Work · Evidence record
How Should AI Localize Safety Instructions Without Diluting Warnings — Three of Five Checks Passed
The completed WFT-062 synthetic field test stopped at 6/10: three of five Safety Localization checks passed after one correction, but Safety Localization task fidelity [WFT-062] and Safety Localization rule accuracy [WFT-062] remained unsupported.
- Exact prompts and outputs
- One correction only
- Synthetic inputs disclosed
01 · The assignment
The task
localize workplace safety instructions while preserving mandatory warnings
02 · Scope before score
Test disclosures
Input disclosure
Synthetic blind-test input pack WFT-062: A safety team will provide approved source instructions, a terminology list, reading-level target, and legally fixed warning language. Source facts: English WFT-062-S01–S08; mandatory 'LOCK OUT BEFORE SERVICE'; glossary isolator/guard/permit; Spanish (Mexico); diagram labels A–F. Governing rule card: no warning omission, softened modality, changed measurement, or label drift. 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-062 for “localize workplace safety instructions while preserving mandatory warnings” 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-062. Task: localize workplace safety instructions while preserving mandatory warnings. Context: A safety team will provide approved source instructions, a terminology list, reading-level target, and legally fixed warning language. Fictional source facts: English WFT-062-S01–S08; mandatory 'LOCK OUT BEFORE SERVICE'; glossary isolator/guard/permit; Spanish (Mexico); diagram labels A–F. Governing policy, formula, or rubric: no warning omission, softened modality, changed measurement, or label drift. 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 localized instruction sheet, protected-warning table, and terminology review. 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 bilingual clause alignment and terminology audit will verify warning preservation, omissions, substitutions, and readability.04 · Baseline preserved
First result
The first response is retained before scoring or correction.
Frozen first response WFT-062 produced a localized instruction sheet, protected-warning table, and terminology review for “localize workplace safety instructions while preserving mandatory warnings.” Its first artifact row read “WFT-062-S05 | translate procedural prose, preserve lockout warning force and capitalization, use glossary terms, and retain A–F labels | status: proposed | source: fictional fixture.” A second row named the mandatory lockout warning and locale-specific permit term and recorded a disposition. The rule cell mentioned without verifying no warning omission, softened modality, changed measurement, or label drift. No message, transaction, system change, or learner outcome occurred. The audit passed Safety Localization exception handling [WFT-062] and Safety Localization source traceability [WFT-062]. It found for Safety Localization task fidelity [WFT-062], the draft did not link WFT-062-S05 to the full task boundary; for Safety Localization rule accuracy [WFT-062], the draft mentioned but did not verify no warning omission, softened modality, changed measurement, or label drift; for Safety Localization handoff usability [WFT-062], the draft left the localized instruction sheet, protected-warning table, and terminology review without a reviewer-ready acceptance marker. Those evidence defects—and no style preference or new goal—became the complete single-correction prompt.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-062 first-draft failures, using no new input or goal: 1) Safety Localization task fidelity [WFT-062] — the draft did not link WFT-062-S05 to the full task boundary; 2) Safety Localization rule accuracy [WFT-062] — the draft mentioned but did not verify no warning omission, softened modality, changed measurement, or label drift; 3) Safety Localization handoff usability [WFT-062] — the draft left the localized instruction sheet, protected-warning table, and terminology review without a reviewer-ready acceptance marker.06 · Corrected output
Corrected final result
Corrected response WFT-062 preserved all supplied identifiers and the central decision: translate procedural prose, preserve lockout warning force and capitalization, use glossary terms, and retain A–F labels. Its corrected row read “WFT-062-S05 | rule: no warning omission, softened modality, changed measurement, or label drift | decision: translate procedural prose, preserve lockout warning force and capitalization, use glossary terms, and retain A–F labels | static status: 6/10.” It changed only failed dimensions, adding support for Safety Localization handoff usability [WFT-062]. The final audit passed Safety Localization exception handling [WFT-062], Safety Localization source traceability [WFT-062], and Safety Localization handoff usability [WFT-062]. It still lacked Safety Localization task fidelity [WFT-062] and Safety Localization rule accuracy [WFT-062]; those failures remain visible. The localized instruction sheet, protected-warning table, and terminology review earned 6/10 from 3 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: 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.
| Check | First | Final | Evidence |
|---|---|---|---|
| Safety Localization task fidelity [WFT-062] | Fail | Fail | WFT-062 static check 1 inspected “Safety Localization task fidelity [WFT-062]” against WFT-062-S05, the rule “no warning omission, softened modality, changed measurement, or label drift,” and the saved localized instruction sheet, protected-warning table, and terminology review. The first transcript failed; the corrected transcript failed. Only fictional text counted, never a live outcome. |
| Safety Localization rule accuracy [WFT-062] | Fail | Fail | WFT-062 static check 2 inspected “Safety Localization rule accuracy [WFT-062]” against WFT-062-S05, the rule “no warning omission, softened modality, changed measurement, or label drift,” and the saved localized instruction sheet, protected-warning table, and terminology review. The first transcript failed; the corrected transcript failed. Only fictional text counted, never a live outcome. |
| Safety Localization exception handling [WFT-062] | Pass | Pass | WFT-062 static check 3 inspected “Safety Localization exception handling [WFT-062]” against WFT-062-S05, the rule “no warning omission, softened modality, changed measurement, or label drift,” and the saved localized instruction sheet, protected-warning table, and terminology review. The first transcript passed; the corrected transcript passed. Only fictional text counted, never a live outcome. |
| Safety Localization source traceability [WFT-062] | Pass | Pass | WFT-062 static check 4 inspected “Safety Localization source traceability [WFT-062]” against WFT-062-S05, the rule “no warning omission, softened modality, changed measurement, or label drift,” and the saved localized instruction sheet, protected-warning table, and terminology review. The first transcript passed; the corrected transcript passed. Only fictional text counted, never a live outcome. |
| Safety Localization handoff usability [WFT-062] | Fail | Pass | WFT-062 static check 5 inspected “Safety Localization handoff usability [WFT-062]” against WFT-062-S05, the rule “no warning omission, softened modality, changed measurement, or label drift,” and the saved localized instruction sheet, protected-warning table, and terminology review. 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-062 bounded “localize workplace safety instructions while preserving mandatory warnings” to disclosed fictional inputs and froze the first response.
- WFT-062 exposed WFT-062-S05—translate procedural prose, preserve lockout warning force and capitalization, use glossary terms, and retain A–F labels—inside the saved localized instruction sheet, protected-warning table, and terminology review.
- WFT-062 earned inspectable passes for Safety Localization exception handling [WFT-062] and Safety Localization source traceability [WFT-062] under the unchanged rubric.
What failed or remained weak
- WFT-062 still lacked saved-text evidence for Safety Localization task fidelity [WFT-062]; that failure remains published.
- WFT-062 still lacked saved-text evidence for Safety Localization rule accuracy [WFT-062]; that failure remains published.
09 · Inspectable record
Evidence notes
A bilingual clause alignment and terminology audit will verify warning preservation, omissions, substitutions, and readability.
- WFT-062 preserves the exact synthetic prompt, frozen first-response account, failure-only correction, corrected-response account, and five boolean decisions together.
- WFT-062 scores are arithmetic: 2 first-pass checks × 2 = 4/10; 3 final-pass checks × 2 = 6/10.
- WFT-062 evaluated only the text/static portion of the declared evidence plan—A bilingual clause alignment and terminology audit will verify warning preservation, omissions, substitutions, and readability.—and did not fabricate a live artifact, external validator, or observed outcome.
10 · Boundary of the claim
Limitations
- WFT-062 is a synthetic benchmark, so its mixed verdict measures fit to the disclosed fictional Safety Localization fixtures rather than effectiveness in a real workplace or learning setting.
- WFT-062 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.