# Prompt — Contradiction Log (multi-source reconciliation) > The third prompt in the Appendix E § Section 2 starter pack. Use this when you are merging multiple transcripts that may disagree — the failure mode where the [template-fill prompt](prompt-template-fill.md) will happily produce a confident-looking draft from inconsistent input. > > **When to use this**: Two participants in one meeting describe different success criteria; a follow-up email contradicts the kickoff transcript; the requestor's manager has a different audience in mind than the requestor. Skip reconciliation and the BRD will read as confirming each party's own view — exactly the failure pattern *The Verification Habit* (Ch 7) describes as *consensus that no one actually agrees to*. --- ## The reconciliation workflow 1. Run the [extraction prompt](prompt-extraction.md) **separately against each source** (transcript, email thread, project charter). Do not concatenate sources before extraction — the per-source extraction is what makes the contradictions visible. 2. Run the contradiction-log prompt below against all per-source extraction outputs. 3. Read the contradiction log. For each contradiction, either resolve it through follow-up (preferred) or document it as an open question in the Constraints & Assumptions section of the eventual BRD (acceptable only when follow-up is genuinely blocked). 4. Once contradictions are resolved (or explicitly logged), run the [template-fill prompt](prompt-template-fill.md) against the reconciled fact list — not against the original transcripts. --- ## The prompt Copy from `--- BEGIN PROMPT ---` to `--- END PROMPT ---` verbatim. Replace the `[paste extraction outputs ...]` block with the per-source extraction outputs from step 1, clearly labeled by source. ```text --- BEGIN PROMPT --- You are an analyst reconciling structured fact lists extracted from multiple meeting transcripts, email threads, and project artifacts. Your job is to produce a contradiction log that names every disagreement across sources, so a human can resolve them before drafting the BRD. Rules: - Compare facts across the labeled sources. Treat facts that agree as confirmed (do not list them in the contradiction log unless the agreement is itself noteworthy, e.g., three independent sources confirming an unusual constraint). - For each disagreement, produce a contradiction entry with the following fields: * BRD section the contradiction touches (Executive Summary, Business Context, Key Business Questions, Success Criteria, Stakeholder Requirements, Data Requirements, Deliverables, Constraints & Assumptions, or "Out of scope" if it does not map to any section) * The contradicting claims, each attributed to its source ("[Source A — kickoff transcript]: ..."; "[Source B — follow-up email]: ...") * The contradiction type, chosen from: factual (only one can be true), scope-disagreement (parties want different things), authority-ambiguity (who decides this is unclear), or definitional (parties are using the same word to mean different things) * A recommended resolution path, chosen from: "ask source A to confirm", "ask source B to confirm", "convene a focused meeting with both", "document in Constraints & Assumptions and proceed", or "escalate to the project sponsor" - Do not attempt to resolve the contradictions in this output. The resolution belongs to the human, after follow-up. Your job is to surface the disagreement, not to silently pick one side. - Where the sources implicitly disagree (the topic comes up in two sources but each treats it as obvious from a different angle), flag it as a contradiction even if no explicit statement conflicts. Implicit disagreement is harder to catch later and this prompt is built around catching it now. Output: the contradiction log only. Do not include the confirmed (agreeing) facts; they will be drafted into the BRD separately. Per-source extraction outputs: [paste extraction outputs from each source here, each clearly labeled with its source identifier] --- END PROMPT --- ``` --- ## What good output looks like A contradiction log organized by BRD section, with each entry showing the contradicting claims attributed to their sources, the contradiction type, and a recommended resolution path. The log is short (often three to seven entries from a kickoff-plus-email reconciliation; longer from a multi-stakeholder workshop). The output is most useful when read alongside the per-source extraction outputs — the contradiction log says *here is what disagrees*, the extraction outputs say *here is what each source said*, and together they let the human reviewer see whether the contradiction is real or an extraction artifact. ## What bad output looks like (the failure mode this prompt prevents) A reconciliation that silently picks one source as "correct" and uses its facts in the eventual BRD without flagging the disagreement. The BRD reads cleanly to whoever's source won, and the losing-source stakeholder pushes back at sign-off with "this doesn't reflect what I said in the meeting." The elicitation loop then gets blamed for producing a BRD that misrepresents the conversation, when the actual cause is the reconciliation step that was skipped. If your output looks like that: split the sources back out, re-run extraction per source, and re-run this contradiction-log prompt. The contradictions are there; the failure was in collapsing them too early. --- **Source**: Operational expansion of Appendix E § Section 2 "When the simple loop does not work" — specifically the subsection on *When stakeholder claims conflict across captures* and the contradiction-log technique referenced there.