Work · Evidence record
Finding Broken Spreadsheet Formulas with AI: The One-Pass Revision Reached 8/10
The completed WFT-009 synthetic field test reached 8/10 after one failure-only correction: 4 of five static checks passed for Formula Auditing, while 1 check remained unresolved.
- Exact prompts and outputs
- One correction only
- Synthetic inputs disclosed
01 · The assignment
The task
find broken formulas in a monthly reporting workbook
02 · Scope before score
Test disclosures
Input disclosure
Synthetic blind-test input pack WFT-009: A finance team will provide a workbook seeded with reference, range, sign, and copied-formula errors. Source facts: workbook WFT-009-B2:H14; #REF! in F8; SUM(B9:F9) copied incorrectly into G10; reversed sign in D12; June omitted from H14; monthly control values $13,200, $13,680, $14,050, $14,210, $14,590, and $14,590. Governing rule card: formula lineage, range boundaries, sign convention, and independent recalculation. Apply every supplied rule in its stated order, abstain on incomplete rows, preserve conflicts as exceptions, and expose the source locator and arithmetic or rationale for every decision. 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-009 for “find broken formulas in a monthly reporting workbook” 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-009. Task: find broken formulas in a monthly reporting workbook. Context: A finance team will provide a workbook seeded with reference, range, sign, and copied-formula errors. Fictional source facts: workbook WFT-009-B2:H14; #REF! in F8; SUM(B9:F9) copied incorrectly into G10; reversed sign in D12; June omitted from H14; monthly control values $13,200, $13,680, $14,050, $14,210, $14,590, and $14,590. Governing policy, formula, or rubric: formula lineage, range boundaries, sign convention, and independent recalculation. Apply every supplied rule in its stated order, abstain on incomplete rows, preserve conflicts as exceptions, and expose the source locator and arithmetic or rationale for every decision. Produce a cell-level formula audit, repair proposal, and recalculation log. 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 marked-up workbook and recalculation tests will verify every flagged formula and missed defect.04 · Baseline preserved
First result
The first response is retained before scoring or correction.
Frozen first response WFT-009 produced a cell-level formula audit, repair proposal, and recalculation log for “find broken formulas in a monthly reporting workbook.” Its first artifact row read “WFT-009-G10 | flag F8, repair G10 to SUM(C10:G10), reverse D12’s expense sign, extend H14 through June, and recompute $84,320 | status: proposed | source: fictional fixture.” A second row named the copied-range fault in G10 and omitted June value in H14 and recorded a disposition. The rule cell verified formula lineage, range boundaries, sign convention, and independent recalculation. No message, transaction, system change, or learner outcome occurred. The audit passed Formula Auditing rule accuracy [WFT-009], Formula Auditing exception handling [WFT-009], and Formula Auditing source traceability [WFT-009]. It found for Formula Auditing task fidelity [WFT-009], the draft did not link WFT-009-G10 to the full task boundary; for Formula Auditing handoff usability [WFT-009], the draft left the cell-level formula audit, repair proposal, and recalculation log without a reviewer-ready acceptance marker. 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-009 first-draft failures, using no new input or goal: 1) Formula Auditing task fidelity [WFT-009] — the draft did not link WFT-009-G10 to the full task boundary; 2) Formula Auditing handoff usability [WFT-009] — the draft left the cell-level formula audit, repair proposal, and recalculation log without a reviewer-ready acceptance marker.06 · Corrected output
Corrected final result
Corrected response WFT-009 preserved all supplied identifiers and the central decision: flag F8, repair G10 to SUM(C10:G10), reverse D12’s expense sign, extend H14 through June, and recompute $84,320. Its corrected row read “WFT-009-G10 | rule: formula lineage, range boundaries, sign convention, and independent recalculation | decision: flag F8, repair G10 to SUM(C10:G10), reverse D12’s expense sign, extend H14 through June, and recompute $84,320 | static status: 8/10.” It changed only failed dimensions, adding support for Formula Auditing handoff usability [WFT-009]. The final audit passed Formula Auditing rule accuracy [WFT-009], Formula Auditing exception handling [WFT-009], Formula Auditing source traceability [WFT-009], and Formula Auditing handoff usability [WFT-009]. It still lacked Formula Auditing task fidelity [WFT-009]; those failures remain visible. The cell-level formula audit, repair proposal, and recalculation log 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 |
|---|---|---|---|
| Formula Auditing task fidelity [WFT-009] | Fail | Fail | WFT-009 static check 1 inspected “Formula Auditing task fidelity [WFT-009]” against WFT-009-G10, the rule “formula lineage, range boundaries, sign convention, and independent recalculation,” and the saved cell-level formula audit, repair proposal, and recalculation log. The first transcript failed; the corrected transcript failed. Only fictional text counted, never a live outcome. |
| Formula Auditing rule accuracy [WFT-009] | Pass | Pass | WFT-009 static check 2 inspected “Formula Auditing rule accuracy [WFT-009]” against WFT-009-G10, the rule “formula lineage, range boundaries, sign convention, and independent recalculation,” and the saved cell-level formula audit, repair proposal, and recalculation log. The first transcript passed; the corrected transcript passed. Only fictional text counted, never a live outcome. |
| Formula Auditing exception handling [WFT-009] | Pass | Pass | WFT-009 static check 3 inspected “Formula Auditing exception handling [WFT-009]” against WFT-009-G10, the rule “formula lineage, range boundaries, sign convention, and independent recalculation,” and the saved cell-level formula audit, repair proposal, and recalculation log. The first transcript passed; the corrected transcript passed. Only fictional text counted, never a live outcome. |
| Formula Auditing source traceability [WFT-009] | Pass | Pass | WFT-009 static check 4 inspected “Formula Auditing source traceability [WFT-009]” against WFT-009-G10, the rule “formula lineage, range boundaries, sign convention, and independent recalculation,” and the saved cell-level formula audit, repair proposal, and recalculation log. The first transcript passed; the corrected transcript passed. Only fictional text counted, never a live outcome. |
| Formula Auditing handoff usability [WFT-009] | Fail | Pass | WFT-009 static check 5 inspected “Formula Auditing handoff usability [WFT-009]” against WFT-009-G10, the rule “formula lineage, range boundaries, sign convention, and independent recalculation,” and the saved cell-level formula audit, repair proposal, and recalculation log. The first transcript failed; 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-009 bounded “find broken formulas in a monthly reporting workbook” to disclosed fictional inputs and froze the first response.
- WFT-009 exposed WFT-009-G10—flag F8, repair G10 to SUM(C10:G10), reverse D12’s expense sign, extend H14 through June, and recompute $84,320—inside the saved cell-level formula audit, repair proposal, and recalculation log.
- WFT-009 earned inspectable passes for Formula Auditing rule accuracy [WFT-009] and Formula Auditing exception handling [WFT-009] under the unchanged rubric.
What failed or remained weak
- WFT-009 still lacked saved-text evidence for Formula Auditing task fidelity [WFT-009]; that failure remains published.
09 · Inspectable record
Evidence notes
A marked-up workbook and recalculation tests will verify every flagged formula and missed defect.
- WFT-009 preserves the exact synthetic prompt, frozen first-response account, failure-only correction, corrected-response account, and five boolean decisions together.
- WFT-009 scores are arithmetic: 3 first-pass checks × 2 = 6/10; 4 final-pass checks × 2 = 8/10.
- WFT-009 evaluated only the text/static portion of the declared evidence plan—A marked-up workbook and recalculation tests will verify every flagged formula and missed defect.—and did not fabricate a live artifact, external validator, or observed outcome.
10 · Boundary of the claim
Limitations
- WFT-009 is a synthetic benchmark, so its worked verdict measures fit to the disclosed fictional Formula Auditing fixtures rather than effectiveness in a real workplace or learning setting.
- WFT-009 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.