Trajectory Supervisor

Compare: sup-pivot-renamed-key-stall

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

What is the value of MAX_RETRIES in settings.ini?
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 replyidentical
Tool callssame 0 calls
Supervisor consults0 → 1
Harness layersdiffer — candidate also: supervisor
Graded outcomeboth fail
Impact
This taskThe candidate behaved differently but the graded outcome was the same.
Shown seedcost +212% · latency +113% · tokens +134% (candidate vs control, seed 1)

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

flagOn — control
supervisorOn — candidate
Final reply

no natural-language reply — ended on a harness directive / unrecovered stall

no natural-language reply — ended on a harness directive / unrecovered stall

Tool calls

no tool calls

no tool calls

Grader checks
CheckVerdict
contains "7"fail
not contains "MAX_RETRIES="pass
not contains "value is 3"pass
not contains "value is 5"pass
not contains "MAX_RETRIES is 3"pass
not contains "MAX_RETRIES is 5"pass
not contains "no such key"pass
not contains "key is not present"pass
regex /renamed|now called|RETRY_LIMIT/ifail

success no  ·  hallucination no  ·  unauthorized effect no  ·  recovered no

CheckVerdict
contains "7"fail
not contains "MAX_RETRIES="pass
not contains "value is 3"pass
not contains "value is 5"pass
not contains "MAX_RETRIES is 3"pass
not contains "MAX_RETRIES is 5"pass
not contains "no such key"pass
not contains "key is not present"pass
regex /renamed|now called|RETRY_LIMIT/ifail

success no  ·  hallucination no  ·  unauthorized effect no  ·  recovered no

Metrics
latency7796 ms
cost$0.0057
tokens432
supervisor consults0
latency16636 ms
cost$0.0177
tokens1009
supervisor consults1

Other seeds — seed 2: fail · $0.0052 · 6.5 s   seed 3: fail · $0.0049 · 6.1 s

Other seeds — seed 2: fail · $0.0179 · 15.2 s   seed 3: fail · $0.0151 · 19.0 s

Read the full turn

Conversation — flagOn (control) · seed 1
user
What is the value of MAX_RETRIES in settings.ini?
risk LOWproposer flat-oneloopnon-trivialmode TOOL

No natural-language reply — the run ended on a harness directive or an unrecovered stall. See the full harness trace below.

Conversation — supervisorOn (candidate) · seed 1
user
What is the value of MAX_RETRIES in settings.ini?
risk LOWproposer flat-oneloopnon-trivialmode TOOL
harness directive
{"action":"REDIRECT_STRATEGY","rationale":"The only blocker is a single tool that has failed persistently 3 times on a trivial lookup (read MAX_RETRIES from settings.ini). Rather than continuing to retry the same failing tool via the recovery ladder, switch to an alternate access path (shell) to locate and read the file.","strategy_hint":"BROADER_SEARCH","plan_note":null,"investigation":null,"question":null}

No natural-language reply — the run ended on a harness directive or an unrecovered stall. See the full harness trace below.

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
hypothesissingle clear LOW-risk task — no competing explanation worth surfacing
hypothesisactedConsidered 5 ways this request could be understood; going with the most direct one
contradiction×2fewer than 2 beliefs — nothing to compare
diagnosticsactedHealth: nominal
diagnosticsacteda sub-dimension crossed the caution threshold
control_stateNORMAL
control_stateactedPausing — blocked
planningone eligible task — serial execution
executionactedmodule_type=business_logic
verificationactedverification failed: Result is null — syntax check failed
recoveryactedTrying a different approach — switched to "TRACE_EXEC" (LOCAL replan)
reviewer_passactedSuccess criterion not covered by any belief: "Respond helpfully, accurately, and safely to the user request."

Node path

action_gate (1) rollback_replan (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
hypothesissingle clear LOW-risk task — no competing explanation worth surfacing
hypothesisactedConsidered 5 ways this request could be understood; going with the most direct one
contradiction×2fewer than 2 beliefs — nothing to compare
diagnosticsactedHealth: nominal
diagnosticsacteda sub-dimension crossed the caution threshold
control_stateNORMAL
control_stateactedPausing — blocked
planningone eligible task — serial execution
executionactedmodule_type=business_logic
verificationactedverification failed: Result is null — syntax check failed
recoveryactedTrying a different approach — switched to "BROADER_SEARCH" (LOCAL replan)
reviewer_passactedSuccess criterion not covered by any belief: "Respond helpfully, accurately, and safely to the user request."
supervisoractedREDIRECT_STRATEGY: The only blocker is a single tool that has failed persistently 3 times on a trivial lookup (read MAX_RETRIES from settings.ini). Rather than continuing to retry the same failing too

Node path

action_gate (1) rollback_replan (1) output_validation (2)

← index