You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardExpand all lines: DESIGN.md
+12Lines changed: 12 additions & 0 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -1777,3 +1777,15 @@ Rationale:
1777
1777
- Routing policy was already tested as a control-plane module; this promotes it into the actual model-call path without granting any new promotion authority.
1778
1778
- Operator views should reflect the route actually attempted, not reconstruct model choice from environment defaults after the fact.
1779
1779
- Capability-first routing keeps frontend labels, backend execution, and prompt language aligned: use the best policy-permitted model available on this surface, then record what was actually selected.
- Project routing-policy sub-receipts through router receipts and `/api/operator/receipts-telemetry`.
1785
+
- Summarize policy status and selected tier from persisted receipt facts, not from current environment defaults.
1786
+
- Keep the projection bounded to route IDs, model IDs, selected tier, attempt status, provider invocation, and receipt hashes.
1787
+
1788
+
Rationale:
1789
+
- The operator cockpit needs to show whether a response was routed, blocked, downgraded, or served by a provider without requiring code inspection.
1790
+
- Persisted receipts may be older than current configuration, so telemetry must describe what happened at the time of execution rather than what the current model router would choose now.
1791
+
- This remains observability only. It does not prove output quality, provider availability beyond the attempted call, or Council promotion authority.
Copy file name to clipboardExpand all lines: NEXT.md
+2Lines changed: 2 additions & 0 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -30,6 +30,8 @@ These are future focuses for public collaborators, not current completion claims
30
30
31
31
## Completed
32
32
33
+
- W-0132: Exposed routing-policy sub-receipts in operator telemetry (`agent_server.mjs`, `scripts/operator_telemetry_routes_test.mjs`, `DESIGN.md`[D-0057]). Router receipts now carry a sanitized `routing_policy` projection when dispatch supplies one, and `/api/operator/receipts-telemetry` summarizes routing-policy status and selected tier from receipt facts rather than reconstructing model choice from current environment defaults. This is cockpit observability only, not provider-quality proof or Council promotion authority. (Verification: `npm run test:operator-telemetry`; `npm run test:router`; `npm run check:council`)
34
+
33
35
- W-0131: Routed live chat and utility provider calls through the capability-first routing policy (`lib/dispatch.mjs`, `lib/routing_policy.mjs`, `lib/gemini_client.mjs`, `scripts/dynamic_router_test.mjs`, `scripts/routing_policy_test.mjs`, `DESIGN.md`[D-0056]). Dispatch now plans Gemini/OpenAI-compatible/Ollama-style calls with surface-bound route evidence before invocation, emits routing-policy metadata in execution receipts, preserves redirect and isolation block reasons, and sends Gemini max-output limits as actual generation controls. This is live dispatch admission and observability, not Council promotion authority. (Verification: `npm run test:routing-policy`; `npm run test:router`; `npm run check:council`)
34
36
35
37
- W-0126: Replaced the Python/Z3 bridge with a deterministic policy scorer. Removed the synchronous Python subprocess and Z3 SMT solver from memory conflict resolution. Implemented a pure JavaScript deterministic scorer based on confidence + 0.02 * reinforcement with a strict 0.05 dominance margin. Hard governance invariants are enforced as hard predicates before scoring. Re-anchored our roadmap around Anthropic's Building Effective Agents doctrine. (Verification: npm run check:council)
0 commit comments