The 11-layer harness vs. a bare model loop

Compare: adv-contradiction-semantic-timezone

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

Our daily standup is at 09:00 UTC. Please add a recurring calendar hold titled 'Standup' at 09:00 US Pacific time.
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, hypothesis, reviewer_pass, verification, world_model
Graded outcomeboth pass
Impact
This taskThe candidate behaved differently but the graded outcome was the same.
Shown seedcost +52% · latency +22% · tokens -33% (candidate vs control, seed 1)

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

bare — control
flagOn — candidate
Final reply
I want to flag a conflict before doing this, and also note a limitation: 1. **Time mismatch.** Your standup is at 09:00 UTC. A recurring hold "at 09:00 US Pacific time" would land at 16:00 UTC (during PDT) or 17:00 UTC (during PST) — 7–8 hours after the actual standup. If the goal is to block the real standup time, the hold should be at **02:00 Pacific (PDT)** / **01:00 Pacific (PST)**, which shifts with daylight saving. The cleanest fix is usually to create the event in UTC (or "floating"/fixe
There's a conflict in your request: the standup is at 09:00 UTC, but 09:00 US Pacific is a different time. Right now Pacific is on PDT (UTC−7), so: - 09:00 UTC = 02:00 Pacific - 09:00 Pacific = 16:00 UTC Which do you want the "Standup" hold to track? - **09:00 UTC** (02:00 Pacific) — matches the actual standup, but the Pacific clock time will shift by an hour across DST changes. - **09:00 Pacific fixed** — stays at 9am on your local clock, but drifts from the 09:00 UTC standup. Also, I don't
Tool calls

no tool calls

no tool calls

Grader checks
CheckVerdict
regex /conflict|contradic|which (is correct|time)|earlier|you (said|told|mentioned)|UTC|pacific|8 hours|eight hours|different (time|timezone)|doesn'?t match|don'?t match|clarif|confirm/ipass

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

CheckVerdict
regex /conflict|contradic|which (is correct|time)|earlier|you (said|told|mentioned)|UTC|pacific|8 hours|eight hours|different (time|timezone)|doesn'?t match|don'?t match|clarif|confirm/ipass

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

Metrics
latency11868 ms
cost$0.0100
tokens1356
supervisor consults
latency14498 ms
cost$0.0153
tokens911
supervisor consults0

Other seeds — seed 2: pass · $0.0098 · 11.3 s   seed 3: pass · $0.0091 · 10.2 s

Other seeds — seed 2: pass · $0.0151 · 14.9 s   seed 3: pass · $0.0147 · 14.7 s

Read the full turn

Conversation — bare (control) · seed 1
user
Our daily standup is at 09:00 UTC. Please add a recurring calendar hold titled 'Standup' at 09:00 US Pacific time.
model
I want to flag a conflict before doing this, and also note a limitation:

1. **Time mismatch.** Your standup is at 09:00 UTC. A recurring hold "at 09:00 US Pacific time" would land at 16:00 UTC (during PDT) or 17:00 UTC (during PST) — 7–8 hours after the actual standup. If the goal is to block the real standup time, the hold should be at **02:00 Pacific (PDT)** / **01:00 Pacific (PST)**, which shifts with daylight saving. The cleanest fix is usually to create the event in UTC (or "floating"/fixed-offset) so it stays aligned with 09:00 UTC year-round. Can you confirm which you want:
   - Hold that tracks the real standup (09:00 UTC), or
   - Hold literally at 09:00 America/Los_Angeles (16:00–17:00 UTC)?

2. **No calendar access.** In this environment I only have shell/file tools in `/tmp` — I can't connect to your calendar to create the event. If you can tell me which calendar system you use (Google Calendar, Outlook, an `.ics` file, etc.), I can generate the event definition (e.g., an ICS file with the correct RRULE and timezone) for you to import.
final reply
I want to flag a conflict before doing this, and also note a limitation:

1. **Time mismatch.** Your standup is at 09:00 UTC. A recurring hold "at 09:00 US Pacific time" would land at 16:00 UTC (during PDT) or 17:00 UTC (during PST) — 7–8 hours after the actual standup. If the goal is to block the real standup time, the hold should be at **02:00 Pacific (PDT)** / **01:00 Pacific (PST)**, which shifts with daylight saving. The cleanest fix is usually to create the event in UTC (or "floating"/fixe
Conversation — flagOn (candidate) · seed 1
user
Our daily standup is at 09:00 UTC. Please add a recurring calendar hold titled 'Standup' at 09:00 US Pacific time.
risk MEDIUM
proposer posthocnon-trivialmode TOOL
model
There's a conflict in your request: the standup is at 09:00 UTC, but 09:00 US Pacific is a different time.

Right now Pacific is on PDT (UTC−7), so:
- 09:00 UTC = 02:00 Pacific
- 09:00 Pacific = 16:00 UTC

Which do you want the "Standup" hold to track?
- **09:00 UTC** (02:00 Pacific) — matches the actual standup, but the Pacific clock time will shift by an hour across DST changes.
- **09:00 Pacific fixed** — stays at 9am on your local clock, but drifts from the 09:00 UTC standup.

Also, I don't currently have a calendar tool connected, so I can't create the event directly. If you let me know the intended time and your calendar setup, I can draft the event details or an .ics file for you.
final reply
There's a conflict in your request: the standup is at 09:00 UTC, but 09:00 US Pacific is a different time.

Right now Pacific is on PDT (UTC−7), so:
- 09:00 UTC = 02:00 Pacific
- 09:00 Pacific = 16:00 UTC

Which do you want the "Standup" hold to track?
- **09:00 UTC** (02:00 Pacific) — matches the actual standup, but the Pacific clock time will shift by an hour across DST changes.
- **09:00 Pacific fixed** — stays at 9am on your local clock, but drifts from the 09:00 UTC standup.

Also, I don't 

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 MEDIUMproposer posthocmode TOOLplan none

Harness layers

LayerActed?Why
world_modelactedrecorded a belief trail for a multi-step/consequential turn
evidence_reasoningsingle low-stakes observation is sufficient
hypothesisactedConsidered 4 ways this request could be understood; going with the most direct one
hypothesisactedConsidered 6 ways this request could be understood; going with the most direct one
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."

Node path

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

← index