Completed field testSynthetic benchmark

Work · Evidence record

AI Contract Redaction Under a Defined Confidentiality Policy — Three of Five Checks Passed

The completed WFT-014 synthetic field test stopped at 6/10: three of five Document Redaction checks passed after one correction, but Document Redaction task fidelity [WFT-014] and Document Redaction handoff usability [WFT-014] remained unsupported.

  • 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
6/10
Verdict
mixed

01 · The assignment

The task

redact confidential data from a contract bundle

02 · Scope before score

Test disclosures

Input disclosure

Synthetic blind-test input pack WFT-014: A legal operations team will provide synthetic contracts containing personal, commercial, and account information under a defined redaction policy. Source facts: controlled excerpts WFT-014-D01 through WFT-014-D05; clauses 2.1, 3.4, 6.2, and 8.7; effective dates 2026-09-01 and 2026-10-15; one defined-term conflict in WFT-014-D04; and a mandatory exception in clause 6.2. Governing rule card: the effective dates and the distinction between mandatory and optional language. 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-014 for “redact confidential data from a contract bundle” in a Codex multi-agent session. We froze the first response, returned only its 3 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-014. Task: redact confidential data from a contract bundle. Context: A legal operations team will provide synthetic contracts containing personal, commercial, and account information under a defined redaction policy. Fictional source facts: controlled excerpts WFT-014-D01 through WFT-014-D05; clauses 2.1, 3.4, 6.2, and 8.7; effective dates 2026-09-01 and 2026-10-15; one defined-term conflict in WFT-014-D04; and a mandatory exception in clause 6.2. Governing policy, formula, or rubric: the effective dates and the distinction between mandatory and optional language. 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 clause matrix, proposed output, and unresolved-source register. 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: Redacted copies and a hidden-answer comparison will verify removals, leaks, and unnecessary redactions.

04 · Baseline preserved

First result

The first response is retained before scoring or correction.

Frozen first response WFT-014 produced a clause matrix, proposed output, and unresolved-source register for the task “redact confidential data from a contract bundle.” It treated the supplied pack as fictional and proposed this central handling: trace the proposed treatment to clauses 2.1 and 6.2, preserve the mandatory exception, and mark the definition conflict in WFT-014-D04 for review. Concrete saved artifact row WFT-014-ROW1 reads: “WFT-014-D01 | trace the proposed treatment to clauses 2.1 and 6.2, preserve the mandatory exception, and mark the definition conflict in WFT-014-D04 for review | source: supplied fictional pack | review state: first-draft.” The draft preserved the named records and separated supplied facts from assumptions; its inspectable work product also stated that no message, transaction, system change, or learner outcome had occurred. On the five predeclared static checks, it passed Document Redaction rule accuracy [WFT-014] and Document Redaction exception handling [WFT-014]. The audit found concrete failures: for Document Redaction task fidelity [WFT-014], the saved draft did not connect WFT-014-D04 to the full boundary of “redact confidential data from a contract bundle”; for Document Redaction source traceability [WFT-014], the saved draft gave the central WFT-014-D04 decision no source-to-output locator; for Document Redaction handoff usability [WFT-014], the saved draft left the clause matrix, proposed output, and unresolved-source register without a complete reviewer handoff and acceptance marker. Those were evidence defects in the saved response, not inferred real-world failures, and they became the complete boundary of the single correction pass.

Initial score: 4/10

05 · One pass only

Exact correction prompt

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

Correct only these detected WFT-014 first-draft failures, using no new input or goal: 1) Document Redaction task fidelity [WFT-014] — the draft did not connect WFT-014-D04 to the full boundary of “redact confidential data from a contract bundle”; 2) Document Redaction source traceability [WFT-014] — the draft gave the central WFT-014-D04 decision no source-to-output locator; 3) Document Redaction handoff usability [WFT-014] — the draft left the clause matrix, proposed output, and unresolved-source register without a complete reviewer handoff and acceptance marker.

06 · Corrected output

Corrected final result

Corrected response WFT-014 retained the original fictional inputs, task boundary, and central decision: trace the proposed treatment to clauses 2.1 and 6.2, preserve the mandatory exception, and mark the definition conflict in WFT-014-D04 for review. Concrete corrected artifact row WFT-014-ROW1 reads: “WFT-014-D01 | trace the proposed treatment to clauses 2.1 and 6.2, preserve the mandatory exception, and mark the definition conflict in WFT-014-D04 for review | evidence locator: WFT-014-D01 | static status: 6/10 after one correction.” It changed only the detected failure areas, adding inspectable rows for Document Redaction source traceability [WFT-014]. The frozen final text passed Document Redaction rule accuracy [WFT-014], Document Redaction exception handling [WFT-014], and Document Redaction source traceability [WFT-014] and still failed Document Redaction task fidelity [WFT-014] and Document Redaction handoff usability [WFT-014]. The final clause matrix, proposed output, and unresolved-source register therefore earned 6/10 from 3 passing checks; no second repair was attempted. This result reports only a bounded transcript-and-fixture evaluation. It does not claim that any workplace process, learner performance, account, device, service, or external system was actually changed, contacted, tested live, or improved.

Final score: 6/10

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
Document Redaction task fidelity [WFT-014] Fail FailWFT-014 static check 1 inspected the saved wording for “Document Redaction task fidelity [WFT-014].” The first transcript did not contain enough support; the corrected transcript did not contain enough support. The comparison used only fictional fixture WFT-014-D04, the declared Document Redaction rules, and the record’s frozen text; no live outcome or external action counted as evidence.
Document Redaction rule accuracy [WFT-014] Pass PassWFT-014 static check 2 inspected the saved wording for “Document Redaction rule accuracy [WFT-014].” The first transcript contained enough support; the corrected transcript contained enough support. The comparison used only fictional fixture WFT-014-D04, the declared Document Redaction rules, and the record’s frozen text; no live outcome or external action counted as evidence.
Document Redaction exception handling [WFT-014] Pass PassWFT-014 static check 3 inspected the saved wording for “Document Redaction exception handling [WFT-014].” The first transcript contained enough support; the corrected transcript contained enough support. The comparison used only fictional fixture WFT-014-D04, the declared Document Redaction rules, and the record’s frozen text; no live outcome or external action counted as evidence.
Document Redaction source traceability [WFT-014] Fail PassWFT-014 static check 4 inspected the saved wording for “Document Redaction source traceability [WFT-014].” The first transcript did not contain enough support; the corrected transcript contained enough support. The comparison used only fictional fixture WFT-014-D04, the declared Document Redaction rules, and the record’s frozen text; no live outcome or external action counted as evidence.
Document Redaction handoff usability [WFT-014] Fail FailWFT-014 static check 5 inspected the saved wording for “Document Redaction handoff usability [WFT-014].” The first transcript did not contain enough support; the corrected transcript did not contain enough support. The comparison used only fictional fixture WFT-014-D04, the declared Document Redaction rules, and the record’s frozen text; no live outcome or external action counted as evidence.
Initial4/10
Final6/10
Verdictmixed
RecommendedNo

08 · No cleanup by omission

What worked—and what failed

What worked

  • WFT-014 kept “redact confidential data from a contract bundle” bounded to disclosed fictional inputs and preserved an auditable first-response snapshot.
  • WFT-014 made the central handling—trace the proposed treatment to clauses 2.1 and 6.2, preserve the mandatory exception, and mark the definition conflict in WFT-014-D04 for review—inspectable rather than implying unseen work.
  • WFT-014 earned final passes for Document Redaction rule accuracy [WFT-014] and Document Redaction exception handling [WFT-014] under the same frozen scoring rules.

What failed or remained weak

  • WFT-014 still lacked enough saved-text evidence for Document Redaction task fidelity [WFT-014]; the record leaves that final failure visible.
  • WFT-014 still lacked enough saved-text evidence for Document Redaction handoff usability [WFT-014]; the record leaves that final failure visible.

09 · Inspectable record

Evidence notes

Redacted copies and a hidden-answer comparison will verify removals, leaks, and unnecessary redactions.

  • WFT-014 preserves the exact synthetic prompt, frozen first-response account, failure-only correction, corrected-response account, and five boolean decisions together.
  • WFT-014 scores are arithmetic: 2 first-pass checks × 2 = 4/10; 3 final-pass checks × 2 = 6/10.
  • WFT-014 evaluated only the text/static portion of the declared evidence plan—Redacted copies and a hidden-answer comparison will verify removals, leaks, and unnecessary redactions.—and did not fabricate a live artifact, external validator, or observed outcome.
Download this case record

10 · Boundary of the claim

Limitations

  • WFT-014 is a synthetic benchmark, so its mixed verdict measures fit to the disclosed fictional Document Redaction fixtures rather than effectiveness in a real workplace or learning setting.
  • WFT-014 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