fix(report): an execution with no scores is the worst one, not the best

[lopadova]

Work item

Branch or work item: claude/review-round-1 — two Codex findings on the merged report-UI work.

Summary

An errored execution (no scores at all) was ranking as the best candidate for a row and hiding the failure, and the precision panel read fields only from the top level while single-run reports nest them under precision.run.

Changes

Files/subsystems changed: resources/js/utils/reportBlocks.ts, resources/js/components/report/PrecisionPanel.tsx, resources/js/i18n/messages.ts, resources/js/components/report/reportPanels.test.tsx.

reportBlocks.ts — unscored ranks last

Picking the best execution per row by mean score returned positive infinity when an execution had no scores. An execution with no scores is an errored one, and it is the most interesting thing that can have happened to a row — ranking it best hid it behind a sibling that happened to score. It now sorts last (Number.NEGATIVE_INFINITY), so the failure is what the reader sees.

This is the same "absent is not zero" shape already handled correctly elsewhere in the same file — applied inconsistently here.

PrecisionPanel.tsx — nested block + three-state resolvability

• resolution and target_resolvable were read from the top level only, but the report writes both nested under precision.run for a single-run report: the panel silently rendered nothing. It now falls back to the nested block.
• target_resolvable is read as three-state. Absent means unknown, not "not resolvable" — a missing field was rendering the assertive "target not reachable" copy on reports that never computed it.

UI/UX changes

New text_target_unknown string (en + it): a neutral line when resolvability was not computed, replacing a wrong assertion.

Documentation changes

None — no documented behaviour changes; the panel now renders on report shapes where it previously rendered nothing.

Test gate

• Local gates run for changed scope
• Backend/package tests — not applicable, frontend-only change
• UI tests — 29 vitest tests (26 → 29, three regressions added: nested precision block, unknown resolvability, unscored-execution ranking)
• tsc --noEmit clean
• Playwright scenarios — no new user flow; panel copy and ordering only
• CI checks green for PR — pending on this push
• GitHub Copilot Code Review requested

Stability impact

• @api classes/methods/constants changed? No
• Contract/README updated if needed: not needed — the report contract is unchanged; this reads a field it already emits.

Security / privacy impact

• Token/secret handling reviewed — untouched
• Redaction rules preserved — untouched
• No plaintext secrets in logs/docs/UI

Risk / rollback

• Risk: Low. Both changes surface information that was previously hidden or misstated.
• Mitigation: covered by three regression tests pinning the exact shapes.
• Rollback: revert the merge commit; no data or contract migration involved.

---

Generated by Claude Code