# Bleeding-Edge Scorecard v1.0

The reproducible score is generated by:

```bash
python scripts/check_repo_health.py
```

The supplemental scorecard currently checks:

- Python project metadata.
- Streamlit native AppTest coverage.
- Pydantic contracts and versioned registries.
- Concern-boundary scanner.
- Conductor templates, agents, skills and workflows.
- Main CI lane.
- Latest-compatible dependency canary.
- GitHub Pages deployment.
- Dependabot coverage.
- Public claim-control documents.
- Enhancement shortlist decision record (Track 049).

## Analytical Enhancement Commitments (v1.8.5)

The following enhancements are classified and scheduled by the Track 049 shortlist (`docs/decisions/enhancement-shortlist-v1.8.5.md`):

### Wave 1 — Complete (Must)

| # | Enhancement | Category |
|---|---|---|
| 1 | Subgroup equity/rural stratification on existing charts | Subgroup + Visual |
| 2 | Tornado sensitivity chart | Analysis + Visual |
| 3 | Seeded Monte Carlo ensembles (all-scenario) | Simulation |
| 4 | Waterfall / decomposition chart | Visual |
| 5 | Cohort-stratified runs | Simulation |

### Wave 2 — Complete (Should)

| # | Enhancement | Category |
|---|---|---|
| 6 | Variance decomposition | Analysis |
| 7 | Heatmap matrix (scenario × subgroup) | Visual |
| 8 | Policy shock sequences | Simulation |
| 9 | Uncertainty ribbon on stock-flow path | Visual |
| 10 | Ridge/violin distribution with subgroup facets | Visual |
| 11 | Stress-test scenarios | Simulation |

### Wave 3 — Complete (Could)

| # | Enhancement | Category |
|---|---|---|
| 12 | Phase portraits / vector fields | Visual |
| 13 | Agent-based subgroup replay | Simulation |
| 14 | Regime sweep simulation | Simulation |
| 15 | Payoff surface (3D) | Visual |
| 16 | Interaction scan | Analysis |

### Explicit Exclusions

| Item | Reason |
|---|---|
| Person-level analysis | Aggregate-subgroup boundary |
| Private / patient-level data | Public-data boundary |
| Black-box solvers | Transparency requirement |
| Solver promotion | Delegated to Track 048 |

Target posture: `bleeding-edge controlled`, meaning current tooling is used with explicit public-claim, dependency, testing and deployment gates, and analytical enhancements follow the ranked shortlist.

## Verification

```powershell
rg -n "rural|depriv|equity|subgroup|tornado|variance decomposition|phase portrait|vector field|heatmap|violin|waterfall|frontier|ensemble|stochastic" models docs conductor
python -m pytest -q models/tests
```

## Public-only SOTA gate

SOTA/bleeding-edge status now requires public-only registry purity, parameter traceability, public aggregate calibration, structural uncertainty, full VOI, visual grammar, and release reproducibility gates.

## Release-note gate

Bleeding-edge/SOTA release notes are governed by [docs/release/bleeding-edge-sota-release-handling.md](release/bleeding-edge-sota-release-handling.md). Historical root release notes have been moved to [docs/release/archive/](release/archive/); new releases should use GitHub Releases, `CHANGELOG.md`, generated model cards, and release manifests.