37 lines
2.0 KiB
Markdown
37 lines
2.0 KiB
Markdown
|
# Deep Interpretation Schema v1
|
||
|
|
|
||
|
|
This document records the current evidence-first deep interpretation output contract used by `build_evidence_report`.
|
||
|
|
|
||
|
|
## `structured_result_interpretation_v1`
|
||
|
|
|
||
|
|
One object per stage. It reports local stage facts and the interpretation boundary for article-style Results writing.
|
||
|
|
|
||
|
|
Required fields:
|
||
|
|
- `schema`: always `structured_result_interpretation_v1`.
|
||
|
|
- `stage_group`: coarse evidence group, such as `classification_structure_support` or `rna_response_candidate_prioritization`.
|
||
|
|
- `interpretation_level`: one of `publishable_result`, `descriptive_only`, `candidate_only`, or `boundary_only`.
|
||
|
|
- `local_facts`: compact manifest/parsed facts used by the interpretation.
|
||
|
|
- `evidence_refs`: local evidence handles from extracted stage evidence.
|
||
|
|
- `missing_slots`: missing outputs or partial-success gaps that prevent stronger claims.
|
||
|
|
- `downgrade_reasons`: explicit reasons for lowering interpretation strength.
|
||
|
|
- `boundary`: reviewer-facing wording for what must not be claimed.
|
||
|
|
|
||
|
|
## `grouped_result_interpretation_v1`
|
||
|
|
|
||
|
|
One object per evidence group. It gives cross-stage Results narrative while preserving the weakest required boundary.
|
||
|
|
|
||
|
|
Required fields:
|
||
|
|
- `schema`: always `grouped_result_interpretation_v1`.
|
||
|
|
- `group`, `title`, `stages`: group identity and included completed stages.
|
||
|
|
- `readiness`: the lowest readiness among included structured stage interpretations.
|
||
|
|
- `result_interpretation`: cross-stage Results-style narrative based only on local stage evidence.
|
||
|
|
- `evidence_refs`, `missing_slots`, `downgrade_reasons`: merged from stage-level entries.
|
||
|
|
- `boundary`: group-level claim boundary.
|
||
|
|
|
||
|
|
## Boundary Rules
|
||
|
|
|
||
|
|
- High MSA gap or missing support prevents strong topology and subfamily-function claims.
|
||
|
|
- `partial_success` stages are descriptive unless missing subproducts are irrelevant to the claim.
|
||
|
|
- RNA-seq and regulatory evidence can prioritize candidates, but cannot prove causal regulation without experimental validation.
|
||
|
|
- RAG/literature context can support wording and background, but cannot replace missing local results.
|