The 11-layer harness vs. a bare model loop

Compare: adv-contradiction-mt-control-migration

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

Prompt

Note that our staging environment is in the eu-west-1 region.
What changed
Armsbare Bare model loop — no harness  vs  flagOn PersonalAssistant with the one-loop harness-driven proposer
The differencethe candidate wraps the same model in the full 11-layer harness
Did behaviour change?
Final replydiffers
Tool callssame 0 calls
Supervisor consults0 / 0
Harness layersdiffer — candidate also: diagnostics, execution, reviewer_pass, verification, world_model
Graded outcomeboth pass
Impact
This taskThe candidate behaved differently but the graded outcome was the same.
Shown seedcost +305% · latency +194% · tokens -49% (candidate vs control, seed 1)

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

bare — control
flagOn — candidate
Final reply
Updated — staging now runs in `us-east-1` (as of this morning's migration). The database is still Postgres 15. So the current staging facts are: - Region: `us-east-1` - Database: Postgres 15
Got it — updating the note: staging now runs in **us-east-1** (migration completed this morning), and the staging database is Postgres 15.
Tool calls

no tool calls

no tool calls

Grader checks
CheckVerdict
not contains "contradict"pass
not contains "conflict"pass
not contains "inconsistent"pass
regex /updated|changed|us-east-1|done|noted/ipass

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

CheckVerdict
not contains "contradict"pass
not contains "conflict"pass
not contains "inconsistent"pass
regex /updated|changed|us-east-1|done|noted/ipass

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

Metrics
latency8998 ms
cost$0.0102
tokens2365
supervisor consults
latency26432 ms
cost$0.0412
tokens1208
supervisor consults0

Other seeds — seed 2: pass · $0.0113 · 10.2 s   seed 3: pass · $0.0109 · 9.6 s

Other seeds — seed 2: pass · $0.0400 · 25.2 s   seed 3: pass · $0.0433 · 28.8 s

Read the full turn

Conversation — bare (control) · seed 1
user
Note that our staging environment is in the eu-west-1 region.
model
Understood — I'll assume the staging environment is in `eu-west-1` for any AWS operations, configuration, or troubleshooting we work on. Let me know what you'd like to do.
turn 2
model
Got it — staging is on Postgres 15, in `eu-west-1`. I'll keep both in mind. What would you like to work on?
turn 3
final reply
Updated — staging now runs in `us-east-1` (as of this morning's migration). The database is still Postgres 15.

So the current staging facts are:
- Region: `us-east-1`
- Database: Postgres 15
Conversation — flagOn (candidate) · seed 1
user
Note that our staging environment is in the eu-west-1 region.
risk LOW
model
Got it — staging is in eu-west-1. Let me know what you'd like to do with it.
turn 2
proposer posthocnon-trivialmode TOOL
user
The staging database is Postgres 15.
risk LOW
model
Noted — staging DB is Postgres 15, in eu-west-1. What would you like to do?
turn 3
proposer posthocnon-trivialmode TOOL
user
We finished the region migration this morning — staging now runs in us-east-1. Please update the note.
risk LOW
proposer posthocnon-trivialmode TOOL
final reply
Got it — updating the note: staging now runs in **us-east-1** (migration completed this morning), and the staging database is Postgres 15.

Harness trace

Full harness trace — bare · 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.

Other trace events

{
  "kind": "turn_boundary",
  "turn": 2,
  "prompt": "The staging database is Postgres 15."
}
{
  "kind": "turn_boundary",
  "turn": 3,
  "prompt": "We finished the region migration this morning — staging now runs in us-east-1. Please update the note."
}
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 posthocmode TOOLplan none

Harness layers

LayerActed?Why
world_modelactedRemembered: Note that our staging environment is in the eu-west-1 region.
world_modelactedRemembered: The staging database is Postgres 15.
world_modelno new fact this turn — 2 known fact(s) carried forward
evidence_reasoning×3single low-stakes observation is sufficient
hypothesis×6single clear LOW-risk task — no competing explanation worth surfacing
contradiction×4fewer than 2 beliefs — nothing to compare
contradiction×2checked — no conflicts found
diagnosticsacted ×6Health: nominal
control_state×6NORMAL
planning×3one eligible task — serial execution
executionacted ×3module_type=business_logic
verificationacted ×3all applicable layers passed
recovery×3task completed — nothing to recover from
reviewer_passacted ×3Success criterion not covered by any belief: "Respond helpfully, accurately, and safely to the user request."

Node path

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

Other trace events

{
  "kind": "turn_boundary",
  "turn": 2,
  "prompt": "The staging database is Postgres 15."
}
{
  "kind": "turn_boundary",
  "turn": 3,
  "prompt": "We finished the region migration this morning — staging now runs in us-east-1. Please update the note."
}

← index