# Editorial Completeness and Manuscript Scale

Use this playbook for full-paper journal, conference, and competition work. It
sets the outcome the Agent should pursue without prescribing one universal
outline or turning editorial judgment into another form.

## North Star

The manuscript is the product. Internal artifacts are private scaffolding.
Prefer a stronger paper over a more elaborate record of how it was planned.

A reader should be able to answer, in order:

1. Why did this work need to be done?
2. What was done?
3. What does the evidence show?
4. What do the findings mean, within their real limits?
5. What conclusion or next understanding follows?

Keep these answers aligned across title, abstract, Introduction, Results,
Discussion, figures, and the ending. Do not let those surfaces tell different
versions of the paper.

## Let the Agent Design the Paper

Choose the architecture that best fits the target venue, article type, evidence,
and user request. IMRAD is a useful default for original research, not a naming
test. Results and Discussion may be combined when the venue supports it;
Conclusion may be a separate section or an unmistakable final movement of the
Discussion. What matters is that each intellectual job is actually done.

Do not use word quotas, paragraph counts, top-level-section budgets, or a
rationale matrix as substitutes for this judgment. Venue length ranges and
section proportions are orientation, not targets to pad or ceilings that permit
underdevelopment.

When space is tight, compress repeated numbers and generic background before
method definitions, update/decision rules and evaluation settings. Keep the
method and comparison self-contained; a compact algorithm or settings table
can preserve the necessary detail from the existing study record.

`tier` controls how broadly the Agent researches examples and how much optional
process detail it records. It never reduces the paper promised by `scene`,
`target_name`, the template, or the user's request.

## Develop Evidence into Results

Treat a figure contract or evidence map as the start of a Results unit, not its
finished prose. A strong unit usually gives the reader:

- why this analysis is needed at this point in the argument;
- the visible result, with the decisive comparison or quantitative anchor;
- what that result establishes for the contribution;
- the relevant boundary or alternative explanation when it changes meaning;
- a bridge to the next question.

Use this as a thinking sequence, not a fixed paragraph template. Avoid both
panel-by-panel narration and caption-like compression. Supporting figures should
advance the argument in the text, while legends preserve panel-level detail.

## Make Discussion Do New Work

Discussion should not merely repeat bounded claims from Results. It should
synthesize the main findings, relate them to prior work, explain agreements or
disagreements, consider alternatives, state specific limitations, and make the
field-level implication legible. End with the strongest conclusion the evidence
earns and, when useful, the next unresolved question.

## Read the Rendered Paper in Sequence

After layout, read the PDF from start to finish as a reader would. Check not only
whether every figure exists and is legible, but whether each figure appears near
the Results text that needs it, whether a float interrupts or reverses the
argument, and whether captions and surrounding prose share the same claim.
Reposition floats or revise transitions when reading order obscures the story.

## Integrated Editorial Pass

In balanced mode, perform one substantive editorial pass in free prose. State
the manuscript's central claim, where the evidence is strongest, where the
argument remains thin or discontinuous, and the highest-leverage revision.
Revise the paper before declaring `Review status: PASS`.

The pass should judge the manuscript itself. Do not reward the existence of
plans, matrices, receipts, or review files, and do not convert stylistic advice
into blocking paperwork. Only truth, unsupported primary claims, unresolved
figure/text identity, missing user authority, or unusable deliverables require a
hard stop in balanced mode.