feat/discovery

Best-effort guard means CI never catches export_columns anchor regression

e2e/platform/discovery-anchors.spec.ts:41

The if (count > 0) pattern means the test only fires when export_columns elements are present. Since the preview seed has no contracts, CI always has zero rows — so this assertion is structurally disabled in the required check on every PR. The comment documents this as intentional ('catches rot on dogfooding / future seed'), but the net effect is that removing export_columns from PipelineCell would go undetected in CI until the seed gains contracts. Worth noting as a known coverage gap.

No timing gap between .count() and .first().toBeVisible()

hasTable waited 10 s for the DOM to settle. No navigation or mutation occurs between .count() and .toBeVisible(), so the count→assert pattern is timing-safe.

Missing annotation when export_columns count is zero — inconsistent with existing pattern

e2e/platform/discovery-anchors.spec.ts:41

The hasTable=false branch pushes a skipped-assertion annotation so reviewers can see the skip in the HTML report. When the table is present but has no data rows (the CI case), the export_columns check silently no-ops with no annotation. Adding an else branch with test.info().annotations.push({ type: 'skipped-assertion', description: 'export_columns anchor skipped — no contract rows in seed' }) would make every CI run's skip visible, consistent with the existing pattern.

Structural asymmetry of inner conditional is justified

The three column-header anchors (monitoring, payment_monitoring, internal_site_ids) are on the header row and assert unconditionally inside hasTable. export_columns is on per-data-row elements — the extra inner guard is semantically correct. The updated comments explain this clearly.

Comment quality is good — distinction between header and row anchors is non-obvious

The new comments correctly distinguish between column-HEADER anchors (render when table is present, even on empty body) and per-DATA-ROW anchors (export_columns). This is non-obvious without reading PipelineCell. Comments add genuine value.

export_columns locator constructed twice — extract to a const

e2e/platform/discovery-anchors.spec.ts:41

page.locator('[data-tour="export_columns"]') is evaluated twice (once for .count(), once for .first()). No race risk here (DOM is settled after the hasTable wait), but a one-line extraction — const exportAnchor = page.locator('[data-tour="export_columns"]') — removes duplication and matches the style used for the table locator above.