Files
claw/docs/superpowers/plans/2026-04-19-scene-skill-102-full-coverage-framework-plan.md

299 lines
7.3 KiB
Markdown

# Scene Skill 102 Full Coverage Framework Plan
> Date: 2026-04-19
> Status: Draft
> Upstream Design: `docs/superpowers/specs/2026-04-19-scene-skill-102-full-coverage-framework-design.md`
## Plan Intent
Turn the current sgClaw post-roadmap work into a single controlled framework for driving the `102` scene set toward full bounded coverage.
This plan is the parent roadmap for all later bounded plans. Future bounded plans must fit inside one of the routes defined here.
## Current Baseline
Current integrated baseline:
| Status | Count |
| --- | ---: |
| `auto-pass` | 48 |
| `fail-closed-known` | 47 |
| `adjudicated-valid-host-bridge` | 4 |
| raw `source-unreadable` | 3 |
| Total | 102 |
Timeout hygiene overlay:
| Hygiene interpretation | Count |
| --- | ---: |
| `timeout-as-pass-candidate` | 2 |
| `timeout-as-fail-closed-candidate` | 1 |
| `timeout-still-unreadable` | 0 |
| `timeout-rerun-error` | 0 |
## Overall Goal
The overall goal is:
`100% bounded framework coverage for the current 102 scene set`
This means:
1. every scene is covered by a supported framework path
2. every non-pass scene has a structured and named reason
3. no unresolved timeout, unsupported-family, or route-conflict bucket remains
It does not require `100% auto-pass`.
## Scope Guardrails
1. do not start `G4/G5`
2. do not add new families unless this parent framework is updated first
3. do not treat diagnostics as promotions
4. do not update `scene_execution_board_2026-04-18.json` inside diagnostic or bounded recovery plans
5. do not mix timeout policy work with contract recovery work in the same bounded implementation plan
6. do not create semantics-only micro-plans that are not tied to one of the routes below
## Workstreams
1. `WS1` Coverage and Reporting Integrity
2. `WS2` Mainline Contract Closure
3. `WS3` Boundary Bucket Handling
4. `WS4` Promotion and Board Reconciliation
## Phase 0: Freeze the Parent Framework
### Objective
Make this plan the single parent framework for the next improvement cycle.
### Tasks
1. freeze the current integrated baseline
2. freeze the five framework layers
3. freeze the route order
4. forbid out-of-framework micro-plan drift
### Deliverables
1. parent framework design
2. parent framework plan
### Acceptance Criteria
1. future bounded plans can be mapped to one framework layer
2. future bounded plans can be mapped to one route
## Phase 1: Close Reporting Integrity
### Objective
Finish the reporting-side work so the `102` scene set is measured correctly before further implementation.
### Route
`Route 1: Layer E hygiene integration`
### Tasks
1. preserve raw timeout counts
2. preserve hygiene-aware timeout interpretation
3. preserve route adjudication
4. preserve structured fail-closed buckets
5. produce reconciliation-friendly current-state reporting
### Deliverables
1. timeout hygiene integration assets
2. reconciliation-friendly integrated reporting
### Acceptance Criteria
1. no unresolved timeout interpretation remains
2. no unresolved route conflict remains
## Phase 2: Mainline G3 Contract Closure
### Objective
Reduce the largest remaining fail-closed bucket in a controlled way.
### Route
`Route 2: G3 / paginated_enrichment`
### Tasks
1. freeze the current `G3` fail-closed subgrouping
2. select the top repeated recoverable pattern
3. implement bounded contract recovery
4. rerun only the bounded validation needed by that slice
5. measure delta against the parent baseline
### Deliverables
1. bounded G3 implementation plan(s)
2. bounded G3 implementation report(s)
3. updated coverage delta assets
### Acceptance Criteria
1. no scene-name hardcoding
2. no gate relaxation
3. canonical `G3` and real-sample `G3` remain stable
## Phase 3: Mainline G2 Closure
### Objective
Reduce the remaining `multi_mode_request` fail-closed bucket.
### Route
`Route 3: G2 / multi_mode_request`
### Tasks
1. freeze the current `4` G2 fail-closed records
2. identify the common missing contract
3. implement one bounded G2 correction slice
4. rerun bounded validation
### Deliverables
1. bounded G2 implementation plan(s)
2. bounded G2 implementation report(s)
### Acceptance Criteria
1. real-sample `G2` pass remains stable
2. no route drift into host-bridge or other families
## Phase 4: Mainline G1-E Closure
### Objective
Reduce the remaining `single_request_enrichment` fail-closed bucket.
### Route
`Route 4: G1-E / single_request_enrichment`
### Tasks
1. freeze the current `2` G1-E fail-closed records
2. identify the common missing contract
3. implement one bounded G1-E correction slice
4. rerun bounded validation
### Deliverables
1. bounded G1-E implementation plan(s)
2. bounded G1-E implementation report(s)
### Acceptance Criteria
1. real-sample `G1-E` pass remains stable
2. no route drift into host-bridge or page-state families
## Phase 5: Boundary Buckets After Mainline
### Objective
Touch boundary-family fail-closed buckets only after the mainline buckets have been reduced or explicitly deferred.
### Route
`Route 5: local_doc_pipeline and host_bridge_workflow remaining fail-closed`
### Tasks
1. inspect the `5` local-doc records
2. inspect the `1` host-bridge fail-closed record
3. decide whether to defer or open one bounded boundary correction slice
### Deliverables
1. boundary bucket decision report
2. optional bounded boundary plan
### Acceptance Criteria
1. no boundary slice starts before mainline routes are resolved or deferred
## Phase 6: Promotion and Board Policy
### Objective
Define how stronger framework-resolved statuses may flow back into official scene status assets.
### Route
`Route 6: promotion and board reconciliation`
### Tasks
1. define promotion thresholds
2. define how hygiene-aware timeout results are represented
3. define how structured fail-closed progress is represented
4. define what can and cannot update the execution board
### Deliverables
1. promotion policy design
2. execution-board reconciliation plan
### Acceptance Criteria
1. diagnostics remain distinct from promotion
2. execution board updates become rule-driven instead of ad hoc
## Route Order
The route order is fixed:
1. finish reporting integrity
2. reduce `G3` fail-closed bucket
3. reduce `G2` fail-closed bucket
4. reduce `G1-E` fail-closed bucket
5. inspect boundary fail-closed buckets
6. define promotion and board reconciliation policy
No bounded plan may skip upward in this order unless this parent plan is revised.
## Required Contents for Future Bounded Plans
Every future bounded plan must include:
1. parent route reference
2. parent framework layer
3. fixed input bucket
4. exact files allowed to change
5. files that must not change
6. expected coverage delta
7. stop statement
If one of these is missing, the bounded plan is not valid under this framework.
## Completion Criteria
This parent framework remains active until all of the following are true:
1. `unsupported-family = 0`
2. `missing-source = 0`
3. `misclassified-unresolved = 0`
4. `timeout-still-unreadable = 0`
5. every remaining non-pass scene is either:
- structured fail-closed
- adjudicated valid host-bridge
- policy-recognized timeout rerun hygiene result
6. board reconciliation policy exists
## Stop Statement
This is a parent framework plan.
Do not implement code directly from this plan.
All implementation must happen through later bounded plans that explicitly declare which route and which layer they belong to.