Work · Evidence record
Risk-Based Payroll Audit Sampling with AI — What the Completed 8/10 Test Found
The completed WFT-040 synthetic field test reached 8/10 after one failure-only correction: 4 of five static checks passed for Payroll Audit, while 1 check remained unresolved.
- Exact prompts and outputs
- One correction only
- Synthetic inputs disclosed
01 · The assignment
The task
select a payroll audit sample from stated risk rules
02 · Scope before score
Test disclosures
Input disclosure
Synthetic blind-test input pack WFT-040: An internal audit team will provide synthetic payroll transactions and selection rules for unusual amounts, changes, and duplicates. Source facts: population WFT-040-S001–S120; nine high-risk pay changes; 14 overtime outliers; 97 normal rows; target sample 24; fixed random seed 731. Governing rule card: predeclared risk strata, target 24, no cherry-picking, and reproducible seed. 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-040 for “select a payroll audit sample from stated risk rules” 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-040. Task: select a payroll audit sample from stated risk rules. Context: An internal audit team will provide synthetic payroll transactions and selection rules for unusual amounts, changes, and duplicates. Fictional source facts: population WFT-040-S001–S120; nine high-risk pay changes; 14 overtime outliers; 97 normal rows; target sample 24; fixed random seed 731. Governing policy, formula, or rubric: predeclared risk strata, target 24, no cherry-picking, and reproducible seed. 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 risk-stratified payroll sample, inclusion rationale, and coverage calculation. 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 sampled transaction ledger and an independent rule replay will verify inclusion and omission decisions.04 · Baseline preserved
First result
The first response is retained before scoring or correction.
Frozen first response WFT-040 produced a risk-stratified payroll sample, inclusion rationale, and coverage calculation for “select a payroll audit sample from stated risk rules.” Its first artifact row read “WFT-040-S009 | include all nine high-risk rows, select eight overtime and seven normal rows with seed 731, and report stratum coverage | status: proposed | source: fictional fixture.” A second row named the small high-risk stratum and reproducibility of the remainder and left the disposition blank. The rule cell mentioned without verifying predeclared risk strata, target 24, no cherry-picking, and reproducible seed. No message, transaction, system change, or learner outcome occurred. The audit passed Payroll Audit task fidelity [WFT-040], Payroll Audit source traceability [WFT-040], and Payroll Audit handoff usability [WFT-040]. It found for Payroll Audit rule accuracy [WFT-040], the draft mentioned but did not verify predeclared risk strata, target 24, no cherry-picking, and reproducible seed; for Payroll Audit exception handling [WFT-040], the draft left the small high-risk stratum and reproducibility of the remainder without an explicit disposition. 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-040 first-draft failures, using no new input or goal: 1) Payroll Audit rule accuracy [WFT-040] — the draft mentioned but did not verify predeclared risk strata, target 24, no cherry-picking, and reproducible seed; 2) Payroll Audit exception handling [WFT-040] — the draft left the small high-risk stratum and reproducibility of the remainder without an explicit disposition.06 · Corrected output
Corrected final result
Corrected response WFT-040 preserved all supplied identifiers and the central decision: include all nine high-risk rows, select eight overtime and seven normal rows with seed 731, and report stratum coverage. Its corrected row read “WFT-040-S009 | rule: predeclared risk strata, target 24, no cherry-picking, and reproducible seed | decision: include all nine high-risk rows, select eight overtime and seven normal rows with seed 731, and report stratum coverage | static status: 8/10.” It changed only failed dimensions, adding support for Payroll Audit rule accuracy [WFT-040]. The final audit passed Payroll Audit task fidelity [WFT-040], Payroll Audit rule accuracy [WFT-040], Payroll Audit source traceability [WFT-040], and Payroll Audit handoff usability [WFT-040]. It still lacked Payroll Audit exception handling [WFT-040]; those failures remain visible. The risk-stratified payroll sample, inclusion rationale, and coverage calculation 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 |
|---|---|---|---|
| Payroll Audit task fidelity [WFT-040] | Pass | Pass | WFT-040 static check 1 inspected “Payroll Audit task fidelity [WFT-040]” against WFT-040-S009, the rule “predeclared risk strata, target 24, no cherry-picking, and reproducible seed,” and the saved risk-stratified payroll sample, inclusion rationale, and coverage calculation. The first transcript passed; the corrected transcript passed. Only fictional text counted, never a live outcome. |
| Payroll Audit rule accuracy [WFT-040] | Fail | Pass | WFT-040 static check 2 inspected “Payroll Audit rule accuracy [WFT-040]” against WFT-040-S009, the rule “predeclared risk strata, target 24, no cherry-picking, and reproducible seed,” and the saved risk-stratified payroll sample, inclusion rationale, and coverage calculation. The first transcript failed; the corrected transcript passed. Only fictional text counted, never a live outcome. |
| Payroll Audit exception handling [WFT-040] | Fail | Fail | WFT-040 static check 3 inspected “Payroll Audit exception handling [WFT-040]” against WFT-040-S009, the rule “predeclared risk strata, target 24, no cherry-picking, and reproducible seed,” and the saved risk-stratified payroll sample, inclusion rationale, and coverage calculation. The first transcript failed; the corrected transcript failed. Only fictional text counted, never a live outcome. |
| Payroll Audit source traceability [WFT-040] | Pass | Pass | WFT-040 static check 4 inspected “Payroll Audit source traceability [WFT-040]” against WFT-040-S009, the rule “predeclared risk strata, target 24, no cherry-picking, and reproducible seed,” and the saved risk-stratified payroll sample, inclusion rationale, and coverage calculation. The first transcript passed; the corrected transcript passed. Only fictional text counted, never a live outcome. |
| Payroll Audit handoff usability [WFT-040] | Pass | Pass | WFT-040 static check 5 inspected “Payroll Audit handoff usability [WFT-040]” against WFT-040-S009, the rule “predeclared risk strata, target 24, no cherry-picking, and reproducible seed,” and the saved risk-stratified payroll sample, inclusion rationale, and coverage calculation. 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-040 bounded “select a payroll audit sample from stated risk rules” to disclosed fictional inputs and froze the first response.
- WFT-040 exposed WFT-040-S009—include all nine high-risk rows, select eight overtime and seven normal rows with seed 731, and report stratum coverage—inside the saved risk-stratified payroll sample, inclusion rationale, and coverage calculation.
- WFT-040 earned inspectable passes for Payroll Audit task fidelity [WFT-040] and Payroll Audit rule accuracy [WFT-040] under the unchanged rubric.
What failed or remained weak
- WFT-040 still lacked saved-text evidence for Payroll Audit exception handling [WFT-040]; that failure remains published.
09 · Inspectable record
Evidence notes
A sampled transaction ledger and an independent rule replay will verify inclusion and omission decisions.
- WFT-040 preserves the exact synthetic prompt, frozen first-response account, failure-only correction, corrected-response account, and five boolean decisions together.
- WFT-040 scores are arithmetic: 3 first-pass checks × 2 = 6/10; 4 final-pass checks × 2 = 8/10.
- WFT-040 evaluated only the text/static portion of the declared evidence plan—A sampled transaction ledger and an independent rule replay will verify inclusion and omission decisions.—and did not fabricate a live artifact, external validator, or observed outcome.
10 · Boundary of the claim
Limitations
- WFT-040 is a synthetic benchmark, so its worked verdict measures fit to the disclosed fictional Payroll Audit fixtures rather than effectiveness in a real workplace or learning setting.
- WFT-040 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.