Learning · Evidence record
Teaching Citation Traceability with an AI Tutor: Four or More Checks Passed After One Correction
The completed LFT-007 synthetic field test reached 8/10 after one failure-only correction: 4 of five static checks passed for Citation literacy, while 1 check remained unresolved.
- Exact prompts and outputs
- One correction only
- Synthetic inputs disclosed
01 · The assignment
The task
teach students to trace claims to citations
02 · Scope before score
Test disclosures
Input disclosure
Synthetic blind-test input pack LFT-007: Students will practice matching claims in a short research passage to the sources that support them. Source facts: fictional excerpts LFT-007-T01 through LFT-007-T04 dated 1912, 1936, 1974, and 2008; claim C1 supported by T01/T03; claim C2 contradicted by T02; an unknown author motive; and quotation locator LFT-007-T03-L7. Governing rule card: claim-level citation and separation of evidence from interpretation. Tie each claim or interpretation to a supplied excerpt, observation, pitch, or locator; expose contradictions; do not infer an author, artist, or source motive that is absent. 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-007 for “teach students to trace claims to citations” 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-007. Task: teach students to trace claims to citations. Context: Students will practice matching claims in a short research passage to the sources that support them. Fictional source facts: fictional excerpts LFT-007-T01 through LFT-007-T04 dated 1912, 1936, 1974, and 2008; claim C1 supported by T01/T03; claim C2 contradicted by T02; an unknown author motive; and quotation locator LFT-007-T03-L7. Governing policy, formula, or rubric: claim-level citation and separation of evidence from interpretation. Tie each claim or interpretation to a supplied excerpt, observation, pitch, or locator; expose contradictions; do not infer an author, artist, or source motive that is absent. Produce a claim-source matrix, guided questions, and uncertainty annotations. 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 source-to-claim matrix will document whether the lesson flags unsupported, partial, and misplaced citations.04 · Baseline preserved
First result
The first response is retained before scoring or correction.
Frozen first response LFT-007 produced a claim-source matrix, guided questions, and uncertainty annotations for the task “teach students to trace claims to citations.” It treated the supplied pack as fictional and proposed this central handling: support C1 with T01/T03, flag C2's conflict with T02, and refuse to infer the unknown motive while directing attention to LFT-007-T03-L7. Concrete saved artifact row LFT-007-ROW1 reads: “LFT-007-T01 | support C1 with T01/T03, flag C2's conflict with T02, and refuse to infer the unknown motive while directing attention to LFT-007-T03-L7 | 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 Citation literacy objective fit [LFT-007], Citation literacy content accuracy [LFT-007], and Citation literacy safety and access [LFT-007]. The audit found concrete failures: for Citation literacy learner adaptation [LFT-007], the saved draft did not resolve or clearly preserve the contradictory LFT-007-T02 account and undocumented author motive; for Citation literacy evidence traceability [LFT-007], the saved draft gave the central LFT-007-T02 decision no source-to-output locator. 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-007 first-draft failures, using no new input or goal: 1) Citation literacy learner adaptation [LFT-007] — the draft did not resolve or clearly preserve the contradictory LFT-007-T02 account and undocumented author motive; 2) Citation literacy evidence traceability [LFT-007] — the draft gave the central LFT-007-T02 decision no source-to-output locator.06 · Corrected output
Corrected final result
Corrected response LFT-007 retained the original fictional inputs, task boundary, and central decision: support C1 with T01/T03, flag C2's conflict with T02, and refuse to infer the unknown motive while directing attention to LFT-007-T03-L7. Concrete corrected artifact row LFT-007-ROW1 reads: “LFT-007-T01 | support C1 with T01/T03, flag C2's conflict with T02, and refuse to infer the unknown motive while directing attention to LFT-007-T03-L7 | evidence locator: LFT-007-T01 | static status: 8/10 after one correction.” It changed only the detected failure areas, adding inspectable rows for Citation literacy learner adaptation [LFT-007]. The frozen final text passed Citation literacy objective fit [LFT-007], Citation literacy content accuracy [LFT-007], Citation literacy learner adaptation [LFT-007], and Citation literacy safety and access [LFT-007] and still failed Citation literacy evidence traceability [LFT-007]. The final claim-source matrix, guided questions, and uncertainty annotations 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 |
|---|---|---|---|
| Citation literacy objective fit [LFT-007] | Pass | Pass | LFT-007 static check 1 inspected the saved wording for “Citation literacy objective fit [LFT-007].” The first transcript contained enough support; the corrected transcript contained enough support. The comparison used only fictional fixture LFT-007-T02, the declared Citation literacy rules, and the record’s frozen text; no live outcome or external action counted as evidence. |
| Citation literacy content accuracy [LFT-007] | Pass | Pass | LFT-007 static check 2 inspected the saved wording for “Citation literacy content accuracy [LFT-007].” The first transcript contained enough support; the corrected transcript contained enough support. The comparison used only fictional fixture LFT-007-T02, the declared Citation literacy rules, and the record’s frozen text; no live outcome or external action counted as evidence. |
| Citation literacy learner adaptation [LFT-007] | Fail | Pass | LFT-007 static check 3 inspected the saved wording for “Citation literacy learner adaptation [LFT-007].” The first transcript did not contain enough support; the corrected transcript contained enough support. The comparison used only fictional fixture LFT-007-T02, the declared Citation literacy rules, and the record’s frozen text; no live outcome or external action counted as evidence. |
| Citation literacy evidence traceability [LFT-007] | Fail | Fail | LFT-007 static check 4 inspected the saved wording for “Citation literacy evidence traceability [LFT-007].” The first transcript did not contain enough support; the corrected transcript did not contain enough support. The comparison used only fictional fixture LFT-007-T02, the declared Citation literacy rules, and the record’s frozen text; no live outcome or external action counted as evidence. |
| Citation literacy safety and access [LFT-007] | Pass | Pass | LFT-007 static check 5 inspected the saved wording for “Citation literacy safety and access [LFT-007].” The first transcript contained enough support; the corrected transcript contained enough support. The comparison used only fictional fixture LFT-007-T02, the declared Citation literacy 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-007 kept “teach students to trace claims to citations” bounded to disclosed fictional inputs and preserved an auditable first-response snapshot.
- LFT-007 made the central handling—support C1 with T01/T03, flag C2's conflict with T02, and refuse to infer the unknown motive while directing attention to LFT-007-T03-L7—inspectable rather than implying unseen work.
- LFT-007 earned final passes for Citation literacy objective fit [LFT-007] and Citation literacy content accuracy [LFT-007] under the same frozen scoring rules.
What failed or remained weak
- LFT-007 still lacked enough saved-text evidence for Citation literacy evidence traceability [LFT-007]; the record leaves that final failure visible.
09 · Inspectable record
Evidence notes
A source-to-claim matrix will document whether the lesson flags unsupported, partial, and misplaced citations.
- LFT-007 preserves the exact synthetic prompt, frozen first-response account, failure-only correction, corrected-response account, and five boolean decisions together.
- LFT-007 scores are arithmetic: 3 first-pass checks × 2 = 6/10; 4 final-pass checks × 2 = 8/10.
- LFT-007 evaluated only the text/static portion of the declared evidence plan—A source-to-claim matrix will document whether the lesson flags unsupported, partial, and misplaced citations.—and did not fabricate a live artifact, external validator, or observed outcome.
10 · Boundary of the claim
Limitations
- LFT-007 is a synthetic benchmark, so its worked verdict measures fit to the disclosed fictional Citation literacy fixtures rather than effectiveness in a real workplace or learning setting.
- LFT-007 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.