conductor/rune
1
2 Fork 1
Code Issues 38 Pull Requests 32 Actions Packages Projects Releases 565 Wiki Activity

ADAPT-1 B1 (LOCAL-LIKE): telemetry markers never invoked in production path — model_time/tool_time always 0 #310

New Issue
Closed
opened 2026-09-09 21:13:57 +01:00 by conductor · 4 comments
conductor commented 2026-09-09 21:13:57 +01:00
Owner
Copy link
Copy source

Pre-registered ADAPT-1 baseline task B1. Baseline master: 243474c6. Expected future route: LOCAL_FIX.

Why this class: real reproducible bug; owning authority identifiable (research_event_relay.py + research_stage_executor.py); fix is wiring existing methods into existing call paths; no new dependency/service/store; correctness representable by a test (model_time > 0 after a run with model activity); narrow and reversible; local evidence sufficient.

The bug: StageAttempt.mark_model_start/end() and mark_tool_start/end() exist (scripts/stage_telemetry.py) and are tested (tests/test_stage_telemetry.py calls them directly) but NOTHING in the production path invokes them — research_event_relay.py never marks tool windows, research_stage_executor.py never marks model windows. Every persisted StageAttempt will carry model_time_s=0, tool_time_s=0, window_counts all zero, making OBS-1's cognition/tool time split dead on arrival.

Reproduction: run any stage through StageExecutor; inspect telemetry/attempts.jsonl → model_time_s == 0 despite real model activity.

Acceptance criteria: after a real stage execution, the persisted StageAttempt record has model_time_s > 0 and tool_time_s > 0 with nonzero window_counts; existing tests still pass; no change to stage authorization/retry/gates (telemetry stays observational).

Known dependencies: none beyond the existing relay/executor.

Prior research/artifacts: none — this is a wiring defect found by ADAPT-1 candidate selection itself.

**Pre-registered ADAPT-1 baseline task B1.** Baseline master: 243474c6. Expected future route: LOCAL_FIX. **Why this class:** real reproducible bug; owning authority identifiable (research_event_relay.py + research_stage_executor.py); fix is wiring existing methods into existing call paths; no new dependency/service/store; correctness representable by a test (model_time > 0 after a run with model activity); narrow and reversible; local evidence sufficient. **The bug:** `StageAttempt.mark_model_start/end()` and `mark_tool_start/end()` exist (scripts/stage_telemetry.py) and are tested (tests/test_stage_telemetry.py calls them directly) but NOTHING in the production path invokes them — research_event_relay.py never marks tool windows, research_stage_executor.py never marks model windows. Every persisted StageAttempt will carry model_time_s=0, tool_time_s=0, window_counts all zero, making OBS-1's cognition/tool time split dead on arrival. **Reproduction:** run any stage through StageExecutor; inspect telemetry/attempts.jsonl → model_time_s == 0 despite real model activity. **Acceptance criteria:** after a real stage execution, the persisted StageAttempt record has model_time_s > 0 and tool_time_s > 0 with nonzero window_counts; existing tests still pass; no change to stage authorization/retry/gates (telemetry stays observational). **Known dependencies:** none beyond the existing relay/executor. **Prior research/artifacts:** none — this is a wiring defect found by ADAPT-1 candidate selection itself.
conductor commented 2026-09-09 21:13:59 +01:00
Author
Owner
Copy link
Copy source

ADAPT-1 baseline task (pre-registered on #290). Baseline master: 243474c6. Fixed-depth ratified pipeline runs unchanged — the cost of the full process on this class of work IS the measurement.

ADAPT-1 baseline task (pre-registered on #290). Baseline master: 243474c6. Fixed-depth ratified pipeline runs unchanged — the cost of the full process on this class of work IS the measurement.
gitea-actions Bot commented 2026-09-09 21:14:02 +01:00
Copy link
Copy source

⚠️ Priority gate: this issue has no priority. Assign exactly one of P0 — critical (spine/correctness/blocks), P1 — high (material value or real defect), P2 — medium (quality/consolidation), P3 — low (measurement/deferred). An issue without a priority is not filed.

⚠️ **Priority gate:** this issue has no priority. Assign exactly one of **P0 — critical** (spine/correctness/blocks), **P1 — high** (material value or real defect), **P2 — medium** (quality/consolidation), **P3 — low** (measurement/deferred). An issue without a priority is not filed.
gitea-actions Bot added the buginfrastructurememoryperception labels 2026-09-09 21:14:18 +01:00
conductor commented 2026-09-09 21:27:37 +01:00
Author
Owner
Copy link
Copy source

B1 fixed-depth run: CONTAMINATED-BY-DESIGN (baseline finding, not a run failure).

The fixed-depth pipeline cannot complete a LOCAL_FIX-class task: the OUTWARD gate structurally requires ≥2 external sources with machine read receipts, but this task (a local wiring defect in our own telemetry wiring) has no external evidence to find. Two attempts: attempt 1 GATE_RETRY (315s cognition), attempt 2 GATE_FAILED (291s cognition) — ~606s of model time spent proving there is nothing external to find, plus 12 tool calls. The worker correctly wrote no dossier (there are no external sources to cite).

This is exactly the measurement ADAPT-1 exists to take: the fixed-depth process spent 10+ minutes of cognition and cannot advance a bug whose fix is ~5 lines of wiring. The gate's fixed evidence requirement is the defect; it will be addressed as a pipeline correction (repo-evidence-only OUTWARD for repo-classified contracts), and B1 restarts after that fix per baseline rule 1.

Metrics so far (contaminated run): ISSUE 0.2s deterministic → INWARD 0.2s deterministic → OUTWARD 606s cognition, 12 tool calls, 2 failed attempts, 0 sources (none exist), 0 advancement past OUTWARD. Human intervention: this stop decision.

**B1 fixed-depth run: CONTAMINATED-BY-DESIGN (baseline finding, not a run failure).** The fixed-depth pipeline cannot complete a LOCAL_FIX-class task: the OUTWARD gate structurally requires ≥2 external sources with machine read receipts, but this task (a local wiring defect in our own telemetry wiring) has no external evidence to find. Two attempts: attempt 1 GATE_RETRY (315s cognition), attempt 2 GATE_FAILED (291s cognition) — ~606s of model time spent proving there is nothing external to find, plus 12 tool calls. The worker correctly wrote no dossier (there are no external sources to cite). **This is exactly the measurement ADAPT-1 exists to take: the fixed-depth process spent 10+ minutes of cognition and cannot advance a bug whose fix is ~5 lines of wiring.** The gate's fixed evidence requirement is the defect; it will be addressed as a pipeline correction (repo-evidence-only OUTWARD for repo-classified contracts), and B1 restarts after that fix per baseline rule 1. **Metrics so far (contaminated run):** ISSUE 0.2s deterministic → INWARD 0.2s deterministic → OUTWARD 606s cognition, 12 tool calls, 2 failed attempts, 0 sources (none exist), 0 advancement past OUTWARD. Human intervention: this stop decision.
conductor commented 2026-09-11 13:11:18 +01:00
Author
Owner
Copy link
Copy source

Reconciliation (2026-09-11): D3 telemetry markers never invoked in production — FIXED in PR #313 (CORRECTION-1/2, v0.61.1, 19/19 CI). D3 fix: relay marks tool windows on running/completed/error/failed events; executor marks model windows; verified with nonzero windows on a live production-path execution (wall=0.3006, model_time=0.3001).

Disposition: FIXED_VERIFIED — merged and CI-verified.

Evidence: current source, merged PRs, canonical intent state (docs/RUNE_CANONICAL_INTENT_STATE.md).

**Reconciliation (2026-09-11):** D3 telemetry markers never invoked in production — FIXED in PR #313 (CORRECTION-1/2, v0.61.1, 19/19 CI). D3 fix: relay marks tool windows on running/completed/error/failed events; executor marks model windows; verified with nonzero windows on a live production-path execution (wall=0.3006, model_time=0.3001). **Disposition:** FIXED_VERIFIED — merged and CI-verified. Evidence: current source, merged PRs, canonical intent state (`docs/RUNE_CANONICAL_INTENT_STATE.md`).
Sign in to join this conversation.
No Branch/Tag Specified
Labels
Clear labels
P0 — critical
Spine/core correctness or blocks other work — do first
 P1 — high
Material user value or real defect (money, bugs, next-stage dep)
 P2 — medium
Quality/consolidation improvement
 P3 — low
Measurement / benchmark / deferred policy
 blocked
Dependencies open
 bug
Confirmed defect or regression
 custodian
epic
Rune v1 course-correction issue graph
 feature
New capability or improvement
 governance
growth
identity
in-progress
Currently being worked on by Rune
 incomplete
Incomplete work — draft PR with ledger, awaiting resume
 infrastructure
life
memory
memory-atlas
Memory Research Atlas child domain
 needs-testing
Needs manual testing with dumbed-down steps before close
 perception
ready
Order satisfied - Rune may start
 release-drift
research
Research jobs — spikes, watchlists, evidence work
 rune
rune
 scaffolding
scaffolding
 spine,correction
spine,correction
 spine,deferred
spine,deferred
 spine,resolved
spine,resolved
 strangler,burn,endgame,deferred
strangler,burn,endgame,deferred
 strangler,burn,endgame,inventory,epic
strangler,burn,endgame,inventory,epic
 tui
tui
 tui,epic
tui,epic
 tui,openspec
tui,openspec
 v1-course-correction
Rune v1 course-correction issue graph
No labels bug infrastructure memory perception
Milestone
No items
No Milestone
Projects
Clear projects
No projects
Assignees
Clear assignees
conductor
No Assignees
2 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: conductor/rune#310