M2: Corrected the deck title in authored HTML, build.mjs, the source manifest and quotation record; included the sixth quoted line. Rebuilt summary.json, public data.json and film figures.json from their scripts.
• [fixed] M3: Changed the Manchester event source and date in events.mjs, regenerated all dependent JSON, and corrected the README row. The page computes the 5.5-year lead and 5.6-year median from the regenerated data. The rendered film is not fixed.
• [fixed] M4: Named Cork Airport's own account in the page, README, event source, film caption and film-data comment, and verifier description. Kept the CASA quotation and its attribution intact. The rendered film is not fixed.
• [fixed] M5: Set Cork's event and bearing-comparison epoch to 2018.32 and its displayed day to 26 April, corrected the README and source commentary, and regenerated the event outputs and film figures. The rendered film is not fixed.
• [fixed] M6: Replaced the blanket polar explanation with three polar sites and the named South African case; no data changed.
• [fixed] M7: Changed two slips made here to one in the page and otherness.mjs, following the author's affirmative record.
• [fixed] M8: Changed closer to further away in the page and layer README, using the two recomputed benchmark gaps.
• [fixed] M9: Made the signed -0.109 statistic explicit in the dek, page, README and verifier description; removed Math.abs from the film placeholder derivation and clarified its caption. The rendered film is not fixed.
• [fixed] M10: Changed within a tenth to about 0.13 degrees short in the dek, page and README. The film source now uses a derived 0.127-degree gap; the rendered film is not fixed.
• [fixed] M11: Scoped the live-arithmetic claim to the slider and map.
• [fixed] M12: Scoped the FAA rule to azimuths ending in five and separated that exact boundary from the page's uncertainty band in HTML, build comments and film captions. The rendered film is not fixed.
• [fixed] M13: Replaced every-airport completeness language with fields that are not always filled, and specified the 13,806 complete records in HTML. Updated the corresponding faa.mjs comment.
• [fixed] M14: Replaced almost evenly and the coincidence gloss with about three to two, magnetic to true.
• [fixed] M15: Renamed the implemented alternative strict rounding in HTML, README, otherness.mjs, its generated data label, film captions and verifier description. The numbers and scoring algorithm are unchanged. The rendered film is not fixed.
• [fixed] M18: Narrowed the page, markdown, generator comment and verifier header to the main computed results, and replaced the loose spread predicate with sentence assertions. The published check mirrors have not been refreshed by their shared generator.