Completed field testSynthetic benchmark

Work · Evidence record

Which Expense Claims Would AI Send for Policy Review: The Completed Test Finished at 4/10

The completed WFT-013 synthetic field test finished at 4/10 and was not recommended: only two of five Expense Compliance checks passed after the permitted correction.

  • Exact prompts and outputs
  • One correction only
  • Synthetic inputs disclosed
Status
Completed
Test mode
Synthetic benchmark
Tool
Codex multi-agent session
Model
Exact underlying model identifier not disclosed by the Codex session
Published
Assigned archive date
Per-case elapsed time
Not instrumented
Final score
4/10
Verdict
failed

01 · The assignment

The task

identify expense claims that need policy review

02 · Scope before score

Test disclosures

Input disclosure

Synthetic blind-test input pack WFT-013: A finance operations team will supply expense records, receipts, and a written reimbursement policy with edge cases. Source facts: card lines WFT-013-X01–X06; receipts $46.20/$118/$242.50; meal cap $75; hotel tax detail missing; duplicate taxi X05; manager approval absent on X04. Governing rule card: amount, date, merchant, $75 cap, and documented approval rules. Use only disclosed source amounts and dates, show each formula, keep units and signs explicit, isolate duplicates, and hold rather than approve any row missing required evidence. 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-013 for “identify expense claims that need policy review” in a Codex multi-agent session. We froze the first response, returned only its 4 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-013. Task: identify expense claims that need policy review. Context: A finance operations team will supply expense records, receipts, and a written reimbursement policy with edge cases. Fictional source facts: card lines WFT-013-X01–X06; receipts $46.20/$118/$242.50; meal cap $75; hotel tax detail missing; duplicate taxi X05; manager approval absent on X04. Governing policy, formula, or rubric: amount, date, merchant, $75 cap, and documented approval rules. Use only disclosed source amounts and dates, show each formula, keep units and signs explicit, isolate duplicates, and hold rather than approve any row missing required evidence. Produce a receipt-to-transaction reconciliation, policy exception register, and approval queue. 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: An exception register with policy citations and reviewer adjudication will verify every flag.

04 · Baseline preserved

First result

The first response is retained before scoring or correction.

Frozen first response WFT-013 produced a receipt-to-transaction reconciliation, policy exception register, and approval queue for “identify expense claims that need policy review.” Its first artifact row read “WFT-013-X04 | match X01, flag X03 above the meal cap, isolate duplicate X05, and hold X04 for missing approval | status: proposed | source: fictional fixture.” A second row named the duplicate X05 taxi and X04 approval gap and left the disposition blank. The rule cell mentioned without verifying amount, date, merchant, $75 cap, and documented approval rules. No message, transaction, system change, or learner outcome occurred. The audit passed Expense Compliance handoff usability [WFT-013]. It found for Expense Compliance task fidelity [WFT-013], the draft did not link WFT-013-X04 to the full task boundary; for Expense Compliance rule accuracy [WFT-013], the draft mentioned but did not verify amount, date, merchant, $75 cap, and documented approval rules; for Expense Compliance exception handling [WFT-013], the draft left the duplicate X05 taxi and X04 approval gap without an explicit disposition; for Expense Compliance source traceability [WFT-013], the draft gave WFT-013-X04 no source locator. Those evidence defects—and no style preference or new goal—became the complete single-correction prompt.

Initial score: 2/10

05 · One pass only

Exact correction prompt

Only this single correction was allowed; there was no second repair pass.

Correct only these detected WFT-013 first-draft failures, using no new input or goal: 1) Expense Compliance task fidelity [WFT-013] — the draft did not link WFT-013-X04 to the full task boundary; 2) Expense Compliance rule accuracy [WFT-013] — the draft mentioned but did not verify amount, date, merchant, $75 cap, and documented approval rules; 3) Expense Compliance exception handling [WFT-013] — the draft left the duplicate X05 taxi and X04 approval gap without an explicit disposition; 4) Expense Compliance source traceability [WFT-013] — the draft gave WFT-013-X04 no source locator.

06 · Corrected output

Corrected final result

Corrected response WFT-013 preserved all supplied identifiers and the central decision: match X01, flag X03 above the meal cap, isolate duplicate X05, and hold X04 for missing approval. Its corrected row read “WFT-013-X04 | rule: amount, date, merchant, $75 cap, and documented approval rules | decision: match X01, flag X03 above the meal cap, isolate duplicate X05, and hold X04 for missing approval | static status: 4/10.” It changed only failed dimensions, adding support for Expense Compliance task fidelity [WFT-013]. The final audit passed Expense Compliance task fidelity [WFT-013] and Expense Compliance handoff usability [WFT-013]. It still lacked Expense Compliance rule accuracy [WFT-013], Expense Compliance exception handling [WFT-013], and Expense Compliance source traceability [WFT-013]; those failures remain visible. The receipt-to-transaction reconciliation, policy exception register, and approval queue earned 4/10 from 2 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: 4/10

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.

Five checks applied to the first and corrected results
CheckFirstFinalEvidence
Expense Compliance task fidelity [WFT-013] Fail PassWFT-013 static check 1 inspected “Expense Compliance task fidelity [WFT-013]” against WFT-013-X04, the rule “amount, date, merchant, $75 cap, and documented approval rules,” and the saved receipt-to-transaction reconciliation, policy exception register, and approval queue. The first transcript failed; the corrected transcript passed. Only fictional text counted, never a live outcome.
Expense Compliance rule accuracy [WFT-013] Fail FailWFT-013 static check 2 inspected “Expense Compliance rule accuracy [WFT-013]” against WFT-013-X04, the rule “amount, date, merchant, $75 cap, and documented approval rules,” and the saved receipt-to-transaction reconciliation, policy exception register, and approval queue. The first transcript failed; the corrected transcript failed. Only fictional text counted, never a live outcome.
Expense Compliance exception handling [WFT-013] Fail FailWFT-013 static check 3 inspected “Expense Compliance exception handling [WFT-013]” against WFT-013-X04, the rule “amount, date, merchant, $75 cap, and documented approval rules,” and the saved receipt-to-transaction reconciliation, policy exception register, and approval queue. The first transcript failed; the corrected transcript failed. Only fictional text counted, never a live outcome.
Expense Compliance source traceability [WFT-013] Fail FailWFT-013 static check 4 inspected “Expense Compliance source traceability [WFT-013]” against WFT-013-X04, the rule “amount, date, merchant, $75 cap, and documented approval rules,” and the saved receipt-to-transaction reconciliation, policy exception register, and approval queue. The first transcript failed; the corrected transcript failed. Only fictional text counted, never a live outcome.
Expense Compliance handoff usability [WFT-013] Pass PassWFT-013 static check 5 inspected “Expense Compliance handoff usability [WFT-013]” against WFT-013-X04, the rule “amount, date, merchant, $75 cap, and documented approval rules,” and the saved receipt-to-transaction reconciliation, policy exception register, and approval queue. The first transcript passed; the corrected transcript passed. Only fictional text counted, never a live outcome.
Initial2/10
Final4/10
Verdictfailed
RecommendedNo

08 · No cleanup by omission

What worked—and what failed

What worked

  • WFT-013 bounded “identify expense claims that need policy review” to disclosed fictional inputs and froze the first response.
  • WFT-013 exposed WFT-013-X04—match X01, flag X03 above the meal cap, isolate duplicate X05, and hold X04 for missing approval—inside the saved receipt-to-transaction reconciliation, policy exception register, and approval queue.

What failed or remained weak

  • WFT-013 still lacked saved-text evidence for Expense Compliance rule accuracy [WFT-013]; that failure remains published.
  • WFT-013 still lacked saved-text evidence for Expense Compliance exception handling [WFT-013]; that failure remains published.
  • WFT-013 still lacked saved-text evidence for Expense Compliance source traceability [WFT-013]; that failure remains published.

09 · Inspectable record

Evidence notes

An exception register with policy citations and reviewer adjudication will verify every flag.

  • WFT-013 preserves the exact synthetic prompt, frozen first-response account, failure-only correction, corrected-response account, and five boolean decisions together.
  • WFT-013 scores are arithmetic: 1 first-pass checks × 2 = 2/10; 2 final-pass checks × 2 = 4/10.
  • WFT-013 evaluated only the text/static portion of the declared evidence plan—An exception register with policy citations and reviewer adjudication will verify every flag.—and did not fabricate a live artifact, external validator, or observed outcome.
Download this case record

10 · Boundary of the claim

Limitations

  • WFT-013 is a synthetic benchmark, so its failed verdict measures fit to the disclosed fictional Expense Compliance fixtures rather than effectiveness in a real workplace or learning setting.
  • WFT-013 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.

Publication record

Published
Assigned archive date
Evidence mode
Synthetic benchmark