# Prompt — Template-Fill (one-pass)

> The single prompt from Appendix E § Section 2 that does both jobs: extract the facts the transcript supports, then fill the eight-section template using only those facts, marking every gap explicitly. Engineered to keep the LLM from smoothing over gaps with plausible-sounding inference.
>
> **When to use this**: Common case. One stakeholder transcript that fits in the LLM's working context, no cross-source contradictions to track.
>
> **When NOT to use this**: Long transcripts that exceed context (use the [extraction prompt](prompt-extraction.md) as a first pass instead), high-stakes BRDs where you want to audit the extraction before it shapes the draft, or conflicting stakeholder claims (use the [contradiction-log prompt](prompt-contradiction-log.md)).

---

## The prompt

Copy from `--- BEGIN PROMPT ---` to `--- END PROMPT ---` verbatim. Replace the two `[paste ...]` blocks with the [eight-section template](brd-template.md) and your transcript.

```text
--- BEGIN PROMPT ---
You are an analyst authoring a Business Requirements Document (BRD)
from a meeting transcript, using the template below.

Rules:
- Every claim you write must be supported by something in the
  transcript. Do not infer from general knowledge of the domain.
- Where the transcript does not contain the answer for a section,
  sub-section, or table cell, insert "[PLACEHOLDER: <brief description
  of what is missing>]" exactly as shown. Do not soften placeholders
  with surrounding narrative; they are the most useful output.
- Preserve direct quotes that capture the stakeholder's phrasing of
  audience, decision, or success criterion.
- Keep each filled section concise but complete enough to support
  stakeholder sign-off review; cut content that is not load-bearing
  for the decision, but do not compress past the point of usefulness.

Output only the filled template.

Template:
[paste the eight-section template from Section 1 here]

Transcript:
[paste transcript here]
--- END PROMPT ---
```

---

## What good output looks like

The draft BRD with `[PLACEHOLDER: <description>]` markers in every gap. The markers are the gap inventory — each one names a specific question you now have to answer through follow-up (email an obvious participant, schedule a focused session, defer with a flag in Constraints, or pause and escalate scope risk).

A useful diagnostic on its own: a transcript that produces placeholders in *most* of the eight sections is a meeting that did not, in fact, scope the work. The next move is to ask for a follow-up meeting rather than to draft a BRD against thin material.

## What bad output looks like (the failure mode this prompt prevents)

A fluent-looking BRD with every section filled in confidently — but the content of half the sections is plausible-sounding inference from domain knowledge rather than from the transcript. The first stakeholder review surfaces the inference, the stakeholder loses trust in the BRD, and the elicitation discipline gets blamed for the LLM's hallucination.

If your output looks like that: switch to a model that respects the refusal instruction, or run the [extraction prompt](prompt-extraction.md) as a separate first pass so you can see what the transcript actually supports before the template-fill obscures it.

---

**Source**: Appendix E § Section 2 "The prompt" of *The Defensible Decision*. Reproduced verbatim.