Files
claw/docs/superpowers/reports/2026-04-19-102-full-sweep-dry-run-triage-report.md

118 lines
3.9 KiB
Markdown

# 102 Full Sweep Dry-Run Triage Report
> Date: 2026-04-19
> Plan: `docs/superpowers/plans/2026-04-19-102-full-sweep-dry-run-triage-plan.md`
> Source Dry-Run: `tests/fixtures/generated_scene/full_sweep_dry_run_2026-04-19.json`
> Triage Result: `tests/fixtures/generated_scene/full_sweep_dry_run_triage_2026-04-19.json`
## Scope
This triage explains the `62` non-pass records from the `102` scene full sweep.
It is classification-only:
1. no analyzer implementation was changed
2. no generator implementation was changed
3. `scene_execution_board_2026-04-18.json` was not updated
4. no scene was promoted
5. no implementation correction was started
## Input Buckets
| Bucket | Count |
| --- | ---: |
| Timeout / `source-unreadable` | 31 |
| Misclassified | 5 |
| No-report failure | 25 |
| Bootstrap target failure | 1 |
| Total non-pass records | 62 |
## Timeout Triage
All `31` timeout records originally had:
`generator timeout after 30s`
The second-level labels are:
| Timeout label | Count |
| --- | ---: |
| `timeout-unvalidated-source` | 19 |
| `timeout-large-source` | 8 |
| `timeout-known-family-sample` | 4 |
No timeout record is treated as unsupported family by default.
The `timeout-known-family-sample` records are important because they include scenes that already have current family mapping. These should be separated from the unvalidated-source and large-source buckets before any coverage conclusion is drawn.
## Misclassification Triage
All `5` misclassified records share the same second-level label:
`route-overprefer-host-bridge`
| Scene | Expected group | Expected archetype | Inferred archetype |
| --- | --- | --- | --- |
| `95598报修工单日管控` | `G3` | `paginated_enrichment` | `host_bridge_workflow` |
| `95598重要服务事项报备统计表` | `G3` | `paginated_enrichment` | `host_bridge_workflow` |
| `用电报装信息统计列表` | `G1-E` | `single_request_enrichment` | `host_bridge_workflow` |
| `配网支撑月报(95598抢修统计报表)` | `G3` | `paginated_enrichment` | `host_bridge_workflow` |
| `高低压新增报装容量月度统计表` | `G1-E` | `single_request_enrichment` | `host_bridge_workflow` |
This is the clearest routing-quality signal in the dry-run. The current generic route can over-prefer `host_bridge_workflow` when board-level expectations already point to `G3` or `G1-E`.
This report does not correct routing logic.
## No-Report Failure Triage
All `25` generic no-report failures were assigned the same failure stage:
`readiness-before-report`
The stderr archetype distribution is:
| Stderr archetype | Count |
| --- | ---: |
| `paginated_enrichment` | 17 |
| `local_doc_pipeline` | 4 |
| `multi_mode_request` | 2 |
| `host_bridge_workflow` | 1 |
| `single_request_enrichment` | 1 |
This means these failures are not source-missing failures. They reached a known archetype path, then failed before a structured `generation-report.json` was emitted.
The dominant no-report shape is:
`workflow evidence is incomplete for archetype paginated_enrichment`
## Bootstrap Target Failure
One failure remains separately tracked:
| Scene | Inferred archetype | Reason |
| --- | --- | --- |
| `用户停电频次分析监测` | `page_state_eval` | `bootstrap_target` |
It is not merged into the no-report bucket.
## Recommended Next Blocker
The next blocker should be handled in this order:
1. `timeout-known-family-sample`
2. `route-overprefer-host-bridge`
3. `readiness-before-report`
Reasoning:
1. `timeout-known-family-sample` affects already-mapped scenes and can distort coverage if left mixed with unvalidated timeouts.
2. `route-overprefer-host-bridge` is a clean routing-quality issue across existing `G3` and `G1-E` expectations.
3. `readiness-before-report` is high volume, but it first needs structured fail-closed reporting before implementation correction.
## Stop Statement
This triage is complete and stops here.
It does not start an implementation plan, does not promote scenes, and does not update the execution board.