Add Esperanto, Scottish Gaelic, Guaraní, Hawaiian engines (Beta); park Navajo

.github/workflows/ci.yml

@@ -981,5 +981,51 @@ jobs:
         run: ./scripts/zul/fetch_kaikki.sh
       - name: Zulu golden harness with two-oracle regression gate
         run: cargo run --release --bin golden_zul -- data/zul/unimorph.tsv data/zul/kaikki.tsv --check
+      - name: Cache kaikki epo
+        id: kaikki-epo
+        uses: actions/cache@v5
+        with:
+          path: data/epo/kaikki.tsv
+          key: kaikki-epo-v1
+      - name: Fetch kaikki epo
+        if: steps.kaikki-epo.outputs.cache-hit != 'true'
+        run: ./scripts/epo/fetch_kaikki.sh
+      - name: Esperanto golden harness with single-oracle regression gate (Beta)
+        run: cargo run --release --bin golden_epo -- data/epo/kaikki.tsv --check
+      - name: Cache kaikki gla
+        id: kaikki-gla
+        uses: actions/cache@v5
+        with:
+          path: data/gla/kaikki.tsv
+          key: kaikki-gla-v1
+      - name: Fetch kaikki gla
+        if: steps.kaikki-gla.outputs.cache-hit != 'true'
+        run: ./scripts/gla/fetch_kaikki.sh
+      - name: Scottish Gaelic golden harness with single-oracle regression gate (Beta)
+        run: cargo run --release --bin golden_gla -- data/gla/kaikki.tsv --check
+      - name: Cache kaikki grn
+        id: kaikki-grn
+        uses: actions/cache@v5
+        with:
+          path: data/grn/kaikki-raw.jsonl
+          key: kaikki-grn-v1
+      - name: Fetch kaikki grn
+        if: steps.kaikki-grn.outputs.cache-hit != 'true'
+        run: ./scripts/grn/fetch_kaikki.sh
+      - name: Build grn gold
+        run: python3 scripts/grn/build.py
+      - name: Guaraní golden harness with single-oracle regression gate (Beta)
+        run: cargo run --release --bin golden_grn -- data/grn/kaikki.tsv --check
+      - name: Cache kaikki haw
+        id: kaikki-haw
+        uses: actions/cache@v5
+        with:
+          path: data/haw/kaikki.tsv
+          key: kaikki-haw-v1
+      - name: Fetch kaikki haw
+        if: steps.kaikki-haw.outputs.cache-hit != 'true'
+        run: ./scripts/haw/fetch_kaikki.sh
+      - name: Hawaiian golden harness with single-oracle regression gate (Beta)
+        run: cargo run --release --bin golden_haw -- data/haw/kaikki.tsv --check
       - name: Reverse-lookup gate (fra, spa, eng; deu needs the kaikki dump)
         run: cargo run --release --bin reverse_gate -- --check --langs fra,spa,eng

.gitignore

@@ -9,5 +9,6 @@ Cargo.lock
 __pycache__/
 !/data/*/parts.tsv
 !/data/*/aux.tsv
+!/data/*/overrides.tsv
 .claude/worktrees/
 errors.log

Cargo.toml

@@ -72,6 +72,14 @@ include = [
     "data/ind/overrides.tsv",
     "data/zul/parts.tsv",
     "data/zul/overrides.tsv",
+    "data/epo/parts.tsv",
+    "data/epo/overrides.tsv",
+    "data/gla/verbs.tsv",
+    "data/gla/parts.tsv",
+    "data/grn/parts.tsv",
+    "data/grn/overrides.tsv",
+    "data/haw/parts.tsv",
+    "data/haw/overrides.tsv",
     "data/por/classes.tsv",
     "data/por/verbs.tsv",
     "data/ron/classes.tsv",

data/epo/overrides.tsv

@@ -0,0 +1,2 @@
+# Per-cell overrides: none. Esperanto has no irregular verbs.
+# lemma	features	form

data/epo/parts.tsv

@@ -0,0 +1,5 @@
+# Esperanto is perfectly regular: no irregular verbs, no principal
+# parts to mine. This file exists only so the reverse-index include
+# (src/reverse.rs) and Cargo package include resolve. Column 1 (lemma)
+# is intentionally empty.
+# lemma

data/gla/parts.tsv

@@ -0,0 +1,7 @@
+# Suppletive / irregular overrides that the mined principal parts
+# cannot express. Columns:
+#   lemma  past  future  vn  ptcp  fut-dep  pst-dep  cond-stem
+# "-" = keep the mined/derived value. cond-stem, when set, replaces the
+# lemma as the base for the derived relative future, conditional and
+# imperative forms (thoir → bheir-/thug-).
+thoir	-	-	-	-	-	-	beir

data/grn/overrides.tsv

@@ -0,0 +1,21 @@
+# Guarani per-cell overrides: lemma ⇥ features ⇥ form.
+# Consulted before the productive rules in src/grn.rs.
+# soro (nasoral) forms its coactive causative with the nasal-mutated
+# stem mo+ndoro (s→nd), not the regular mbo+soro. It is the only
+# nasoral verb attested, so this nasal mutation is listed, not ruled.
+soro	V;COACTIVE;HORTATIVE;1PL.EXCL	toromondoro
+soro	V;COACTIVE;HORTATIVE;1PL.INCL	tañamondoro
+soro	V;COACTIVE;HORTATIVE;1SG	tamondoro
+soro	V;COACTIVE;HORTATIVE;2PL	tapemondoro
+soro	V;COACTIVE;HORTATIVE;2SG	teremondoro
+soro	V;COACTIVE;HORTATIVE;3PL	tomondoro
+soro	V;COACTIVE;HORTATIVE;3SG	tomondoro
+soro	V;COACTIVE;IMPERATIVE;2PL	pemondoro
+soro	V;COACTIVE;IMPERATIVE;2SG	emondoro
+soro	V;COACTIVE;INDICATIVE;1PL.EXCL	romondoro
+soro	V;COACTIVE;INDICATIVE;1PL.INCL	ñamondoro
+soro	V;COACTIVE;INDICATIVE;1SG	amondoro
+soro	V;COACTIVE;INDICATIVE;2PL	pemondoro
+soro	V;COACTIVE;INDICATIVE;2SG	remondoro
+soro	V;COACTIVE;INDICATIVE;3PL	omondoro
+soro	V;COACTIVE;INDICATIVE;3SG	omondoro

data/grn/parts.tsv

@@ -0,0 +1,73 @@
+# Guarani principal parts: lemma ⇥ conjugation class.
+# The class is read verbatim from the kaikki gug-conj-* inflection
+# template (see scripts/grn/build.py); the engine in src/grn.rs
+# derives the whole paradigm from it. gn-conj-* is folded into gug-.
+lemma	class
+'a	h
+'u	h
+guata	areal-oral
+gueraha	areal-oral
+g̃uahe	areal-nasal
+hai	areal-oral
+hecha	areal-oral
+hechaga'u	areal-oral
+heka	areal-oral
+hekuavo	areal-oral
+hendu	areal-nasal
+hesape	areal-oral
+hupi	areal-oral
+hupyty	areal-oral
+ime	areal-nasal
+jaho'i	areal-oral
+jajái	areal-oral
+japo	areal-oral
+jehekýi	areal-oral
+jehu	areal-oral
+jepy'amongeta	areal-oral
+jeroky	areal-oral
+jerure	areal-oral
+jogua	areal-oral
+johéi	areal-oral
+jopy	areal-oral
+juka	areal-oral
+jupi	areal-oral
+juvy	areal-oral
+karu	areal-oral
+kañy	areal-nasal
+ke	aireal-oral
+kuaa	aireal-oral
+kuave'ẽ	areal-oral
+kytĩ	aireal-nasal
+ma'ẽ	areal-nasal
+mano	areal-nasal
+mbo'e	areal-nasal
+mbohovái	areal-nasal
+mbohéra	areal-nasal
+mboja'o	areal-nasal
+mbopiro'y	areal-nasal
+mbopu	areal-nasal
+mboty	areal-nasal
+mbotyryry	areal-nasal
+mbyepoti	areal-nasal
+monde	areal-nasal
+muña	areal-nasal
+myenyhẽ	areal-nasal
+pe'a	aireal-oral
+poi	areal-oral
+porandu	areal-oral
+puka	areal-oral
+purahéi	areal-oral
+puru	aireal-oral
+pytyvõ	areal-oral
+rambosa	areal-nasal
+reko	areal-oral
+soro	areal-nasoral
+sãmbyhy	areal-nasal
+timbo	areal-nasal
+veve	areal-oral
+y'u	h
+ñani	areal-nasal
+ñapytĩ	areal-nasal
+ñe'ẽ	areal-nasal
+ñembo'y	areal-nasal
+ñembosarái	areal-nasal

data/haw/overrides.tsv

@@ -0,0 +1,24 @@
+# Hawaiian derivational overrides: lemma ⇥ feature ⇥ form.
+# Lexicalised forms the productive rules in src/haw.rs do not
+# predict — causative+reduplication fusions, prefix-triggered
+# vowel lengthening, ʻokina/long-vowel edge cases. Mined by
+# scripts/haw/mine_overrides.py from the kaikki oracle.
+hele	V;CAUS	hoʻohelehele
+hoʻi	V;CAUS	hoʻihoʻi
+hoʻokolo	V;CAUS	hoʻokolohua
+hoʻokolo	V;CAUS	hoʻokolokolo
+hoʻoluli	V;CAUS	hoʻoluliluli
+kāhili	V;CAUS	hoʻokahili
+lōʻihi	V;CAUS	hoʻoloʻihi
+mahele	V;CAUS	hoʻomāhele
+malū	V;CAUS	hoʻomālū
+pio	V;CAUS	hoʻopiopio
+pono	V;CAUS	hoʻoponopono
+ā	V;CAUS	hōʻā
+ʻaʻā	V;CAUS	hoʻaʻā
+ʻāhewa	V;CAUS	hoʻāhewa
+ʻēʻē	V;CAUS	hoʻēʻē
+ʻō	V;CAUS	hoʻō
+ʻōiwi	V;CAUS	hoʻōiwi
+ʻōlapa	V;CAUS	hoʻōlapa
+ʻōnaha	V;CAUS	hoʻōnaha