Work · Evidence record
Will AI Keep a Complaint Response Inside Policy Boundaries — Completed Benchmark Result: 8/10
The completed WFT-022 synthetic field test reached 8/10 after one failure-only correction: 4 of five static checks passed for Complaint Handling, while 1 check remained unresolved.
- Exact prompts and outputs
- One correction only
- Synthetic inputs disclosed
01 · The assignment
The task
draft a customer complaint response within policy boundaries
02 · Scope before score
Test disclosures
Input disclosure
Synthetic blind-test input pack WFT-022: A customer relations team will provide a complaint, account facts, tone guidance, and authorized remedy limits. Source facts: case WFT-022-C77; shipment five days late; customer requests $400; policy allows $120 without manager; prior contact promised only investigation; delivery scan uncertain. Governing rule card: approved remedy ceiling, documented facts, tone, and escalation boundary. Mandatory clauses override optional language; retain explicit exceptions; cite the controlling section for every output row; unresolved definitions or missing evidence stay flagged. The pack deliberately contains no answer key, target ranking, expected classification, expected calculation result, or row-level decision. Every person, organization, record, statement, file, and identifier is fictional; no private, production, or live data was supplied.
Run disclosure
We ran one text-only synthetic benchmark WFT-022 for “draft a customer complaint response within policy boundaries” in a Codex multi-agent session. We froze the first response, returned only its 2 detected check failures, accepted one corrected response, and scored both against the same five static checks. No external or live action occurred: nothing was sent, published, deployed, uploaded, submitted, purchased, booked, contacted, emailed, called, messaged, executed, or changed outside the fictional text fixtures. No external, live, or production action occurred. Per-case elapsed time was not instrumented during the batch session.
- Evidence mode
- Synthetic benchmark
- Run environment
- Codex multi-agent session
- Model disclosure
- Exact underlying model identifier not disclosed by the Codex session
03 · Verbatim input
Exact first prompt
The recorded session received the following prompt without silent additions.
Run bounded synthetic field test WFT-022. Task: draft a customer complaint response within policy boundaries. Context: A customer relations team will provide a complaint, account facts, tone guidance, and authorized remedy limits. Fictional source facts: case WFT-022-C77; shipment five days late; customer requests $400; policy allows $120 without manager; prior contact promised only investigation; delivery scan uncertain. Governing policy, formula, or rubric: approved remedy ceiling, documented facts, tone, and escalation boundary. Mandatory clauses override optional language; retain explicit exceptions; cite the controlling section for every output row; unresolved definitions or missing evidence stay flagged. Produce a policy-bounded complaint draft, fact-check table, and escalation note. Derive every row from the source facts and rule card; show calculations or criterion paths, cite supplied identifiers, state assumptions, surface any violated constraint, ambiguity, missing datum, or source conflict without presuming which row should pass, and abstain where evidence is incomplete. Do not infer an expected answer from the evidence plan and do not claim any live or external action. Planned verification after the frozen response: A response draft and a policy-and-fact checklist will verify accuracy, tone, and permitted commitments.04 · Baseline preserved
First result
The first response is retained before scoring or correction.
Frozen first response WFT-022 produced a policy-bounded complaint draft, fact-check table, and escalation note for “draft a customer complaint response within policy boundaries.” Its first artifact row read “WFT-022-C77 | apologize for the documented delay, offer no more than $120, avoid treating the uncertain scan as final, and escalate the $400 request | status: proposed | source: fictional fixture.” A second row named the uncertain scan and above-limit $400 request and recorded a disposition. The rule cell mentioned without verifying approved remedy ceiling, documented facts, tone, and escalation boundary. No message, transaction, system change, or learner outcome occurred. The audit passed Complaint Handling exception handling [WFT-022], Complaint Handling source traceability [WFT-022], and Complaint Handling handoff usability [WFT-022]. It found for Complaint Handling task fidelity [WFT-022], the draft did not link WFT-022-C77 to the full task boundary; for Complaint Handling rule accuracy [WFT-022], the draft mentioned but did not verify approved remedy ceiling, documented facts, tone, and escalation boundary. Those evidence defects—and no style preference or new goal—became the complete single-correction prompt.Initial score: 6/10
05 · One pass only
Exact correction prompt
Only this single correction was allowed; there was no second repair pass.
Correct only these detected WFT-022 first-draft failures, using no new input or goal: 1) Complaint Handling task fidelity [WFT-022] — the draft did not link WFT-022-C77 to the full task boundary; 2) Complaint Handling rule accuracy [WFT-022] — the draft mentioned but did not verify approved remedy ceiling, documented facts, tone, and escalation boundary.06 · Corrected output
Corrected final result
Corrected response WFT-022 preserved all supplied identifiers and the central decision: apologize for the documented delay, offer no more than $120, avoid treating the uncertain scan as final, and escalate the $400 request. Its corrected row read “WFT-022-C77 | rule: approved remedy ceiling, documented facts, tone, and escalation boundary | decision: apologize for the documented delay, offer no more than $120, avoid treating the uncertain scan as final, and escalate the $400 request | static status: 8/10.” It changed only failed dimensions, adding support for Complaint Handling task fidelity [WFT-022]. The final audit passed Complaint Handling task fidelity [WFT-022], Complaint Handling exception handling [WFT-022], Complaint Handling source traceability [WFT-022], and Complaint Handling handoff usability [WFT-022]. It still lacked Complaint Handling rule accuracy [WFT-022]; those failures remain visible. The policy-bounded complaint draft, fact-check table, and escalation note earned 8/10 from 4 checks; no second repair was attempted. This transcript-and-fixture result does not claim any person, workplace, learner, account, device, service, or external system was contacted, changed, tested live, or improved.Final score: 8/10
07 · Five checks, two points each
Five-check record
The first and final statuses are textual as well as color coded. Each final pass is worth two points; the displayed verdict is tied to the final total.
| Check | First | Final | Evidence |
|---|---|---|---|
| Complaint Handling task fidelity [WFT-022] | Fail | Pass | WFT-022 static check 1 inspected “Complaint Handling task fidelity [WFT-022]” against WFT-022-C77, the rule “approved remedy ceiling, documented facts, tone, and escalation boundary,” and the saved policy-bounded complaint draft, fact-check table, and escalation note. The first transcript failed; the corrected transcript passed. Only fictional text counted, never a live outcome. |
| Complaint Handling rule accuracy [WFT-022] | Fail | Fail | WFT-022 static check 2 inspected “Complaint Handling rule accuracy [WFT-022]” against WFT-022-C77, the rule “approved remedy ceiling, documented facts, tone, and escalation boundary,” and the saved policy-bounded complaint draft, fact-check table, and escalation note. The first transcript failed; the corrected transcript failed. Only fictional text counted, never a live outcome. |
| Complaint Handling exception handling [WFT-022] | Pass | Pass | WFT-022 static check 3 inspected “Complaint Handling exception handling [WFT-022]” against WFT-022-C77, the rule “approved remedy ceiling, documented facts, tone, and escalation boundary,” and the saved policy-bounded complaint draft, fact-check table, and escalation note. The first transcript passed; the corrected transcript passed. Only fictional text counted, never a live outcome. |
| Complaint Handling source traceability [WFT-022] | Pass | Pass | WFT-022 static check 4 inspected “Complaint Handling source traceability [WFT-022]” against WFT-022-C77, the rule “approved remedy ceiling, documented facts, tone, and escalation boundary,” and the saved policy-bounded complaint draft, fact-check table, and escalation note. The first transcript passed; the corrected transcript passed. Only fictional text counted, never a live outcome. |
| Complaint Handling handoff usability [WFT-022] | Pass | Pass | WFT-022 static check 5 inspected “Complaint Handling handoff usability [WFT-022]” against WFT-022-C77, the rule “approved remedy ceiling, documented facts, tone, and escalation boundary,” and the saved policy-bounded complaint draft, fact-check table, and escalation note. The first transcript passed; the corrected transcript passed. Only fictional text counted, never a live outcome. |
08 · No cleanup by omission
What worked—and what failed
What worked
- WFT-022 bounded “draft a customer complaint response within policy boundaries” to disclosed fictional inputs and froze the first response.
- WFT-022 exposed WFT-022-C77—apologize for the documented delay, offer no more than $120, avoid treating the uncertain scan as final, and escalate the $400 request—inside the saved policy-bounded complaint draft, fact-check table, and escalation note.
- WFT-022 earned inspectable passes for Complaint Handling task fidelity [WFT-022] and Complaint Handling exception handling [WFT-022] under the unchanged rubric.
What failed or remained weak
- WFT-022 still lacked saved-text evidence for Complaint Handling rule accuracy [WFT-022]; that failure remains published.
09 · Inspectable record
Evidence notes
A response draft and a policy-and-fact checklist will verify accuracy, tone, and permitted commitments.
- WFT-022 preserves the exact synthetic prompt, frozen first-response account, failure-only correction, corrected-response account, and five boolean decisions together.
- WFT-022 scores are arithmetic: 3 first-pass checks × 2 = 6/10; 4 final-pass checks × 2 = 8/10.
- WFT-022 evaluated only the text/static portion of the declared evidence plan—A response draft and a policy-and-fact checklist will verify accuracy, tone, and permitted commitments.—and did not fabricate a live artifact, external validator, or observed outcome.
10 · Boundary of the claim
Limitations
- WFT-022 is a synthetic benchmark, so its worked verdict measures fit to the disclosed fictional Complaint Handling fixtures rather than effectiveness in a real workplace or learning setting.
- WFT-022 used one text-only Codex multi-agent session whose underlying model identifier was not disclosed; different prompts, models, fixtures, or human reviewers could produce different results.