Skip to content

Latest commit

 

History

History
21 lines (18 loc) · 2.58 KB

File metadata and controls

21 lines (18 loc) · 2.58 KB

Architecture Decision Records (ADR) index

SPEC 10.1 requires ADRs for major design decisions. This directory is the single index of all ADRs. Each record documents the context, decision, and consequences for a significant architectural choice.

ADR Title Summary
0001 Monorepo Single repo for corpus, formalization, pipelines, kernels, and portal; Lake, uv, pnpm.
0002 Schema-first Canonical JSON schemas for all corpus and manifest artifacts; validation on every CI.
0003 Claim bundle Claim bundle as the unit of traceability; no theorem card without claim ID, no claim without source span.
0004 JSON storage for corpus and manifests Use JSON (not YAML) for all corpus and manifest files.
0005 Schema versioning Schema $id and version; migration notes in contributor-playbook.md; no auto-migration in v0.1.
0006 Domain expansion policy First domain chemistry/adsorption; new domains via metadata and docs, no schema enum.
0007 Pydantic for pipeline models Pipeline internal models mirror schemas via Pydantic; type validation and JSON round-trip.
0008 LLM worker interface deferred No LLM-assisted extraction in v0.1; human-curated or script-based only.
0009 Unit system deferred Kernel unit_constraints (strings) only; formal dimension/unit system in Lean deferred.
0010 Release integrity via checksums Changelog, manifest checksum, per-file hashes, release artifact SHA-256; Sigstore keyless signing of checksums.txt.
0011 LLM worker (suggest-only) Optional Prime Intellect chat completions; claim/mapping/Lean proposal sidecars; human-gated apply (proof-repair-apply for Lean); warn-only schema check for sidecars.
0012 Trust boundary and publish integrity Recompute manifest graphs; content-addressed build hash; normalization preserves unresolved links; canonical vs suggestion artifacts.
0013 LLM evaluation policy Prompt template digests; benchmarks/llm_eval reference fixtures; tasks.llm_eval; human rubric; optional live-eval scripts.

When adding a new ADR, use the next number (e.g. 0014) and add a row to this table.