the-mindscape-of-bro-tsong/reports/schema_change_notes.md

96 lines
2.2 KiB
Markdown

# Schema Change Notes
## Scope
This note covers schema changes made during the second content stabilization pass.
## Changed Schemas
### `schemas/source_article.schema.json`
Change:
- Formalized `source_type` enum:
- `original_article`
- `synthesis_note`
- `placeholder`
- Formalized `source_status` enum:
- `representative`
- `derived_synthesis`
- `placeholder`
Reason:
- Current data already used `synthesis_note` and `derived_synthesis`.
- The schema now matches implementation data instead of leaving enum drift implicit.
### `schemas/source_excerpt.schema.json`
Change:
- Added required `quote_status`.
- Added required `source_location`.
- Formalized `quote_status` enum:
- `exact`
- `condensed`
- `paraphrased`
- Formalized `excerpt_type` enum:
- `definition`
- `taxonomy`
- `mechanism`
- `application_rule`
- `value_claim`
- `boundary_rule`
- `validation_rule`
Reason:
- Distinguishes exact quotes from compressed or paraphrased evidence.
- Prevents `raw_excerpt` from silently containing condensed text.
### `schemas/regression_case.schema.json`
Change:
- Expanded `case_type` enum:
- `positive`
- `boundary`
- `misuse`
- `no_call`
- `selector_gate`
- `pipeline`
- Added optional evaluable fields:
- `should_call_model`
- `expected_primary_model`
- `negative_expected_models`
- `expected_classification`
- `expected_dominant_scarcity`
- `expected_max_depth`
- `minimum_required_elements`
- `forbidden_elements`
- `evaluation_mode`
Reason:
- Supports no-call, false-positive, selector-gate, and structured review cases without adding LLM evaluation.
## Not Changed
### `schemas/model_card.schema.json`
No new required model-card fields were added in this pass.
Reason:
- QPI and Intellectual Archaeology output contracts were strengthened in model JSON and cards.
- Making those fields schema-required is left for CCRA / Owner review to avoid over-weighting v0.2.
### `schemas/model_index.schema.json`
No additional index semantics were added in this pass.
Reason:
- Index continues to mirror model/card state and counts.
- Content review status remains in `reports/model_review_status.json`, not `model_index.json`.