BTS-NUM · Data Standards
BTS-NUM100: Bantu Numeracy Data Standard
The English-anchored data contract for Bantu numeracy, including the calendar. Every numeral, weekday, month, and date a language provides is a governed primitive that resolves to exactly one English reference id and carries a provenance tier.
Draft status.
Draft at the primitive-tracking stage. The contract, schemas, and addressing may change until v1.0.0. Conformance is defined but not assigned: every language's level is currently null, and no language may claim one.
The numeracy data contract.
BTS-NUM100 is the data half of Bantu numeracy. It names the English references every language is asked to answer, and it governs the concrete per-language primitives that answer them: cardinals, ordinals, operators, counted nouns, weekdays, months, and numeral-derived dates. Those primitives drive the calculator, calendar, tutor, benchmark, and corpus surfaces.
Example: (REF-00006, bem) is Bemba's answer to the English reference for five.
Peers, not layers.
| BTS-E100 | BTS-NUM100 | |
|---|---|---|
| Defines | The Bantu equations, the structure | The reference and the primitives, the data |
| Nature | Abstract, Bantu-native, language-agnostic | Concrete, English-anchored, per-language |
| Keyed by | Equation IDs, E100-2xx | English REF-IDs and primitive ids |
Peer of BTS-E100, not a superset
BTS-E100 defines the abstract Bantu equations; BTS-NUM100 defines the concrete, per-language data that instantiates them. A cartridge field satisfies an E100 slot through a shared cross-reference table; NUM100 never renumbers or absorbs E100.
English pivot
The reference contract is a frozen, checksummed registry of 1,415 rows, each a permanent REF-ID. Every translation-bearing datum resolves to exactly one REF-ID, which yields aligned parallel corpora across languages for free.
Primitive model
Every functional datum is a primitive with a deterministic UUIDv5 id, a kind, a status, a version, and a provenance. Ids are reproducible, so synchronisation is idempotent and removal is deprecate-not-delete.
Calendar is first-class
Weekday names, month names, and numeral-derived full dates are governed here, not in BTS-E100, because they sit on the numeral core: day-of-month reuses the cardinal primitives and numeral weekdays are root-verified against them.
Provenance is ranked
Translation forms are tiered native-attested, composed-verified, then composed. A native speaker who consults a dictionary, spellcheck, or an elder is still native-attested; provenance is never downgraded for using an aid.
Round-trip invariant
For each language, unflattening the active primitives of a kind must reproduce the on-disk file exactly, for engine rules, calendar, counted nouns, and the reference contract layer.
The English reference contract.
| Category | Rows | Example |
|---|---|---|
| cardinal | 137 | REF-00006 = 5 / five |
| ordinal | 17 | 1st / first |
| operator/sum | 60 | 1 + 1 = 2 |
| counted-noun | 1,006 | six people |
| weekday | 7 | Monday |
| month | 12 | January |
| other | 176 | symbols and operator words, Plus = + |
| Total | 1,415 |
Six kinds of primitive.
| Kind | Is |
|---|---|
| english-reference | One row of the English reference contract. |
| rule | One leaf of a cartridge tree, the engine ruleset. |
| attested-reference | One language's answer to one REF-ID, the translation layer. |
| variant | A kept alternate form. |
| calendar | One leaf of a language's calendar file: weekday, month, or date. |
| counted-noun | One leaf of a language's concord view. |
Conformance ladder, defined but not yet assigned.
| Level | Name | Means, when grading is enabled |
|---|---|---|
| L0 | Schema-valid | The cartridge schema validates with zero issues. |
| L1 | Numeral core | 137 attested cardinals with full attested-lookup fidelity. |
| L2 | Reference-complete | All 1,415 contract rows captured, reference alignment clean. |
| L3 | Capabilities | Calendar present, 7 weekdays and 12 months, numeral days root-verified, date composition, plus counted-noun concord. |
| L4 | Governed | Exact round-trip, every translation primitive reference-anchored and provenance-tagged. |
| Certified | Adjudicated | External native-speaker adjudication on a sampled set. |
Why this matters.
A system that claims to count from zero to a trillion in a Bantu language is usually claiming three different things at once: what it can render, what a native speaker has actually confirmed, and what a rule composed on the fly. BTS-NUM100 forces those apart. Every numeral carries its provenance tier and its English reference, so a buyer can see exactly which of the three they are licensing.