Problem

modes/interview/debrief.md is append-only against the role-specific prep file — its own Inputs section says "append debrief notes." Every concrete instruction in the mode follows that shape: add a "Gaps to Close Before Round N" section, add stories to interview-prep/story-bank.md, log retractions to interview-prep/retracted-claims.md. Nothing handles the case where a live interview reveals that a specific fact already stated in the prep file is simply wrong — not a new gap, a correction to something already asserted (location, comp range, team size, reporting structure, tech/system stack).

Concrete example: a prep file's header said a role was "Metro Hall, on-site" (inferred from an expired posting, tagged as such). The actual recruiter call revealed it's hybrid. The correct fix was editing that header line in place, not appending a new note under an unchanged, now-wrong claim.

Solution

Adds a new Step 1b — Check for Contradicted Facts to modes/interview/debrief.md, running alongside Step 1's capture phase:

• Flags when the interview directly contradicts a specific existing fact in the role-specific prep file (as opposed to surfacing new information, which still appends via the existing Steps 4/5/8, unchanged).
• Instructs correcting the original line in place with a strikethrough-plus-correction format:

  ~~Metro Hall, on-site~~ **Metro Hall — hybrid** (confirmed on the {date} call)
  

• Instructs resolving any inference tag ([inferred from JD], or prose noting an expired/inaccessible source) once the interview confirms or corrects it — replacing the marker with the confirmed fact and its real source, rather than leaving a settled fact permanently marked uncertain.

The Inputs section (item 9) and the Rules section both get a one-line pointer to this new step for reinforcement. No existing step's behavior changes — this is additive only.

Added test-all.mjs section 63 (following the same mode-doc pinning convention as section 62's stated-comp wiring check) asserting: the new step exists, the append-vs-correct distinction is documented, the strikethrough-correction example format is present verbatim, and the inference-tag resolution instruction is present.

node test-all.mjs: 2048 passed, 0 failed, 2 pre-existing warnings (unrelated to this change).

Coordination note

Overlaps file-wise with #2122 (also touches modes/interview/debrief.md, different section — that PR adds a transcript-input path to Step 1/Step 9, this PR adds a fact-correction step elsewhere in the mode). No content conflict expected; a mechanical rebase may be needed on whichever of the two merges second.

Closes #2125

Summary by CodeRabbit

• Documentation

  • Updated the interview debrief workflow with a new Contradicted Facts sub-step to correct existing role prep-file claims when interviews directly contradict them.
  • Added clear guidance to distinguish new information (append) from contradictions (correct in place).
  • Documented the required in-place correction format and added instructions for resolving inference tags after contradiction/confirmation.
  • Clarified boundaries to avoid touching retracted-claims, and not updating the story bank or rewriting gap additions.

• Tests

  • Added an end-of-suite check to enforce the contradicted-facts workflow requirements.