---
name: pacing-inspector
description: In the audit phase, compare actual scene-by-scene tension and word counts against the target tension-curve.md and telemetry.md baselines. Flag drag zones, energy dips, and pacing inconsistencies. ALWAYS use this skill when orchestrator invokes audit-phase inspection, or when the user asks for "pacing audit," "drag zones," "energy curve."
---

# Pacing Inspector

> **Theme runtime.** Read `theme/active.yaml` before producing user-visible output. Substitute every `[token]` in this skill's prose with the value from that file. If a token is missing from `active.yaml`, fall back to the value in `themes/f1/theme.yaml`. If `theme/active.yaml` itself is absent, defer to the orchestrator (which dispatches `theme-interview`).

An [audit-phase] pass that compares the actual scene-by-scene tension trajectory and word counts of the finished manuscript against the targets set in `tension-curve.md` and `telemetry.md`. Flags drag zones, energy dips, and pacing inconsistencies at manuscript scale — the things that only show up when you look at the whole book at once. Flag-only; never rewrites.

This is the manuscript-wide pacing audit. The per-[unit-of-work] pacing check is the `panel-pacing` skill's job; here you look across the whole book.

## Inputs

- `manuscript/chapter-NN-scene-NN.md` for every scene in the book.
- `tension-curve.md` — target tension per scene (start tension, peak tension, end tension, direction of movement).
- `telemetry.md` — target word count per scene and per chapter.
- `payoffs.md` — to identify the structural high-tension beats (midpoint, climax) so a drag-zone flag near those beats is treated as more severe.

## Output

A section in `final-pass-flags.md` under `## Pacing` with:
- A summary of pacing health: total drag zones flagged, total trajectory misses, total word-count misses.
- Per-flag detail: scene location, target vs. actual, severity, 1–3 reframe candidates (cut, compress, add tension beat, swap with an adjacent scene).
- A simple ASCII or markdown-table chart showing the manuscript-wide actual-vs-target curve for visual scan.

## Procedure

1. For each scene in `manuscript/`, read the prose and compute (or estimate from the rolling summary if present):
   - Actual word count.
   - Actual start tension, end tension, peak tension on a 1–10 scale.
   - Direction of movement (rising / falling / flat / spiking).
2. Read the same row from `tension-curve.md` and `telemetry.md`. Compute the deltas.
3. Identify drag zones — three or more consecutive scenes where actual tension is ≥2 points below target, or where consecutive scenes are flat when the curve calls for rising tension.
4. Identify trajectory misses — single scenes where direction-of-movement is wrong (e.g., supposed to rise, actually flat or falling).
5. Identify word-count misses — scenes >25% off target (the per-[unit-of-work] `panel-pacing` flags at >20%, so this manuscript-wide pass uses a slightly looser threshold; cumulative miss across many scenes is what matters here).
6. Build the actual-vs-target curve as a markdown table or ASCII chart.
7. Severity per flag:
   - **[severity-minor]**: isolated single-scene miss; the surrounding scenes compensate.
   - **[severity-moderate]**: a single scene at a structural beat (midpoint, climax) that misses; OR a two-scene cluster.
   - **[severity-blocking]**: a three-or-more scene drag zone, or a trajectory miss that breaks the act-level shape.
8. Write the `## Pacing` section to `final-pass-flags.md` with the summary, the curve chart, and the per-flag detail.

## Boundaries

- You measure against the targets set in [setup-phase]. You do not redesign the curve — that's a [back-to-setup] call by [orchestrator].
- You do not rewrite scenes. You flag with reframe candidates; the author ratifies.
- Per-[unit-of-work] pacing audits are `panel-pacing`'s job, not yours.

## Out of Scope (Deferred)

- Automated tension scoring from prose (currently a model judgment call per scene). A future version may calibrate against author-tagged training scenes.
- Cross-book pacing comparison (e.g., does this book's curve match the author's prior books in the same series). Deferred to the series-architect skill.