[harrisonoconnorhover]

HubSpot could return a completed association batch with explicit empty relationships, causing the monitor to discard successful neighboring records. Healthy Contact pages could then hide unresolved Deal coverage gaps. Preserve valid batch results, recognize only precisely scoped empty-association responses, and keep coverage degraded until fresh reads actually recover the affected records. Recovery also works when native values and timestamps are unchanged.

Add a resumable, explicitly pinned test-account qualification workflow using the existing messy dataset, production polling and Safe Import, retained native readback, and a private offline report. Native browser testing also fixes the triage note's accessible label and mobile incident-list scrolling. The seeder now uses consistent HubSpot email/company domain suffixes while preserving deliberate fixture mismatches and acknowledged records on resume.

Validation:

• Real HubSpot and Salesforce development accounts: 2,900 seeded records; 420 controlled changes accepted and detected; 29 automatic trims and 24 approved phone fills verified with protected-field checks.
• 158 recorded qualification checks passed: 64 HubSpot monitor, 38 Salesforce monitor, 44 real browser, and 12 import result assertions.
• Combined with current main: all 774 tests, TypeScript, lint, app/public builds and diff checks passed. The preserved public CSV workspace passed desktop/mobile browser checks, including exact downloads and Undo. Native-branch secret scan and report desktop/mobile/filter checks passed.

The failed first HubSpot run remains preserved. Its 237 documented field changes were restored with guarded readback before the successful rerun. HubSpot's existing automation also created 351 companies because of the initial setup's mismatched suffixes; those remain separately attributed manual findings, outside repair authority. The future seed mapping correction was locally tested, without reseeding this retained cohort.

Credentials, native records and detailed evidence remain ignored and private. Qualification monitors are disabled. This proves native polling and local browser workflows; it does not claim hosted execution, webhook/CDC delivery, Slack delivery or live model behavior.