3.1 KiB
Session Log Addendum: 2026-05-12 — Language Pack Design
Project: AssetOverflow/core
Session type: Architecture design addendum
Related ADR: ADR-0005
21:02 — Language Pack Direction Confirmed
What happened: The decision was made to move from general discussion of language integration toward a formal, meticulous language pack contract for CORE.
Why: The foundational languages are not optional output skins. They are part of the native articulation and semantic design of the system. Therefore the next clean artifact is a strict contract and scaffold, not an informal note.
Decision:
Create ADR-0005 and establish the packs/ structure with shared contracts,
schemas, trilingual anchor templates, and initial en, he, and el pack
scaffolds.
Design Notes Recorded
Language choice rationale
The three languages were chosen because of the range of depth they bring:
- English is the default base (any language could serve this role in a custom CORE instance, but English is the default).
- Hebrew and Koine Greek are the depth languages. John 1:1–2: the universe itself was spoken into existence, and John articulated it in both Hebrew thought and Greek expression — almost certainly a nod from the Holy Spirit. The hidden layer of intelligence in CORE's vocabulary manifold is grounded in this. This is why and how CORE finds its truth and power in its design and communication — being primary, being a core.
Pack design decisions
- Packs are lemma-first, not token-first.
- Morphology is native to the pack contract, especially for Hebrew and Koine Greek where morphology is semantically load-bearing.
- Readback belongs to the pack, not to a global postprocessor.
- Semantic lift must target shared field primitives.
- Pack activation is gate-based, not file-existence-based.
- Cross-language alignment is an explicit probe surface, not an emergent hope.
- Runtime boundary files raise
NotImplementedErrorat unimplemented semantic edges — honest structure, not fake behavior.
LLM as ingest extraction: rejected
Using a general-purpose LLM as the extraction engine for large document ingest was considered and explicitly rejected. The reason: it introduces a D3 nondeterministic oracle at the normalization site, and its interpretations are silently embedded in the field state without provenance. Semantic Rigor forbids this. The replacement is a deterministic, structure-aware segmenter (StructuralSegmenter) that carves at form boundaries — document structure, verse markers, code delimiters — not at interpreted content boundaries. For Hebrew and Koine Greek this is even cleaner: canonical verse and pericope boundaries have been fixed for centuries and constitute D0 segmentation.
Build Order Established
| Batch | Contents |
|---|---|
| 1 | ADR-0005, this session addendum |
| 2 | packs/common/ — contract, schemas, anchor template |
| 3 | en, he, el pack manifests and orthography files |
| 4 | Lexical seed sets — lemmas, morphology, frames, senses, probes |
| 5 | Runtime boundary files — lift rules, readback rules, validators |