Learning · Evidence record
Does AI Apply an Essay Rubric Consistently Across Writing Styles — Completed Benchmark Result: 8/10
The completed LFT-014 synthetic field test reached 8/10 after one failure-only correction: 4 of five static checks passed for Rubric calibration, while 1 check remained unresolved.
- Exact prompts and outputs
- One correction only
- Synthetic inputs disclosed
01 · The assignment
The task
apply an essay rubric consistently across writing styles
02 · Scope before score
Test disclosures
Input disclosure
Synthetic blind-test input pack LFT-014: The AI will assess anonymized essays that vary in voice and structure using one fixed analytic rubric. Source facts: fictional learner artifacts LFT-014-W01 through LFT-014-W04; rubric criteria R1–R5; passages P2 and P7 as admissible evidence; an unsupported conclusion in W03; a stylistic variation in W04; and a no-rewrite boundary. Governing rule card: consistent rubric application without replacing learner work. Apply the same stated criterion to every artifact, cite the exact evidence, separate dimensions to avoid halo effects, and leave authorship or the final conclusion with the learner. 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 LFT-014 for “apply an essay rubric consistently across writing styles” 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 LFT-014. Task: apply an essay rubric consistently across writing styles. Context: The AI will assess anonymized essays that vary in voice and structure using one fixed analytic rubric. Fictional source facts: fictional learner artifacts LFT-014-W01 through LFT-014-W04; rubric criteria R1–R5; passages P2 and P7 as admissible evidence; an unsupported conclusion in W03; a stylistic variation in W04; and a no-rewrite boundary. Governing policy, formula, or rubric: consistent rubric application without replacing learner work. Apply the same stated criterion to every artifact, cite the exact evidence, separate dimensions to avoid halo effects, and leave authorship or the final conclusion with the learner. Produce a criterion-level feedback table, evidence citations, and next-step prompt. 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: Criterion-level annotations will be compared with independent educator judgments and cited passage evidence.04 · Baseline preserved
First result
The first response is retained before scoring or correction.
Frozen first response LFT-014 produced a criterion-level feedback table, evidence citations, and next-step prompt for the task “apply an essay rubric consistently across writing styles.” It treated the supplied pack as fictional and proposed this central handling: cite P2/P7 for LFT-014-W01, diagnose the unsupported W03 conclusion, score W04 by the same R1–R5 criteria, and preserve the no-rewrite boundary. Concrete saved artifact row LFT-014-ROW1 reads: “LFT-014-W01 | cite P2/P7 for LFT-014-W01, diagnose the unsupported W03 conclusion, score W04 by the same R1–R5 criteria, and preserve the no-rewrite boundary | source: supplied fictional pack | review state: first-draft.” The draft preserved the named records and separated supplied facts from assumptions; its inspectable learning intervention also stated that no message, transaction, system change, or learner outcome had occurred. On the five predeclared static checks, it passed Rubric calibration objective fit [LFT-014], Rubric calibration evidence traceability [LFT-014], and Rubric calibration safety and access [LFT-014]. The audit found concrete failures: for Rubric calibration content accuracy [LFT-014], the saved draft left consistent rubric application without replacing learner work without an explicit verification row; for Rubric calibration learner adaptation [LFT-014], the saved draft did not resolve or clearly preserve the unsupported LFT-014-W03 conclusion and stylistic variation in W04. 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: 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 LFT-014 first-draft failures, using no new input or goal: 1) Rubric calibration content accuracy [LFT-014] — the draft left consistent rubric application without replacing learner work without an explicit verification row; 2) Rubric calibration learner adaptation [LFT-014] — the draft did not resolve or clearly preserve the unsupported LFT-014-W03 conclusion and stylistic variation in W04.06 · Corrected output
Corrected final result
Corrected response LFT-014 retained the original fictional inputs, task boundary, and central decision: cite P2/P7 for LFT-014-W01, diagnose the unsupported W03 conclusion, score W04 by the same R1–R5 criteria, and preserve the no-rewrite boundary. Concrete corrected artifact row LFT-014-ROW1 reads: “LFT-014-W01 | cite P2/P7 for LFT-014-W01, diagnose the unsupported W03 conclusion, score W04 by the same R1–R5 criteria, and preserve the no-rewrite boundary | evidence locator: LFT-014-W01 | static status: 8/10 after one correction.” It changed only the detected failure areas, adding inspectable rows for Rubric calibration content accuracy [LFT-014]. The frozen final text passed Rubric calibration objective fit [LFT-014], Rubric calibration content accuracy [LFT-014], Rubric calibration evidence traceability [LFT-014], and Rubric calibration safety and access [LFT-014] and still failed Rubric calibration learner adaptation [LFT-014]. The final criterion-level feedback table, evidence citations, and next-step prompt therefore earned 8/10 from 4 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: 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 |
|---|---|---|---|
| Rubric calibration objective fit [LFT-014] | Pass | Pass | LFT-014 static check 1 inspected the saved wording for “Rubric calibration objective fit [LFT-014].” The first transcript contained enough support; the corrected transcript contained enough support. The comparison used only fictional fixture LFT-014-W03, the declared Rubric calibration rules, and the record’s frozen text; no live outcome or external action counted as evidence. |
| Rubric calibration content accuracy [LFT-014] | Fail | Pass | LFT-014 static check 2 inspected the saved wording for “Rubric calibration content accuracy [LFT-014].” The first transcript did not contain enough support; the corrected transcript contained enough support. The comparison used only fictional fixture LFT-014-W03, the declared Rubric calibration rules, and the record’s frozen text; no live outcome or external action counted as evidence. |
| Rubric calibration learner adaptation [LFT-014] | Fail | Fail | LFT-014 static check 3 inspected the saved wording for “Rubric calibration learner adaptation [LFT-014].” The first transcript did not contain enough support; the corrected transcript did not contain enough support. The comparison used only fictional fixture LFT-014-W03, the declared Rubric calibration rules, and the record’s frozen text; no live outcome or external action counted as evidence. |
| Rubric calibration evidence traceability [LFT-014] | Pass | Pass | LFT-014 static check 4 inspected the saved wording for “Rubric calibration evidence traceability [LFT-014].” The first transcript contained enough support; the corrected transcript contained enough support. The comparison used only fictional fixture LFT-014-W03, the declared Rubric calibration rules, and the record’s frozen text; no live outcome or external action counted as evidence. |
| Rubric calibration safety and access [LFT-014] | Pass | Pass | LFT-014 static check 5 inspected the saved wording for “Rubric calibration safety and access [LFT-014].” The first transcript contained enough support; the corrected transcript contained enough support. The comparison used only fictional fixture LFT-014-W03, the declared Rubric calibration rules, and the record’s frozen text; no live outcome or external action counted as evidence. |
08 · No cleanup by omission
What worked—and what failed
What worked
- LFT-014 kept “apply an essay rubric consistently across writing styles” bounded to disclosed fictional inputs and preserved an auditable first-response snapshot.
- LFT-014 made the central handling—cite P2/P7 for LFT-014-W01, diagnose the unsupported W03 conclusion, score W04 by the same R1–R5 criteria, and preserve the no-rewrite boundary—inspectable rather than implying unseen work.
- LFT-014 earned final passes for Rubric calibration objective fit [LFT-014] and Rubric calibration content accuracy [LFT-014] under the same frozen scoring rules.
What failed or remained weak
- LFT-014 still lacked enough saved-text evidence for Rubric calibration learner adaptation [LFT-014]; the record leaves that final failure visible.
09 · Inspectable record
Evidence notes
Criterion-level annotations will be compared with independent educator judgments and cited passage evidence.
- LFT-014 preserves the exact synthetic prompt, frozen first-response account, failure-only correction, corrected-response account, and five boolean decisions together.
- LFT-014 scores are arithmetic: 3 first-pass checks × 2 = 6/10; 4 final-pass checks × 2 = 8/10.
- LFT-014 evaluated only the text/static portion of the declared evidence plan—Criterion-level annotations will be compared with independent educator judgments and cited passage evidence.—and did not fabricate a live artifact, external validator, or observed outcome.
10 · Boundary of the claim
Limitations
- LFT-014 is a synthetic benchmark, so its worked verdict measures fit to the disclosed fictional Rubric calibration fixtures rather than effectiveness in a real workplace or learning setting.
- LFT-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.