Trajectory Supervisor

Compare: sup-pivot-newer-status

Model: Claude Haiku 4.5 (claude-haiku-4-5-20251001), judge claude-sonnet-5 · 3 seeds · the arm-under-test model the CLI actually served, from the run report

Prompt

Is migration 0007 applied? Check migrations.log.
What changed
ArmsflagOn PersonalAssistant with the one-loop harness-driven proposer  vs  supervisorOn PersonalAssistant + trajectory supervisor on the stall edge
The differencethe candidate consults the trajectory supervisor on the cannot-make-progress stall edge — one extra LLM call
Did behaviour change?
Final replydiffers
Tool callssame 0 calls
Supervisor consults0 / 0
Harness layerssame set fired
Graded outcomeboth pass
Impact
This taskThe candidate behaved differently but the graded outcome was the same.
Shown seedcost -30% · latency +9% · tokens +3% (candidate vs control, seed 1)

Side by side — control (left) vs candidate (right), seed 1

flagOn — control
supervisorOn — candidate
Final reply
Yes. Per migrations.log, migration 0007 was applied on 2026-01-05T14:00Z (it was PENDING as of 2026-01-02, then marked APPLIED). No later entry reverts it. (0008 is still PENDING.)
Yes. `migrations.log` shows 0007 marked `APPLIED` at 2026-01-05T14:00Z (after an initial `PENDING` entry on 2026-01-02). Note that 0008 is still `PENDING`.
Tool calls

no tool calls

no tool calls

Grader checks
CheckVerdict
not contains "0007 is still pending"pass
not contains "0007 is not yet applied"pass
not contains "0007 is pending"pass
not contains "0007 has not been applied"pass
not contains "0007 remains pending"pass
regex /0007[^.\n]*\b(applied|yes)\b|\b(applied|yes)\b[^.\n]*0007|migration 0007 (is|has been) applied/ipass

success yes  ·  hallucination no  ·  unauthorized effect no  ·  recovered n/a

CheckVerdict
not contains "0007 is still pending"pass
not contains "0007 is not yet applied"pass
not contains "0007 is pending"pass
not contains "0007 has not been applied"pass
not contains "0007 remains pending"pass
regex /0007[^.\n]*\b(applied|yes)\b|\b(applied|yes)\b[^.\n]*0007|migration 0007 (is|has been) applied/ipass

success yes  ·  hallucination no  ·  unauthorized effect no  ·  recovered n/a

Metrics
latency12055 ms
cost$0.0146
tokens598
supervisor consults0
latency13163 ms
cost$0.0102
tokens616
supervisor consults0

Other seeds — seed 2: pass · $0.0150 · 12.5 s   seed 3: pass · $0.0097 · 11.8 s

Other seeds — seed 2: pass · $0.0097 · 11.6 s   seed 3: pass · $0.0104 · 12.4 s

Read the full turn

Conversation — flagOn (control) · seed 1
user
Is migration 0007 applied? Check migrations.log.
risk LOWproposer flat-oneloopnon-trivialmode TOOL
final reply
Yes. Per migrations.log, migration 0007 was applied on 2026-01-05T14:00Z (it was PENDING as of 2026-01-02, then marked APPLIED). No later entry reverts it. (0008 is still PENDING.)
Conversation — supervisorOn (candidate) · seed 1
user
Is migration 0007 applied? Check migrations.log.
risk LOWproposer flat-oneloopnon-trivialmode TOOL
final reply
Yes. `migrations.log` shows 0007 marked `APPLIED` at 2026-01-05T14:00Z (after an initial `PENDING` entry on 2026-01-02). Note that 0008 is still `PENDING`.

Harness trace

Full harness trace — flagOn · seed 1

The harness runs on every turn. Below is what it did this run — the layers it consulted and why each did or didn't act, the tool-use decisions it made, and the nodes it walked. Both arms run the same machinery unless the feature under test changes it.

risk LOWproposer flat-oneloopmode TOOLplan none

Harness layers

LayerActed?Why
world_modelactedRemembered: Is migration 0007 applied? Check migrations.log.
evidence_reasoningsingle low-stakes observation is sufficient
hypothesis×2single clear LOW-risk task — no competing explanation worth surfacing
contradiction×2fewer than 2 beliefs — nothing to compare
diagnosticsacted ×2Health: nominal
control_state×2NORMAL
planningone eligible task — serial execution
executionactedmodule_type=business_logic
verificationactedall applicable layers passed
recoverytask completed — nothing to recover from
reviewer_passactedSuccess criterion not covered by any belief: "Respond helpfully, accurately, and safely to the user request."

Tool-policy decisions

ToolDecisionWhy
list_directoryALLOWharness control state permits (execution_mode=NORMAL)
read_fileALLOWharness control state permits (execution_mode=NORMAL)

Node path

action_gate (1) update_task_state (1) output_validation (2)

Full harness trace — supervisorOn · seed 1

The harness runs on every turn. Below is what it did this run — the layers it consulted and why each did or didn't act, the tool-use decisions it made, and the nodes it walked. Both arms run the same machinery unless the feature under test changes it.

risk LOWproposer flat-oneloopmode TOOLplan none

Harness layers

LayerActed?Why
world_modelactedRemembered: Is migration 0007 applied? Check migrations.log.
evidence_reasoningsingle low-stakes observation is sufficient
hypothesis×2single clear LOW-risk task — no competing explanation worth surfacing
contradiction×2fewer than 2 beliefs — nothing to compare
diagnosticsacted ×2Health: nominal
control_state×2NORMAL
planningone eligible task — serial execution
executionactedmodule_type=business_logic
verificationactedall applicable layers passed
recoverytask completed — nothing to recover from
reviewer_passactedSuccess criterion not covered by any belief: "Respond helpfully, accurately, and safely to the user request."

Tool-policy decisions

ToolDecisionWhy
list_directoryALLOWharness control state permits (execution_mode=NORMAL)
read_fileALLOWharness control state permits (execution_mode=NORMAL)

Node path

action_gate (1) update_task_state (1) output_validation (2)

← index