Summary

On the dashboard overview, a recipe corroborated on some phases but missing others (e.g. a contributor without the hardware to run the performance phase) was dragged to UNTESTED, hiding the coverage it did have. This adds a PARTIAL rollup state so incomplete-but-clean coverage reads honestly.

Motivation / Context

The recipe-wide overview rollup (rollupRecipe) picks the worst row state, and UNTESTED ranks above SINGLE/CONFIRMED. So a single not-run row flipped the entire recipe badge to UNTESTED — making a well-corroborated recipe look untested purely because one phase (commonly performance) was never run on the available hardware.

Fixes: N/A
Related: #1404 (GP4 corroborate generator), #1400 (interim evidence dashboard epic)

Type of Change

• Bug fix (non-breaking change that fixes an issue)

Component(s) Affected

• Other: pkg/corroborate dashboard renderer (display-only)

Implementation Notes

• New PARTIAL state used only by rollupRecipe (the overview badge, landing tallies, and left-tree dot) — never a per-row or per-phase state.
• Rule: positive coverage on some phases plus any not-run rows → PARTIAL. Real problems still dominate (CONTESTED/FAILING win); fully-covered recipes stay CONFIRMED/SINGLE; recipes with no positive coverage anywhere stay UNTESTED.
• Distinct violet chip (--partial:#a78bfa); amber was already CONTESTED.
• Renderer-only and display-only. PARTIAL is derived client-side from the per-row consensus already present in index.json. The grid, per-phase cells, dashboards, and per-run views keep UNTESTED/not-run unchanged. No change to the evidence/ingest format, meta.json, the baked JSON schema, or the Go consensus model — a re-publish of the same evidence just renders the overview badge correctly. Byte-determinism of the emit is preserved (static asset).

Testing

go test -race ./pkg/corroborate/...          # embed + byte-determinism: PASS
node --check <renderer script>               # valid JS
# functional test against the extracted rollupRecipe (8 cases): all PASS
#   [SINGLE,UNTESTED]->PARTIAL  [CONFIRMED,UNTESTED]->PARTIAL
#   [CONFIRMED,CONFIRMED]->CONFIRMED  [SINGLE,CONFIRMED]->SINGLE
#   [UNTESTED,UNTESTED]->UNTESTED  []->UNTESTED
#   [FAILING,SINGLE,UNTESTED]->FAILING  [CONTESTED,SINGLE,UNTESTED]->CONTESTED

No Go source files changed (HTML renderer only), so the Go lint gate is N/A.

Risk Assessment

• Low — Isolated, display-only renderer change, easy to revert.

Rollout notes: Takes effect on the next dashboard publish; no data migration. Backwards-compatible (reads existing index.json).

Checklist

• Tests pass locally (make test with -race) — go test -race ./pkg/corroborate/...
• Linter passes (make lint) — no Go files changed; renderer JS syntax checked via node --check
• I did not skip/disable tests to make CI green
• I added/updated tests for new functionality — covered by the determinism test + the rollupRecipe functional check above
• I updated docs if user-facing behavior changed — N/A (no doc surface enumerates rollup states)
• Changes follow existing patterns in the codebase
• Commits are cryptographically signed (git commit -S)