# Methodology Changelog

All notable changes to the scoring methodology. Every entry corresponds to an
`INDEX_VERSION` (semver) in `packages/scoring/src/version.ts`.

## [1.0.0] — UNRELEASED (draft)

Initial methodology.

- Task-based scoring on O*NET 30.x task statements, importance-weighted aggregation
  to occupations (O*NET-SOC 2019).
- Five dimensions: automatability 0.35, feasibility 0.20, cost_ratio 0.15,
  barriers 0.20 (inverted), adoption_velocity 0.10.
- Three sub-scores per occupation: exposure, substitution (headline composite),
  augmentation — augmentation rated separately per task, outside the composite.
- Multi-model LLM rater panel (≥2 frontier models via `RATER_MODELS`); confidence
  bounds (`score_low`/`score_high`) derived from rater disagreement envelopes.
- 5% human review sample; expert Delphi overrides replace LLM bands where triggered.
- Grounding research: `docs/research/01-…04-*.md` (compiled 2026-08-05).