fix: add confidence gate to dashboard publishable count (MUSE-6) [Dani Roxberry]

[nebula-ai-agent]

Summary

The dashboard "Ready To Publish" metric was counting items with state === "approved" and no risk flags, but publishListing() also enforces confidence >= 0.75. This caused the metric to overcount — low-confidence approved items would appear ready but fail at publish time with "Low-confidence items must stay in review."

Fix

Added && item.confidence >= 0.75 to the filter in apps/admin-web/src/app/page.tsx line 11, matching the business-logic gate in apps/admin-web/src/lib/workflows.ts line 52.

Testing

• Pre-existing tests in apps/admin-web/tests/workflows.test.ts should still pass
• Manually: items with state=approved, no risk flags, and confidence < 0.75 will no longer increment the metric

Requested by Dani Roxberry via Nebula.

— opened via Nebula by Dani Roxberry (@droxey)

[nebula-ai-agent]

Code Review — PR #3

Verification Summary

Check	Result	Detail
tests	PASS	6 vitest + 4 pytest pass locally; CI (web-and-worker, ios-package) green
lint	PASS	ESLint clean, Ruff clean on worker
scope	PASS	2 files, 3 insertions, 2 deletions — directly addresses MUSE-6
security	PASS	No new deps, no auth changes, no user input handling
ui_behavior	N/A	Server component data filter; no interactive UI to test

Findings

[MAJOR] Missing test for dashboard filter (apps/admin-web/src/app/page.tsx:11)
The confidence gate && item.confidence >= 0.75 has no test coverage. The only test file (workflows.test.ts) covers publishListing(), but page.tsx has zero tests. Without a test that asserts low-confidence approved items are excluded from the "Ready To Publish" count, this fix is unverified at the integration level.

Expected: A test that exercises the filter with mock data including items where confidence < 0.75, verifying they are excluded.

[MINOR] Filter condition order inconsistency
publishListing() in workflows.ts:46-53 checks confidence FIRST (it is the critical gate that throws). The dashboard filter checks confidence LAST. While functionally equivalent, the ordering is inconsistent. Consider matching: item.state === "approved" && item.confidence >= 0.75 && item.riskFlags.length === 0.

[NIT] Evidence bundle inaccuracy
The evidence bundle at triggers/daily-morning-sweep/state/evidence-pr3.md says the PR has a "CI fix commit (6a8bd9a)" but the PR is a single squashed commit (150ce48) containing both changes. The Files Changed section only lists ci.yml.

Verdict

changes_requested — The confidence gate logic is correct and all existing tests pass, but the new dashboard filter has no test coverage. Add a test that verifies low-confidence approved items are excluded from the "Ready To Publish" metric.

Reviewed by Code Reviewer via Nebula