# Real Sample Validation Triage Report

> Date: 2026-04-19
> Roadmap: `docs/superpowers/plans/2026-04-18-scene-skill-real-sample-validation-roadmap-plan.md`

## 1. Triage Result

The three current mainline anchors now classify as:

1. `G2`: `mismatch-driven`
2. `G1-E`: `stable`
3. `G3`: `mismatch-driven`

## 2. Family-Level Reading

### G2

`G2` already had a real mismatch anchor from `台区线损大数据-月_周累计线损率统计分析`.

The key failures remain:

1. archetype mismatch
2. bootstrap mismatch
3. request contract missing
4. column semantics missing

This is a mainline compiler-family gap.

### G1-E

`G1-E` remains the only stable positive real-sample anchor.

Its current role is not to open a new scope but to act as the positive baseline against which future real-sample regressions are judged.

### G3

`G3` now has an executed real mismatch anchor.

The key finding is:

1. the real sample was executed
2. the sample did not stay in `paginated_enrichment`
3. the run collapsed into `local_doc_pipeline`
4. the result then failed closed

This is also a mainline compiler-family gap.

## 3. Scope Decision

The next bounded implementation scope should be:

1. `mainline G3 real-sample archetype correction`
2. followed by `mainline G2 real-sample contract correction`

The next scope should not be:

1. a boundary-family roadmap for `G6/G7/G8`
2. a deferred-family entry round for `G4/G5`

## 4. Reason

The strongest validation pressure still comes from unresolved mainline gaps.

As long as `G3` and `G2` both fail under real execution, opening boundary or deferred families would be scope drift.