Rubric Configs

Source: reference/rubric-configs/README.md · synced 2026-06-12

Per-sub-vertical configuration files consumed by the matching engine (Steps 5–8).

Each config defines the conditional mapping rules + scoring weights for one (sub_function, sub_vertical) cell. Together with the vertical encyclopedias in ../verticals/, these constitute the proprietary domain layer that distinguishes the system from a generic LLM matcher.

Schema source of truth: plans/amendment-3.1+5.3+6.1-bootstrap-rubric-extension.md §4c. Generated by the bootstrap mine.

Status (2026-06-11)

The 4 original grant-vertical stubs (commodities.json, maritime.json, offshore.json, tech-gaming.json) used the wrong schema — they were keyed on the 4 grant-level marketing verticals, but the locked taxonomy fires rubrics at 22 sub-vertical granularity per SOURCE-OF-TRUTH.md §3b.

Action taken: The 4 stubs have been archived to _archive/2026-06-11-rubric-stubs-4-vertical/ for audit trail.

Replacement schedule: 22 sub-vertical rubric files will be emitted by the bootstrap mine per plans/amendment-3.1+5.3+6.1-bootstrap-rubric-extension.md. File names follow {sub-function}.{sub-vertical}.json — e.g. cargo-ops.energy.json, ship-ops.tanker.json, risk-analytics.credit.json.

Locked sub-vertical inventory (22 cells)

Cell breakdown per SOURCE-OF-TRUTH.md §3b.

Schema

See plans/amendment-3.1+5.3+6.1-bootstrap-rubric-extension.md §4c “22 per-sub-vertical rubric configs” for the canonical RubricConfig TypeScript schema and the Signal / TitleNorm sub-schemas.