# Quality Score

**Status:** Generated
**Updated:** 2026-06-11
**Owner:** Project maintainers
**Generated by:** `mars scores export --repo <path>`
**Source DB:** `/path/to/local-redacted`
**Evidence window:** 30d

## Purpose

This file is the repo-visible quality artifact for MARS evidence. It is generated from the scoring database, telemetry, ticket state, and preserved manual notes so agents can inspect quality without treating the dashboard as the source of truth.

## Grading Scale

| Grade | Meaning |
| --- | --- |
| A | Complete, tested, documented, and consistently healthy. |
| B | Functional with minor gaps or hardening work still open. |
| C | Partially healthy; meaningful implementation or proof work remains. |
| D | Unhealthy or under-proven; corrective work is needed. |
| F | Failing repeatedly or missing the expected quality surface. |
| Insufficient evidence | Live SQLite evidence is missing or too sparse to grade honestly. |

## Overall Roll-Up

| Area | Grade | Evidence | Next Action |
| --- | --- | --- | --- |
| Role health | Insufficient evidence | No scored role outcomes. | Keep recording terminal outcomes for every role run. |
| Factory pace | Insufficient evidence | No trace pace evidence. | Use trace pace rows to target high-turn or limit-stop roles before raising runtime limits. |
| Terminal outcomes and checks | Insufficient evidence | No terminal outcomes recorded. | Investigate failed checks, guardrail blocks, no-op runs, and human follow-up. |
| Ticket flow and intervention debt | D | 13 backlog, 1 in-progress, 73 done, 1 open intervention-debt | Drain in-progress and high-priority intervention debt before ordinary backlog work; keep medium/low intervention debt visible without blocking product progress. |
| Telemetry and dogfood | A | No telemetry triage targets; None recorded | Promote recurring telemetry and dogfood failures into bounded remediation. |
| Evidence coverage | Insufficient evidence | SQLite present but no role scores in the selected window. | Run harness jobs with scoring enabled when evidence is insufficient. |

**Current overall grade: Insufficient evidence.**

## Role Health

No role scores were available in the selected evidence window.

## Factory Pace

No trace pace evidence was available in the selected evidence window.

## Convergence And Guardrails

No convergence or guardrail evidence was available in the selected evidence window.

## Evidence Signals

| Signal | Evidence |
| --- | --- |
| Role scores | No scored role outcomes. |
| Factory pace | No trace pace evidence. |
| Terminal outcomes | No terminal outcomes recorded. |
| Convergence failures | No convergence failures or guardrail blocks recorded. |
| Stuck tickets | 1 in-progress: `docs/tickets/in-progress/T-011-measure-and-optimize-factory-pace.md` |
| Failed dogfood | None recorded |
| Guardrail blocks | None recorded |
| Intervention debt | 1 open intervention-debt, 23 total intervention-debt |
| Check results | 0 passed, 0 failed |
| No-op runs | None recorded |
| Human follow-up | None recorded |
| Deterministic remediation | No remediation attempts recorded. |
| Top telemetry triage targets | No telemetry triage targets |

## Top Improvement Targets

1. Run harness jobs with scoring enabled or pass `--db` for the repo-specific SQLite database.
2. Drain 1 in-progress ticket(s) before starting new backlog work.
3. Resolve, downgrade, or leave non-blocking evidence for 1 open intervention-debt ticket(s).

## Source And Target Contract

- Refresh this artifact with `mars scores export --repo <path>`.
- The export reads role scores, terminal outcomes, tickets, telemetry, dogfood, guardrail blocks, no-op runs, human follow-up, deterministic remediation attempts, and check outcomes from the same evidence used by dashboard quality views.
- Convergence failures (circle detection, max-turn/max-tool stops, no-op outcomes) and guardrail block rates are broken out per repo/role from existing trace summaries and terminal outcome counts.
- The dashboard may link to or display this data, but `docs/QUALITY_SCORE.md` remains the repo-visible source of truth for quality claims.
- The quality score separates shipped feature scenarios from enabler work; feature claims still require mapped BDD evidence.
- Low role scores and recurring failures are reported as improvement targets by default; pass `--create-intervention-debt` when ticket materialization is deliberately wanted.
- Missing optional telemetry leaves an explicit evidence warning instead of failing the export.

## Manual Notes

<!-- BEGIN MANUAL NOTES -->
- 2026-05-03 local recovery verification passed `go test ./...`, `go test -race -count=1 ./...`, `go test -cover ./...`, and `go build -o <validation-root> ./cmd/mars`.
- Total coverage from `<validation-root>` is 64.2%; `internal/power` remains platform glue with no direct unit coverage.
- `golangci-lint run` was not available locally (`command not found`); rely on CI or install golangci-lint before claiming local lint evidence.
<!-- END MANUAL NOTES -->

## Generation Notes

- No export warnings.