# Reader Feedback Template

Copy this template to `feedback/round-{N}/{reader-kebab}.md` for each reader
in each feedback round. `reader-kebab` identifies the reader (their name or a
label like `alpha-reader-1`); `{N}` is the round number.

```yaml
---
reader: "{Reader name or label}"
round: {N}
chapters-read: "{e.g. 1-12, or all}"
overall-verdict: "{loved it | liked it with reservations | mixed | didn't connect}"
---

# Feedback — {Reader} (Round {N})

## Overall Impression

{2-4 sentences: what the reader felt about the manuscript as a whole.}

## What Worked

- {Specific things the reader liked, with chapter or scene references}

## Problems

### {Problem title} (Ch {N})

- **What the reader said:** {quote or close paraphrase}
- **Where:** {chapter/scene reference}
- **Canon check:** {verified against the bible | contradicts canon — see note |
  outside canon scope}
- **Severity (reader's):** {blocking | major | minor | nit}

## Questions Raised

- {Questions the reader asked that the manuscript should answer — these are
  candidate `continuity/questions/` entries or signs of a clarity gap}

## Suggested Changes

- {Concrete changes the reader proposed — record as proposals, not decisions}
```

## Canon check discipline

Before a reader's note enters synthesis, check it against the story bible:

- **Verified:** consistent with canon; the note is about craft or clarity.
- **Contradicts canon:** the reader's expectation conflicts with established
  canon — usually means a *setup* problem (canon wasn't conveyed), not a
  canon problem. Note which file establishes the canon fact.
- **Outside canon scope:** the reader wants a different book (different
  genre, different theme). This is a `declined-with-reason` candidate.

Never "fix" canon to satisfy a reader note without user approval — that is a
story decision, not a maintenance fix.

# Synthesis Template

Write the round synthesis to `feedback/round-{N}/synthesis.md` only after
every expected reader file for the round is collected. The synthesis is the
decision record: which notes become revision work, which are adjudicated,
and which are declined with a recorded reason.

```yaml
---
round: {N}
readers: [{reader-1-kebab}, {reader-2-kebab}, ...]
readiness: needs-revision   # ready | needs-revision | not-ready
---

# Synthesis — Round {N}

## Readiness Verdict

**{ready | needs-revision | not-ready}** — {one sentence: why}

- `ready`: no blocking or major convergent issues; proceed to the next
  stage (next round, or export).
- `needs-revision`: convergent or adjudicated issues require a revision
  pass before the next round. Lists the revision plan below.
- `not-ready`: structural problems (broken premise, failed arc, pervasive
  confusion) require developmental rework, not a polish pass.

## Convergent Findings (≥2 readers agree)

| Finding | Readers | Chapters | Canon check | Action |
|---------|---------|----------|-------------|--------|
| {One-line finding} | {a, b} | {Ch N} | {verified / setup-gap} | {Fix planned} |

Convergent findings are the strongest signal: two independent readers
tripping on the same spot means the manuscript, not the reader, is at fault.
These become revision items by default.

## Divergent Findings (readers disagree)

| Finding | For (readers) | Against (readers) | Adjudication | Action |
|---------|---------------|-------------------|--------------|--------|
| {One-line finding} | {a} | {b} | {Which side, and why — cite canon, craft, or intent} | {Fix / decline} |

Adjudication rules:

- Check both sides against the bible and the premise. Canon and theme
  outrank taste.
- A 2-vs-1 split is convergent (see above), not divergent.
- When genuinely 1-vs-1 with no canon tiebreak, prefer the reader whose
  overall verdict was more positive — they engaged more deeply — or ask the
  user. Record who decided.

## Single-Reader Findings

| Finding | Reader | Severity | Weight | Action |
|---------|--------|----------|--------|--------|
| {One-line finding} | {a} | {major/minor} | {accept / investigate / decline} | {...} |

Weigh single-reader findings by specificity: a specific, locatable note
("the alibi in Ch 9 contradicts the timeline") outranks a vague one ("the
middle dragged"). Specific + canon-verifiable = investigate or accept.
Vague + taste-based = usually decline.

## Declined with Reason

| Finding | Reader(s) | Reason for declining |
|---------|-----------|----------------------|
| {One-line finding} | {a} | {e.g. contradicts premise; outside genre contract; reader wanted a different book} |

Every declined finding needs a recorded reason. "The author disagrees" is
not a reason. Reasons reference canon, premise, genre contract, or craft
principle.

## Revision Plan

Numbered, actionable items for the revision pass, each with file targets:

1. {What changes} — targets: {chapter/scene/entity files}
2. ...

Hand this plan to the `revision-continuity` skill for execution.
```

## Synthesis checklist

- [ ] Every expected reader file for the round is collected (no synthesis
      on a partial set — see the discipline rule in `SKILL.md`)
- [ ] Each finding appears in exactly one category
- [ ] Every convergent finding has an action (fix or explicitly deferred)
- [ ] Every divergent finding has a recorded adjudication with a reason
- [ ] Every declined finding has a recorded reason (not just disagreement)
- [ ] The readiness verdict matches the findings (blocking convergent
      issues cannot coexist with `ready`)
- [ ] The revision plan lists concrete file targets for each item