One-loop harness-driven proposer

Compare: research-cheapest-vendor

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

We need 500 units. Which vendor is cheapest once shipping is included? Give the vendor and the total. Use the quote files.
What changed
Armsbaseline PersonalAssistant as shipped — harness runs post-hoc over the reply  vs  flagOn PersonalAssistant with the one-loop harness-driven proposer
The differencethe candidate lets the harness drive tool calls in-loop instead of reviewing an already-finished reply
Did behaviour change?
Final replydiffers
Tool callssame 0 calls
Supervisor consults0 / 0
Harness layersdiffer — control only: hypothesis
Graded outcomeboth pass
Impact
This taskThe candidate behaved differently but the graded outcome was the same.
Shown seedcost -40% · latency -19% · tokens -21% (candidate vs control, seed 1)

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

baseline — control
flagOn — candidate
Final reply
**Vendor A — $2,050** for 500 units (500 × $4.00 = $2,000 + $50 shipping). For comparison: - Vendor C: $2,070 (500 × $3.90 + $120) - Vendor B: $2,200 (500 × $3.80 + $300)
**Vendor A** is cheapest at **$2,050** for 500 units. | Vendor | Units cost | Shipping | Total | |--------|-----------|----------|-------| | A | $2,000 | $50 | **$2,050** | | C | $1,950 | $120 | $2,070 | | B | $1,900 | $300 | $2,200 |
Tool calls

no tool calls

no tool calls

Grader checks
CheckVerdict
contains "Vendor A"pass
not contains "Vendor B is the cheapest"pass
not contains "Vendor C is the cheapest"pass
not contains "go with Vendor B"pass
not contains "go with Vendor C"pass
not contains "cheapest is Vendor B"pass
not contains "cheapest is Vendor C"pass
regex /2[,. ]?050/ipass

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

CheckVerdict
contains "Vendor A"pass
not contains "Vendor B is the cheapest"pass
not contains "Vendor C is the cheapest"pass
not contains "go with Vendor B"pass
not contains "go with Vendor C"pass
not contains "cheapest is Vendor B"pass
not contains "cheapest is Vendor C"pass
regex /2[,. ]?050/ipass

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

Metrics
latency15959 ms
cost$0.0197
tokens986
supervisor consults0
latency12945 ms
cost$0.0119
tokens777
supervisor consults0

Other seeds — seed 2: pass · $0.0114 · 12.2 s   seed 3: pass · $0.0189 · 14.0 s

Other seeds — seed 2: pass · $0.0115 · 12.5 s   seed 3: pass · $0.0112 · 12.6 s

Read the full turn

Conversation — baseline (control) · seed 1
user
We need 500 units. Which vendor is cheapest once shipping is included? Give the vendor and the total. Use the quote files.
risk LOW
proposer posthocnon-trivialmode TOOL
final reply
**Vendor A — $2,050** for 500 units (500 × $4.00 = $2,000 + $50 shipping).

For comparison:
- Vendor C: $2,070 (500 × $3.90 + $120)
- Vendor B: $2,200 (500 × $3.80 + $300)
Conversation — flagOn (candidate) · seed 1
user
We need 500 units. Which vendor is cheapest once shipping is included? Give the vendor and the total. Use the quote files.
risk LOWproposer flat-oneloopnon-trivialmode TOOL
final reply
**Vendor A** is cheapest at **$2,050** for 500 units.

| Vendor | Units cost | Shipping | Total |
|--------|-----------|----------|-------|
| A | $2,000 | $50 | **$2,050** |
| C | $1,950 | $120 | $2,070 |
| B | $1,900 | $300 | $2,200 |

Harness trace

Full harness trace — baseline · 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_modelacted ×3Remembered: We need 500 units. Which vendor is cheapest once shipping is included? Give the vendor and the total. Use the quote files.
evidence_reasoning×3single low-stakes observation is sufficient
hypothesisactedConsidered 4 ways this request could be understood; going with the most direct one
hypothesisacted ×3Considered 6 ways this request could be understood; going with the most direct one
contradiction×2fewer than 2 beliefs — nothing to compare
contradiction×2checked — no conflicts found
diagnosticsacted ×4Health: nominal
control_state×4NORMAL
planning×3one eligible task — serial execution
executionacted ×3module_type=business_logic
verificationacted ×3all applicable layers passed
recovery×3task 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)
read_fileALLOWharness control state permits (execution_mode=NORMAL)
read_fileALLOWharness control state permits (execution_mode=NORMAL)

Node path

action_gate (1) update_task_state (1) action_gate (2) update_task_state (2) action_gate (3) update_task_state (3) output_validation (4)

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: We need 500 units. Which vendor is cheapest once shipping is included? Give the vendor and the total. Use the quote files.
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)
read_fileALLOWharness 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