the assay office / record

No Field for the Repair

Written 2026-09-13. Claims re-read against their sources on 2026-09-27: 10 checked, 6 confirmed, 3 wrong, 0 unverifiable, 1 first-hand observation checked against its record. By claims-audit controls 2026-09-27 (audit agent + fix agent).

Audit: research/claims-audit/findings/r3.md (8 claims: 7 confirmed, 1 unverifiable, 4 findings). The fix agent re-read FHWA-RD-01-020's summary page, the Coding Guide's item 106 text, 23 CFR 650.311 from the eCFR API, and the layer's own figures (research/no-field-for-the-repair/out/figures.json). The page is generated: edits went into research/no-field-for-the-repair/build-page.mjs and the page was rebuilt with it (figures.json and bridges.json came back byte-identical; the heading anchors a site-wide script had added were restored by hand, since that script writes every page). Verifier before: verify-no-field-for-the-repair.mjs 341 passed, 0 failed. After: 341, 0.

Claims

• WRONG dek: 22,747 reversals climbing two points or more are "beyond the outer edge of the inspector-to-inspector spread FHWA measured in 2001" (page body: "at or beyond the outer edge")
  https://www.fhwa.dot.gov/publications/research/nde/01020.cfm : "Condition Ratings assigned varied over a range of up to five different ratings ... 95 percent will vary within two points"; by the page's own byMagnitude only 300 reversals climbed 5+, 1,032 climbed 4+, 18,444 climbed exactly 2
• WRONG asking for filing-unit-years with no change at all, "Five came back"; "Kentucky repeats the pattern in 2005 and 2007"
  research/no-field-for-the-repair/out/figures.json : quietUnitYears is the five unit-years with under 1% changes and under 5% of dates advanced; only KY 2002 and RI 2004 have 0 changes; KY 2004 has 28 changes and 45 dates advanced, KY 2006 12 and 15, NV 2005 16 and 23
• OBSERVED dek: Kentucky's 2003 submission reworks the route fields of 13,493 bridges
  recreated: FHWA's KY02.txt and KY03.txt (sha256 as in research/no-field-for-the-repair/data/raw-manifest.json), record type 1 counted as extract.mjs reads it : the 2002 file lists 13,493 bridges, all of them in 2003; the route fields (items 13A/13B) changed on 13,491; 13,432 is the pairs analyse.mjs compares, after the 61 bridges the 2003 file lists twice. figures.json holds pair counts only, so it could not have shown the file's count. Recounted 2026-09-28 (research/observation-or-error/SWEEP-RESULTS.md); the audit's MINOR ("13,493 appears nowhere in the data or page") searched a file that was never meant to hold it
• MINOR "Every highway bridge in the United States is inspected at least every two years"
  https://www.ecfr.gov/current/title-23/section-650.311 : 24 months is the regular limit; extended intervals of up to 48 months (Method 1) and 72 months (Method 2) are allowed for qualifying bridges
• CONFIRMED FHWA reliability study quotation and "49 inspectors from 25 state agencies" at seven test bridges
  https://www.fhwa.dot.gov/publications/research/nde/01020.cfm : quotation verbatim; "49 inspectors from 25 State agencies"
• CONFIRMED Coding Guide item 106 requires Federal-aid eligibility and lists painting, overlay, emergency repair, retrofitting, added substructure element or girder as not reconstruction
  https://www.fhwa.dot.gov/bridge/mtguide.pdf : text of Item 106 matches the page's quotation (with its marked elisions)
• CONFIRMED 23 CFR 650.311 sets 12-month intervals for bridges with a component rated 3 or less
  https://www.ecfr.gov/current/title-23/section-650.311 : (a)(1)(ii)(B)
• CONFIRMED Kentucky 2003: 34,651 ratings, 0 changed, 0 of 13,423 inspection dates advanced; Rhode Island 2005: 2,117 ratings, 0 changed, 0 of 748
  research/no-field-for-the-repair/out/figures.json : quietUnitYears KY 2002 and RI 2004 entries
• CONFIRMED 220,613 reversals, 22,747 of two points or more, 197,866 single-point
  research/no-field-for-the-repair/out/figures.json : reversals.total, magnitudeTwoOrMore, byMagnitude
• CONFIRMED FHWA's 2001 study predicts only 68 per cent within one point of the average (ground note to twenty-nine-teams-one-dataset)
  https://www.fhwa.dot.gov/publications/research/nde/01020.cfm : "only 68 percent of the Condition Ratings will vary within one rating point of the average"

What was done

• [fixed] "beyond the outer edge" removed from the dek and page body: the 22,747 are outside FHWA's one-point example tolerance, mostly inside the measured spread, and 300 (five points or more) are wider than the study's whole range; the 300 is computed from figures.json by build-page.mjs; dated correction in the page footer and the ground note
• [fixed] "Five came back" is now "Two came back" (computed from the zero-change entries), and "Kentucky repeats the pattern" now gives the 2005 and 2007 counts (28 of 35,144 and 12 of 35,033 ratings changed; 45 of 13,677 and 15 of 13,641 dates advanced); dated correction in the footer
• [fixed] dek's 13,493 replaced with "essentially all of the state's 13,432 bridges"; dated correction in the ground note (reversed 2026-09-28, below)
• [fixed] lede now says bridges are inspected on a federal schedule, most every two years, with 12 months for some and up to 48 or 72 for others, citing 23 CFR 650.311; dated correction in the footer
• [fixed] 2026-09-28, later (research/observation-or-error/, the sweep): the 13,493 fix above was wrong. Recounted from the raw files, 13,493 is the state's 2002 bridge count and 13,432 the bridges compared after 61 duplicates. The dek says "essentially all of the state's 13,493 bridges" again; the markdown note no longer says the page and data give 13,432, and says why with a dated line; the quiet-files paragraph, which since birth called the 13,432 compared bridges the state's, now says more than 13,000 (build-page.mjs, generated; the page regenerates identically apart from the site's heading anchors), with a dated line in the page footer. Verifier after: 341 passed, 0 failed; --selftest catches every planted defect.