Operating Hero Health
Hero Health is the internal reliability dashboard at /hero-health . Granting access There is no admin-management UI. Add the platform admin’s existing profile ID directly: insert into...
Last updated
Hero Health is the internal reliability dashboard at /hero-health.
Granting access
There is no admin-management UI. Add the platform admin’s existing profile ID directly:
insert into public.hero_health_admins (user_id)
values ('USER_PROFILE_UUID');
user_id must match public.user_profiles.id. On the next signed session check, the Hero Health menu and command-palette entry become available. Removing the row removes access.
Dashboard sections
- Reliability ledger: failure clusters, frequency, distinct users, status, guard/output-contract class, and admin-only repair details.
- Learned query queue: pending and flagged global redacted templates.
- Opus escalation effectiveness: raw Sonnet/Opus attempt and rescue counts, costs, and per-cluster outcomes. Rates below 20 Opus-attempted clusters are provisional.
- Digest configuration: read-only schedule and delivery behavior.
Reviewing a learned query
- Open a Pending or Flagged item.
- Review its redacted question, SQL template, trusted source, and allowed tables.
- Approve, reject with a reason, or edit and approve.
- Treat approval as global: runtime still revalidates the template against the requesting user and property context.
Never copy raw resident information into a learned template. The template must use typed :property_id and retain no user/property identifiers.
Scheduled jobs
- Failure clustering: nightly before the healer.
- Healer: nightly, DEV credentials only.
- Weekly digest: Monday at 16:00 UTC.
- Spike check: daily at 05:00 UTC.
- Detailed occurrence redaction: daily after 90 days.
Email recipients
Digest and spike email recipients come from hero_health_admins.user_id → user_profiles.id → user_profiles.email_address. Missing email addresses are skipped.