Phase 05: Simplify Source Status — Context

Gathered: 2026-03-11 Calibration: architect

Phase Boundary

Simplify the source status/health system by separating operational state (active/paused) from health diagnosis (working/declining/improving/failing). Reduce 7 health statuses to 4, fix auto-pause masking health diagnosis, and update all UI/filters/tests.

Decisions

Migration Strategy (7 → 4 Health Statuses)

• Direct rename approach: healthy→working, warning+critical→failing, declining stays, improving stays
• auto_paused removed from health_status — it's an operational concern, not a health diagnosis
• pending/unhealthy in import contexts: map pending→working (new sources start healthy), unhealthy→failing
• Data migration required: UPDATE sources SET health_status = 'working' WHERE health_status = 'healthy', etc.

Two-Axis Model Design

• No new schema columns — operational state already encoded via active boolean + auto_paused_until timestamp
• health_status column stores ONLY health values: working, declining, improving, failing
• Operational state derived from existing columns: active? && !auto_paused? = "active", !active? = "paused", auto_paused? = "auto-paused"
• Filters updated: "Active" filter excludes auto-paused sources (currently includes them)

Auto-Pause + Health Interaction

• Auto-paused sources continue receiving health checks on a slow cadence (every 24h)
• health_status updates independently of pause state — a source can be "failing AND auto-paused"
• This provides visibility into whether a paused source has recovered before manual unpausing
• SourceHealthMonitor.determine_status no longer short-circuits to "auto_paused" — evaluates health first, pause is a separate axis
• Implementation: auto-paused sources get next_fetch_at set to 24h intervals for health probe fetches

Health Transition Rules (Simplified Decision Tree)

• New determine_status logic (evaluated in order):
  1. rate >= healthy_threshold → working
  2. rate < auto_pause_threshold → failing
  3. consecutive_failures >= 3 → declining
  4. improving_streak? (2+ successes after failure) → improving
  5. Fallback (between thresholds, no streak) → declining
• warning_threshold config setting removed entirely
• healthy_threshold and auto_pause_threshold remain as the two boundary points

Open (Claude's discretion)

• Health badge colors: working=green, declining=yellow, improving=blue, failing=red
• Import session health_status values: new imports default to "working" instead of "pending"
• SourceHealthReset behavior: reset should set health_status to "working" (was "healthy")

Deferred Ideas

None captured.