# Audit Stage

For the current host, apply review-policy.md to the real current manuscript,
sources and PDF/DOCX renders. Check scientific support, editorial completeness,
citation identity/context and final cited coverage, template/layout fidelity,
and portable requested packages. Use citation.md, manuscript-format.md and
submission.md at their actual work points. Repair located failures and reuse
valid checks of unaffected content; initial clean review needs no fictitious
revision. Missing author facts limit submission, not safe local improvement.

The remaining ordered commands, named artifacts and progress gate document the
legacy orchestrator. Use them only when that actual tool consumes the old tree;
they do not create a second current-task completion authority.

## Purpose

Audit all PaperSpine outputs before declaring the workflow complete.

Read `execution-efficiency.md`. During repair, rerun checks whose bound inputs
changed. When the manuscript, bibliography, figures, configuration, and target
are frozen, run the complete command set below once. An exact execution receipt
may reuse deterministic computation, but cannot waive a readiness or inspection
receipt.

## Required Checks

1. Artifact completeness.
2. Reference material workspace has `source_index.md`.
3. Contribution and aligned motivation were user-confirmed after research;
   `contribution_check.py` passed before planning.
4. Balanced mode has a substantive `section_blueprints.md`; strict mode also
   requires an ordered `writing_rationale_matrix.md`.
5. `results_validation.md` passed during planning for journal, conference, and
   competition scenes.
6. `structured_review.md` contains a completed, manuscript-grounded editor
   synthesis before LaTeX assembly. Strict mode also requires a passing
   `evidence_review.json`/`evidence_review_check.md`, then `reviewer_audit.md`
   and its checker. Provider failure is never treated as a literature no-hit.
7. No append-only or shallow revision for substantive rewrite tasks.
8. Logic transfer from original draft or materials.
9. Claim support from user evidence.
10. LaTeX citation, label, figure safety.
11. `citation_support_bank.md` follows its literature contract: open-literature
    breadth/recency, or closed-corpus exhaustive deduplication and honest scope.
12. `scientific_evidence_ledger.json` resolves S/C/N/M/O/R links and final
    verification states for evidence-bearing scenes.
13. Final LaTeX source plus hash-bound rendered page/figure inspection.
14. Word output structurally valid by default; skip only when
   `word_output=none` is explicit in config.
15. Submission metadata passes its check. When publication-cycle work is in
    scope, the current target profile passes; a submission/revision has a READY
    immutable bundle and matching archive checksum; a rebuttal passes atomic
    coverage and lineage checks; and a transfer has completed the destination
    rebuild after a READY_TO_REBUILD delta.
16. Publication surface contains no internal evidence IDs or gate narration.
17. Usage telemetry contains measured usage or explicit
    `telemetry_unavailable` receipts; file bytes never stand in for billed tokens.
18. Translation coverage complete when `translation_package=zh`.
19. When `translation_package=zh` and Word output is enabled,
    `final_paper/paper.zh.docx` and `word_report.zh.md` exist and pass. The
    `translation_zh/` folder is an audit/intermediate package, not the final
    Chinese Word deliverable.
20. If the opt-in adaptive shadow ran, its receipt may certify only one
    hash-bound `claim_surface_audited`. It cannot substitute for any canonical
    check, whole-manuscript completion, target-bundle READY, or authorization.

## Scripts

```bash
python scripts/integrity_audit.py paper_rewriting_output --markdown --write
python scripts/artifact_check.py paper_rewriting_output --markdown --write
python scripts/citation_bank_check.py paper_rewriting_output/citation_support_bank.md --markdown --write
python scripts/scientific_evidence_check.py paper_rewriting_output --phase final --markdown --write
python scripts/publication_surface_check.py paper_rewriting_output --markdown --write
python scripts/visual_readiness_check.py paper_rewriting_output --markdown --write
python scripts/metadata_readiness_check.py paper_rewriting_output --markdown --write
python scripts/usage_ledger.py paper_rewriting_output --markdown --write
python scripts/progress_check.py paper_rewriting_output --markdown --write
python scripts/revision_audit.py <original> <revised> --markdown
python scripts/structured_review.py paper_rewriting_output --markdown --write
# strict mode only:
python scripts/structured_review.py paper_rewriting_output --dispatch
python scripts/evidence_grounded_review.py validate paper_rewriting_output/evidence_review.json --manuscript paper_rewriting_output/final_paper/main.tex --markdown --write
python scripts/citation_quality_audit.py paper_rewriting_output --write
python scripts/latex_guard.py <main.tex> --bib <references.bib> --markdown
python scripts/word_guard.py paper_rewriting_output/final_paper/paper.docx --tex paper_rewriting_output/final_paper/main.tex --markdown --output paper_rewriting_output/word_report.md
python scripts/word_guard.py paper_rewriting_output/final_paper/paper.zh.docx --tex paper_rewriting_output/final_paper/main.tex --markdown --output paper_rewriting_output/word_report.zh.md
# when a target-specific publication cycle is in scope:
python scripts/publication_cycle.py profile-check <profile.json> --markdown --write
python scripts/publication_cycle.py rules-check <profile.json> <manuscript.tex> --phase writing --markdown --write
python scripts/publication_cycle.py assemble <profile.json> <plan.json> <bundle-dir> --markdown
python scripts/publication_cycle.py rebuttal-check <review_round.json> --markdown --write
python scripts/publication_cycle.py transfer-plan <origin-profile.json> <destination-profile.json> <transfer-request.json> <delta-dir> --markdown
```

## Required Outputs

- `integrity_audit.md`, `artifact_check.md`, `revision_audit.md`
- `structured_review.md` and `citation_quality_audit.md`; strict mode also
  requires `evidence_review.json`, `evidence_review_check.md`, and
  `reviewer_audit.md`; `logic_transfer_audit.md`
- `scientific_evidence_check.md`, `visual_audit_manifest.json`,
  `visual_readiness_check.md`, `publication_surface_check.md`
- `submission_metadata.json`, `metadata_readiness_check.md`,
  `usage_ledger.jsonl`, `token_budget_by_stage.md`
- A passed `target_profile_check.md` and writing-stage
  `journal_rules_precheck.json/.md`; a READY `bundle_manifest.json`, final
  `journal_rules_final.json/.md`,
  `submission_bundle.zip`, and matching `submission_bundle.sha256` (when applicable)
- A passed `rebuttal_check.md` and rendered response/change artifacts (when applicable)
- `transfer_delta.json` plus the newly rendered destination paper and READY
  destination bundle (when applicable)
- `final_paper/paper.zh.docx` and `word_report.zh.md` when
  `translation_package=zh`

Do not declare the task complete if required artifacts are missing, claims are
unsupported, the integrated editor synthesis is still pending, translation is
partial, a strict-mode rationale matrix is generic, or the final Chinese Word
document is missing. A green canonical manuscript audit does not override a
blocked target package, rebuttal, or transfer rebuild.

## Output Directory Rules

For the current host, keep outputs under the existing task's `paper/` directory and retain valid prior versions. The historical `paper_rewriting_output/` layout may be read in place; it is not a reason to relocate or delete user work.

- **Accidental nesting:** verify which files belong to which version and repair only genuinely incorrect references or placement. Preserve all unique content and resolve filename collisions before a justified move; do not automatically flatten directories.
- **Multiple final-paper directories:** determine the current version from the task and file identities, preserve previous versions, and correct the selected paths. Do not delete a sibling directory merely because its name is `final_paper`.
- **No misplaced artifacts:** `writing_rationale_matrix.md`,
  `citation_support_bank.md`, `research_dossier.md`, and other workflow
  artifacts belong inside `paper_rewriting_output/`, not outside it.

## Completion Hard Gate

Before declaring the workflow complete, run the checks below in order.
`progress_check.py --gate final_audit` is the authoritative hard gate: it
re-runs `artifact_check.py`, `citation_bank_check.py`, `integrity_audit.py`,
`citation_quality_audit.py`, evidence, visual, metadata, publication-surface,
and usage-ledger checks, the required `word_guard.py` check, and — once
`final_paper/main.tex` exists — `latex_guard.py` and `section_economy_check.py`,
then fails on any non-zero exit code. In balanced mode,
`section_economy_check.py` is advisory; strict mode may explicitly enforce the
configured section budget.
Do not treat existing report files as enough evidence of completion.

These last two read the manuscript body, not just report shapes:
`latex_guard.py` fails literal-bracket citations that are not real `\cite`
links and out-of-sync numbering; `integrity_audit.py` fails writing-process /
meta-narrative language (supervisor or reviewer mentions, "reorganized the
paper", transcribed `A -> B -> C` plan chains) leaking into the prose; and
`section_economy_check.py` surfaces unusually fragmented sectioning for
editorial review. Section count alone does not fail a balanced paper. A clean
report file is not enough — the body and completed editor synthesis must agree
that the paper is ready.

```bash
python scripts/artifact_check.py paper_rewriting_output --markdown --write
python scripts/citation_bank_check.py paper_rewriting_output/citation_support_bank.md --markdown --write
python scripts/scientific_evidence_check.py paper_rewriting_output --phase final --markdown --write
python scripts/publication_surface_check.py paper_rewriting_output --markdown --write
python scripts/visual_readiness_check.py paper_rewriting_output --markdown --write
python scripts/metadata_readiness_check.py paper_rewriting_output --markdown --write
python scripts/usage_ledger.py paper_rewriting_output --markdown --write
python scripts/progress_check.py paper_rewriting_output --gate final_audit
python scripts/integrity_audit.py paper_rewriting_output --markdown --write
python scripts/progress_check.py paper_rewriting_output --markdown --write
```

When `word_output` is not explicitly `none` and `output_language` is not `zh`,
also run:

```bash
python scripts/word_guard.py paper_rewriting_output/final_paper/paper.docx --markdown --output paper_rewriting_output/word_report.md
```

When `translation_package=zh` or `output_language=zh`, also run:

```bash
python scripts/word_guard.py paper_rewriting_output/final_paper/paper.zh.docx --markdown --output paper_rewriting_output/word_report.zh.md
```

When `translation_package=zh`, also run and require PASS:

```bash
python scripts/translate_guard.py paper_rewriting_output --markdown --write
```

The final Chinese Word file must be predominantly Chinese and free of visible
Markdown emphasis markers such as `**bold**` or `*italic*`. A `.zh.docx` file
that contains English body prose under a Chinese title is a failed translation
package.

You may declare completion only when `progress_check.py --gate final_audit`
exits 0, `artifact_check.py` exits 0, final progress reports
`is_complete=true`, no `misplaced_artifacts` are reported, integrity audit has
no unresolved BLOCKER, and all five independent readiness dimensions are true:
`scientific_content_ready`, `visual_ready`, `citation_verified`,
`metadata_ready`, and `artifact_portable`. A PASS in one dimension cannot
offset a failure in another. Word output must be present and valid unless the
user explicitly opted out. If pandoc is unavailable, write BLOCKED/FAIL in
`latex_report.md`; do not silently skip Word or claim the workflow is complete.

If the full pass fails, repair the owning input, rerun its affected checks, and
then perform one new final full pass. For the same command and unchanged input
hashes, allow only one corrected retry; a third materially identical attempt is
a blocker, not progress.

For a ProductRunner task, this hard gate consumes the single layered readiness
verdict rather than treating submission metadata as part of every local job.
`manuscript` and `local_delivery` scopes may complete with
`submission_ready=false` when their own blocker lists are empty; author-only
facts remain visible only under `submission_package` and external-action
layers. A `submission_package` request still requires all five non-compensatory
tiers. `external_action_authorized` remains false in every readiness verdict.

## Anti-Pass-Through Rule

**If `artifact_check.md` reports Status: FAIL or Status: BLOCKED, the workflow
is not complete.** Do not declare completion. Do not write `progress.md` with
`is_complete=true`. Return to the failing upstream stage:

- Missing artifacts → run that stage.
- Content issues (weak rationale matrix, thin citation bank) → fix the
  artifact, then re-run `artifact_check.py`.
- Misplaced artifacts → move them into `paper_rewriting_output/`.

**If `citation_bank_check.md` reports Status: FAIL, the citation support bank
is not qualified.** The final audit must not pass until the bank is re-run and
all weak rows are strengthened with reference format + claim-support sentences.

In strict mode, a present but generic rationale matrix fails. Balanced mode does
not require the matrix and must not block a scientifically supportable paper for
missing row-level paperwork.