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
{{ message }}
Repository navigation
Tracer: Attribute multi-model usage across owned Sessions #503
#39 — Usage read model / Typert remote / Profile contract; part of #34.
What to build
The PersonaBot Profile attributes real Orchestrator, Assignment, and DSH Subagent attempts to their owning Bot and execution role; a Turn that uses more than one provider/model contributes to each exact route separately.
Acceptance criteria
Trusted Session ownership attributes Orchestrator, Assignment, and DSH Subagent usage to the correct PersonaBot; unowned or ambiguous work never appears under a guessed Bot.
A Turn using multiple models appears as separate provider/model contributions rather than one mixed bucket, with the correct execution role for each attempt.
Provider-reported usage from retry and failed attempts counts exactly once; a missing provider report is marked unknown rather than zero.
Profile shows the role and route breakdown through the Host read model without exposing private DSH logs or deleted-Session detail.
Focused tests cover mixed-route, retry, failure, Subagent ownership, and unknown cases; an isolated DSH scenario provides a visible result and screenshots for Human QA.
Task: codex/local/01a0e445-0c0f-70e1-bad4-fd99c8cfaa21 (not a DSH Session)
Issue scope: Attribute durable model settlements individually to exact provider/model and trusted Orchestrator, Assignment or Subagent ownership; retain reported failure/retry usage and unknown buckets.
Delivered: per-settlement actual provider/model attribution and trusted Orchestrator / Assignment / DSH Subagent breakdown in the public Profile query and Client.
Verification: 86 focused tests; complete latest-head CI passed (lint, format, typecheck, full tests, build and docs). Real isolated DSH calls independently matched 135,491 tokens across three execution roles and two model routes, unchanged after refresh and cold Host restart. Light/dark screenshots are in the PR and both links return HTTP 200.
Review: CodeRabbit's reusable-verifier aggregation finding fixed and revalidated. The later raw-report versus projection-field finding was checked against the official pinned contracts and actual reported buckets, explained with evidence and resolved; CodeRabbit review check passes.
Human QA: open Execution usage QA 1790792715431 in Bot mode, open its full Profile, verify the three execution rows sum to 135,491 and refresh preserves them. The private runnable link is shared in the chat.
Activity
DoodleBears commentedon Sep 30, 2026
Agent work claim
codex/local/01a0e445-0c0f-70e1-bad4-fd99c8cfaa21(not a DSH Session)codex/503-attempt-model-usagefix(qa): compare all native usage by daily role and route
DoodleBears commentedon Sep 30, 2026
Agent work outcome
codex/local/01a0e445-0c0f-70e1-bad4-fd99c8cfaa21(not a DSH Session)7bb2c0f0b84fd927fd44a6ec56ccf0b42d5c57d3Merge main before Human-approved #503 delivery
Attribute model usage by execution role and actual request route (#598)
DoodleBears commentedon Sep 30, 2026
Agent work outcome
codex/local/01a0e445-0c0f-70e1-bad4-fd99c8cfaa21bf0fdfc85aa7b7cc71b126c09df5039b509bd961.