Observability — later capabilities¶
Target
This page describes intended behaviour or structure. No implementation is claimed unless a section is explicitly marked Current implementation or Evidence.
| Service | Context | Note |
|---|---|---|
| IncidentService | BC-21 | IncidentService (Observability & Feedback group of the June-2026 catalog). |
| TemplateAnalyticsService | BC-21 | TemplateAnalyticsService (Template & Library group of the June-2026 catalog). |
| Capability | Owner when built | Taken up when |
|---|---|---|
IncidentService — incident lifecycle, paging, post-mortems linked to runs and alerts |
BC-21 | alerts alone no longer carry operator coordination, or the fleet (P12) needs incidents per product |
TemplateAnalyticsService — adoption and outcome analytics per template and library |
BC-22 with Template & Library | the generated-catalog (L1) gains usage events |
| Replay and simulation (original handoff §50) — re-execute a recorded timeline against another release in the sandbox | BC-22 | timelines carry enough references to reconstruct inputs deterministically |
| Adversarial evaluation suites — prompt injection, exfiltration, unsafe commands (§44) | BC-22 with Governance | the threat catalogue has test corpora |
| Cross-tenant benchmarks over anonymised datasets with consent | BC-22 | the consent model of memory promotion is in use |
| SLO-driven routing hints to the model router | BC-21 with Model & Compute | objectives per deployment exist for a window |
| Active-learning selection of evaluation tasks | BC-22 | datasets exceed what a suite can run within its budget |
| Projection export to a tenant's warehouse | BC-21 | a tenant asks for it under the export clause of data governance |
Related¶
- Overview · Evaluation · FinOps