v0.6.0-beta.1 — Internal Beta

v0.6.0-beta.1 — Internal Beta

This release hardens reliability metrics correctness and makes After-Action Reports (AARs) durable.

Highlights

• Corrected KPI cost semantics:

  • Suboculo now separates unknown cost from known zero cost.
  • Added estimated_cost_known in task-run derivation.
  • cost_per_success now uses only successful runs with known cost.
  • Cost aggregates and guardrails are aligned to known-cost sample size.

• Durable After-Action Reports:

  • AARs are now persisted in backend storage (task_run_reports), not just transient UI state.
  • AAR endpoint uses read-through caching:
    • returns stored report when fresh
    • regenerates and stores when missing/stale
  • Outcome updates invalidate stale reports for affected runs.

• Task Runs UX reliability:

  • Task Run detail now auto-loads persisted AAR on open.
  • Debounced Task Runs reload behavior remains in place for stable filtering.

API/Behavior Notes

• Reliability KPI outputs now include known-cost counters used for cost metrics.
• /api/reliability/kpi-definitions reflects the updated cost formula and null conditions.
• /api/task-runs/:id/after-action-report now returns cache metadata indicating generated vs stored response path.

Validation

• Backend smoke suite updated and passing.
• Frontend build and Task Runs regression smoke checks passing.

Why this release matters

This release fixes a subtle but important KPI integrity issue: unknown-cost runs could previously bias cost metrics downward. It also turns AARs into persistent artifacts that survive browser reloads and remain consistent across sessions.