the assay office / record

After Z Comes Å

Written 2026-08-24. Claims re-read against their sources on 2026-09-27: 11 checked, 7 confirmed, 4 wrong, 0 unverifiable. By claims-audit controls 2026-09-27 (audit agent + fix agent).

Verifier `node research/after-z-comes-a-ring/verify-after-z-comes-a-ring.mjs`: 110 passed, 0 failed before and after; `--mutate` 4/0 before and after. The German fix is a relabel only (the audit's preferred route): the table, fold map, weights and every shelf order are unchanged, so the page's orders are now correct for the convention it names. The internal id `de-phone` is kept so the blind-reconciliation records (battery.json, engine-b.mjs, reconcile-report.json) still match; the page says the id is a leftover name.

Claims

• CONFIRMED Swedish types å, ä, ö after z
  https://en.wikipedia.org/wiki/Alphabetical_order : "three extra vowels placed at its end (..., X, Y, Z, Å, Ä, Ö)"
• WRONG the German table that files umlauts as base vowels is "the telephone-directory treatment", and ä as ae "the dictionary treatment"
  https://en.wikipedia.org/wiki/German_alphabet : base-letter folding (DIN 5007-1) is preferred for dictionaries; ae/oe/ue (called DIN 5007-2 there) for names and telephone directories; de.wikipedia.org/wiki/Alphabetische_Sortierung puts both in DIN 5007-1 as variants 1 (words) and 2 (name lists)
• MINOR Swedish å, ä, ö are "elements 27, 28 and 29, after z at 26"
  public/strata/after-z-comes-a-ring/index.html : the page's own chips and sort-key readout are 0-indexed (z 25, å 26); the ordinals are right only as 1-based counts
• CONFIRMED Welsh ch dd ff ng ll ph rh th are single letters; order a b c ch d dd e f ff g ng h i j l ll m n o p ph r rh s t th u w y
  https://en.wikipedia.org/wiki/Welsh_orthography : same 29-letter list; la < ly < lla < ma
• WRONG each Welsh digraph ranks "just after its first component"
  https://en.wikipedia.org/wiki/Welsh_orthography : ng follows g, not n (the typed table always had it after g; only the prose was wrong)
• CONFIRMED ß files as ss
  https://en.wikipedia.org/wiki/German_alphabet : "Eszett is sorted as though it were ss"
• CONFIRMED letter-by-letter filing ignores spaces; word-by-word files a whole first word first
  https://en.wikipedia.org/wiki/Alphabetical_order : Oak precedes Oakley word by word; letter by letter ignores spaces
• CONFIRMED Spanish stopped treating ch and ll as separate letters in 1994
  https://en.wikipedia.org/wiki/Alphabetical_order : RAE alphabetization rule of 1994; wording tightened to "alphabetizing" to match (they left the alphabet itself only in 2010)
• MINOR Dutch ij as a single letter is historical practice
  https://en.wikipedia.org/wiki/Alphabetical_order : formerly collated as y, now mostly two letters, "Exceptions are phone directories; IJ is always collated as Y here"
• CONFIRMED whether j is a full Welsh letter was historically contested
  https://en.wikipedia.org/wiki/Welsh_orthography : j "has only recently entered", formally introduced 1987
• CONFIRMED for these names UTF-8 byte order, UTF-16 code-unit order and code-point order coincide
  public/strata/after-z-comes-a-ring/index.html : every character in the aw-tables names is below U+0100, where all three orders agree

What was done

• [fixed] HIGH, German conventions swapped: relabelled the base-vowel-folding table "German dictionary" (DIN 5007-1 variant 1) in the dek, plain, seoDescription, meta/og/twitter/JSON-LD descriptions, body, check panel, verdict sentences, engine.mjs label, verifier check names and README; the ae treatment is now named as the name-list/telephone-directory convention; the "source needed" DIN note replaced with the two Wikipedia sources; dated Corrected line on the page
• [fixed] MINOR, Swedish element numbering: prose now gives the 27th to 29th elements with weights 26 to 28 and z at weight 25, matching the displayed 0-indexed table; dated Corrected line
• [fixed] ng placement (found by the fix agent): prose, engine.mjs and README now say each digraph follows its first component except ng, which follows g; dated Corrected line
• [fixed] Dutch ij (found by the fix agent): now "mostly historical practice, surviving in telephone directories"; dated Corrected line

Second pass, blind to the first

This layer was drawn, one in ten, for the office's audit of itself (the results, and the pre-registration). A second reader, not shown the record above, re-read it on 2026-09-28: 13 claims checked, 1 wrong. Against the first pass: 1 missed by it, 0 introduced by its own fixes, 0 of its fixes disputed, 0 added since by something else. By claude-funny-gauss-cgnurf, one agent on oversight/claims-pass.md blind to the first record, adjudicated by hand.

• [R] the Welsh j: the first record lists "whether j is a full Welsh letter was historically contested" as CONFIRMED against the same Wikipedia article, whose sentence ends "but is still debated"; the text was on the page at the first pass (it predates the 2026-09-27 fix commits and none touched it). Minor in weight, but a verdict disagreement on a claim both passes read

Claims

• CONFIRMED Corrected 2026-09-27: base-vowel folding is the dictionary convention (DIN 5007 variant 1), ä as ae the name-list/telephone one
  https://de.wikipedia.org/wiki/Alphabetische_Sortierung : "Variante 1 (für Wörter verwendet, etwa in Lexika)"; "Variante 2 (spezielle Sortierung für Namenslisten, etwa in Telefonbüchern)"
• CONFIRMED Corrected 2026-09-27: ng follows g in the Welsh alphabet
  https://en.wikipedia.org/wiki/Welsh_orthography : "a b c ch d dd e f ff g ng h i j l ll m n o p ph r rh s t th u w y"
• CONFIRMED Corrected 2026-09-27: å ä ö are the 27th to 29th Swedish elements, weights 26 to 28
  https://en.wikipedia.org/wiki/Swedish_alphabet : "sorted after z"; engine table z=25, å=26, ä=27, ö=28
• CONFIRMED Corrected 2026-09-27: Dutch ij as y survives in phone directories
  https://en.wikipedia.org/wiki/Alphabetical_order : "Exceptions are phone directories; IJ is always collated as Y here"
• CONFIRMED Swedish å, ä, ö come after z
  https://en.wikipedia.org/wiki/Swedish_alphabet : "sorted after z"
• CONFIRMED Welsh digraphs are single letters, 29 letters, lori before llaeth
  https://en.wikipedia.org/wiki/Welsh_orthography : alphabet table
• CONFIRMED Welsh has no k, q, v, x, z
  https://en.wikipedia.org/wiki/Welsh_orthography : "The letters ⟨k, q, v, x, z⟩ are not part of the Welsh Alphabet"
• CONFIRMED ß files as ss in the dictionary convention
  https://de.wikipedia.org/wiki/Alphabetische_Sortierung : "ß und ss sind (zunächst) gleich"
• CONFIRMED Spanish stopped alphabetizing ch and ll separately in 1994
  https://en.wikipedia.org/wiki/Ch_(digraph) : April 1994 vote of the Association of Spanish Language Academies
• CONFIRMED letter-by-letter vs word-by-word filing
  https://en.wikipedia.org/wiki/Alphabetical_order : as described
• CONFIRMED a default Unicode collation table (DUCET) exists
  https://en.wikipedia.org/wiki/Unicode_collation_algorithm : "Default Unicode Collation Element Table (DUCET)"
• CONFIRMED for these names UTF-8, UTF-16 and code-point order coincide; the byte column is a plain code-unit sort
  research/after-z-comes-a-ring/verify-after-z-comes-a-ring.mjs : recomputed; all characters below U+0100
• WRONG whether j counts as a full Welsh letter "was historically contested"
  https://en.wikipedia.org/wiki/Welsh_orthography : "The letter ⟨j⟩ has only recently entered into Welsh orthography, but is still debated"; the page put an open question in the past tense

What was done

• [fixed] the Welsh j sentence now reads "is still debated (it was added in 1987)", with a dated Corrected 2026-09-28 line quoting the source; verifier 110 passed before and after
• [open note:9b802e] the page's "Run it yourself" line cannot be followed by a stranger: the published check needs engine.mjs, which the page never names, and against the live bytes it fails one assertion on the edge-injected Cloudflare beacon (the site-wide fault in research/claims-audit/RESULTS.md); the first pass did not report it either, and it is a reader-side fault, not a false claim, so it is not counted below
• [declined] MINOR, reported: the verifier never reads the prose, so sentence numbers are checked only by hand; the page does not claim otherwise and every sentence number was redone and holds