Skip to content

feat(WO20): exhaustive per-row tests for six thin ALA-LC small-table maps (coverage depth, batch 3) - #289

Merged
ronaldtse merged 1 commit into
mainfrom
wo20-depth-alalc-small
Nov 16, 2026
Merged

ronaldtse merged 1 commit into
mainfrom
wo20-depth-alalc-small

Conversation

@ronaldtse

Copy link
Copy Markdown
Contributor

Summary

WO20 coverage-depth, batch 3 (#283): the six thinnest small-table ALA-LC maps, deepened from the held source PDFs (interscript-private-references/alalc/) via scripts/gen-row-tests.mjs (the #272/#281 pattern). No map rules changed — the tests pin the maps against the printed tables; 31 source-anchored findings recorded (below), never forced green. One audit candidate (alalc-lif) turned out to be not deepenable from the held source and is recorded as a finding instead.

Per-map counts (transcribed = generated + canonically reconciled; every generated test passes)

map transcribed generated (before → after) skipped (why)
alalc-bul-Cyrl-Latn-1997 62 0 → 62 Ъ/ъ second value cells (note-1 word-final context)
alalc-sat-Olck-Latn-2017 32 1 → 32 ᱸ (note-2 mark), ᱻ/ᱼ (omitted marks)
alalc-mni-Mtei-Latn-2022 53 4 → 57 ꯬ heavy tone (mark), ꯭ killer (instruction), punctuation (layout pairing)
alalc-aze-Cyrl-Latn-1997 83 3 → 67 — (19 failing rows are findings)
alalc-ben-Beng-Latn-2017 52 3 → 47 ব duplicate listing (Note 3 va context), ঁ candrabindu (n̐/m̐ context split)
alalc-kat-Geok-Latn-1997 76 5 → 81 ფ/ჷ/ჸ/უ̌ უ̂ Mkhedruli-only tail rows (no Khutsuri key)

alalc-lif-Limb-Latn-2011: not deepenable — the held alalc-limbu.pdf renders no Limbu glyphs at all (the glyph column is blank in both the text layer and the rendered page; verified via rawdict: zero Limbu codepoints on the table page). The printed romanization column is present, but the row KEYS cannot be sourced from the printed page, so per the anti-fabrication rule no rows are generated.

Findings (source-anchored; maps untouched this round)

  • alalc-aze-Cyrl-Latn-1997 — nine letters unencoded or divergent. The map has no rule at all for Ё/ё, Ц/ц, Щ/щ, Ъ/ъ, Ь/ь, Э/э, Ю/ю, Я/я (plain Cyrillic input passes through unchanged); the printed table gives Ë, TS͡/ts͡, Shch, ʺ, ʹ, Ė, IU͡/iu͡, IA͡/ia͡ (tesh digraphs U+0361). Also Й/й: map encodes Y/y, printed Ĭ/ĭ; ј: map emits dotless ı + candrabindu, printed dotted i + candrabindu (i̐).
  • alalc-ben-Beng-Latn-2017 — the map's values diverge from the printed 2017 table in 11 cells and instead match the 1997-era scheme: চ (printed ca, map cha), ছ (cha, chha), ড়/ঢ় (ṛa/ṛha, ṙa/ṙha — dot-above instead of dot-below), য (ya, j̱a), য় (ẏ, ya), শ (śa, sha), ষ (sha, ṣha), ং (ṃ dot-below, ṁ dot-above), ঽ (’ — unencoded), ই (printed capital I, map i).
  • alalc-sat-Olck-Latn-2017 — the map lacks the ᱱ → n rule (the table's last consonant); plain ᱱ passes through unchanged.
  • Version provenance: for bul (map 1997) and aze-Cyrl (map 1997) the only locally-held printed sources are the 2013/2017 revisions; the letter tables match the maps' encodings (except the findings above), and the version difference is recorded in each artifact.
  • Canonical-equivalence reconciliations (not findings): where the runtime emits a canonically equivalent byte form of the printed glyph (bul Й/й breve, aze Ө/ө dot-above), the row is stored in the runtime's byte form so the byte-exact embedded tests pass; documented per artifact.

Audit delta (this batch)

alalc-lif moves from "later batch" to not-deepenable (finding) — its held source carries no script glyphs.

Gates

  • iscrs test-all --maps maps (branch corpus): green, 0 failed (the six maps: 343 tests)
  • ts parser: 435/435 .isc files parse clean
  • Golden recapture delta vs interscript-rs tests/golden/isc-wf0.jsonl: 0 lost, 0 changed (additive; total uncaptured delta to be recaptured in the single rs golden PR after all batches)
  • Every generated row verified end-to-end through iscrs transliterate before being written into the artifact; skipped rows + findings recorded in each scripts/row-sources/*.json note

Test plan

  • iscrs test-all green on the branch corpus
  • golden additivity: 0 lost / 0 changed rows
  • ts parse gate green (435/435)
  • findings + skips recorded per artifact note

…maps (coverage depth)

Rows transcribed from the held ALA-LC PDFs (bulgarian 2013, santali 2017,
manipuri 2022, azerbaijani 2017 Cyrillic section, bengali 2017, georgian
2011 Khutsuri) into row-source provenance artifacts, generated into
embedded tests with scripts/gen-row-tests.mjs. No map rules changed.

358 rows transcribed; 327 generated, all passing end-to-end; 31
source-anchored findings recorded (aze-Cyrl: nine letters unencoded or
divergent; ben-2017: eleven cells diverge from the printed 2017 table;
sat: the ᱱ -> n rule is missing). alalc-lif recorded as not deepenable:
the held PDF carries no Limbu glyphs.
@ronaldtse
ronaldtse merged commit 6b956e1 into main Nov 16, 2026
13 checks passed
ronaldtse added a commit that referenced this pull request Nov 16, 2026
Three batches from held sources (maps #287/#288/#289): eight odni
ICS-630 maps, eight thinnest DIN 1460-2 maps, six small ALA-LC tables
- 906 exhaustive per-row tests, all verified end-to-end; 48
source-anchored findings recorded for follow-up fixes. Golden
refreshed additively (rs #30): 13,745 rows, zero lost/changed.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant