Teachable Decoding Inventory (data/)

Open, fact-based data that the engine uses to resolve cases the pure-grapheme rules leave underdetermined — the small, teachable inventories that structured literacy relies on (base morphemes, origin markers, core GPCs).

License: CC BY 4.0 — any school, curriculum, or product may use, adapt, and redistribute this, with attribution. Credits: see CREDITS.md.

Principles for anything added here

1. Facts only. Every entry is a fact about English orthography (a base morpheme, a grapheme's sound, an origin marker), not a word-by-word answer key and not a copy of any curriculum's curated list.
2. Finite and teachable. These are the bounded inventories a child is actually taught (a few dozen affixes, a handful of origin markers, ~100 base morphemes to start) — the one principled way past the from-letters residue (e.g. farm is a base, so farmer = farm+er, but corn here is not treated as a productive base of corner).
3. Attributed at the row level. Each file carries a _meta block and, where a specific finding backs an entry, a source_notes field citing it.
4. Not sound-from-word. These inventories condition sound/structure on letters and finite teachable sets, never on word frequency or a whole-word pronunciation lookup. (See docs/g2p-research.md §0.)

Files

File	Purpose	Stability
en/gpc-table.json	Core grapheme→phoneme correspondences, ordered by frequency, with positional context	seed
en/gpc-mined.json	Empirical grapheme→phoneme frequencies mined from CMUdict. Regenerate: node src/tests/G2PMiner.js	mined
en/origin-markers.json	Greek/French orthographic markers that disambiguate origin-sensitive graphemes (ch)	seed
en/base-morphemes.json	Prefixes, suffixes, bases and the opaque list — the morpheme-boundary spine	seed
en/articulation.json	Place/manner/voicing, plus how a child FEELS each sound — the multisensory layer	seed
en/base-verbs.json	Verb membership and complement frames (valency)	seed
en/irregular-verbs.json	Principal parts of the strong/irregular verbs	seed
en/conjunctions.json	Coordinators and subordinators	seed
en/definitional-verbs.json	Naming predicates — "referred to as", "hereinafter"	seed
en/standards-crosswalk.json	Rule id → curriculum code. Codes only, never text	seed
en/mined-verbs.json	Wide verb membership mined from Moby, cross-checked against CMUdict	unstable
en/semantic-dimensions.json	Semantic axes — part design sketch, part data	unstable
shared/combining-forms.json	Neoclassical Greek/Latin combining forms	seed

What the stability column promises

seed — deliberately small and correct, meant to grow. The SCHEMA is settled; expect entries to be added, rarely removed. Safe to build against.

mined — regenerated by code from a permissive corpus. Content moves when the corpus or the mining rules do; the shape does not.

unstable — use it, that is what CC BY is for, but do not assume it holds still. Content churns heavily and the schema may change without notice: fields renamed, nesting altered, entries removed. No deprecation period and no migration note is guaranteed. Pin a commit if you depend on the shape. Each such file repeats this in its own _meta.stability, with the reason it is unstable, so a consumer reading only the JSON still finds out.

Nothing here is versioned yet. Until it is, a commit hash is the only stable reference, and that applies to every tier — the tiers describe intent, not a guarantee anyone can enforce.