PRESUBMISSION Mode Integration

How the script-backed PRESUBMISSION layer plugs into each paper-audit mode. For the deterministic rule list, severity calibration, and AI-tone term ban list, see references/PRE_SUBMISSION_RULES.md.

Read next:

• references/PRE_SUBMISSION_RULES.md for the rule taxonomy and term list.
• references/MODE_GUIDE.md for full mode workflows.

Purpose

PRESUBMISSION is a deterministic final-week mechanical layer (em dashes, AI-tone term frequency, abstract completeness, LaTeX citation/label/equation hygiene, paragraph-shape weak signals, concrete captions). It is not a public mode; it runs inside existing modes.

Mode Behavior Matrix

Mode	PRESUBMISSION role	Promotion	Gate effect
quick-audit	inline mechanical readiness signals	n/a	n/a
gate	advisory + Critical blockers	n/a	only Critical fails the gate
re-audit	regression comparison for mechanical findings	n/a	n/a
deep-review Phase 0	script context for reviewer lanes	full/editor focus may promote high-signal items into pre_submission_readiness lane	n/a

For focused theory, literature, methodology, and logic reviews, PRESUBMISSION findings stay in Phase 0 context only. They never become focused-bundle lane issues.

Severity Mapping

PRESUBMISSION source taxonomy	quick/gate severity	deep-review bundle severity
CRITICAL	Critical / P0	major + gate_blocker=true
MAJOR	Major / P1	moderate
MINOR	Minor / P2	minor or Phase 0 only

gate fails only on Critical script findings or failed checklist items. Major and Minor PRESUBMISSION findings stay advisory.

PDF vs Source Behavior

Check group	LaTeX/Typst source	PDF mode
em dash scan	run	run
banned AI-tone frequency	run	run
abstract five-element check	run	run
long paragraph / topic-sentence weak signals	run	run
LaTeX citation tie hygiene	run	skipped
label space/hyphen rules	run	skipped
numbered equation reference rules	run	skipped
source-caption rules	run	skipped

Skips must be visible in script output as ignored comments or metadata, never as issues.

Provenance Rules

• All PRESUBMISSION findings keep [Script] provenance.
• Findings must anchor to source text, line number, or section.
• Reviewer-facing prose must not silently absorb mechanical findings as if they were reviewer judgment.
• The skill audits only; it does not rewrite paper source. Route rewriting requests to the format-specific writing skill.

Integration with gate

1. gate reads PRESUBMISSION findings together with its own checklist.
2. Critical findings convert to gate blockers.
3. Major and Minor findings appear in advisory recommendations after the verdict and EIC screening.
4. IEEE pseudocode rules: distinguish mandatory from IEEE-safe recommendation.

Integration with deep-review

1. audit.py --mode deep-review runs PRESUBMISSION as part of Phase 0.
2. Focus routing decides what happens to those findings:
   • --focus full or --focus editor: high-signal items can be promoted into the pre_submission_readiness lane and consolidated alongside reviewer findings.
   • --focus theory|literature|methodology|logic: keep PRESUBMISSION in Phase 0 context only; do not surface them as focused lane issues.
3. Promotion does not bypass consolidation. Promoted items still flow through consolidate_review_findings.py and verify_quotes.py.

Integration with re-audit

re-audit compares the new PRESUBMISSION findings to the previous run. Status labels follow the standard re-audit schema: FULLY_ADDRESSED, PARTIALLY_ADDRESSED, NOT_ADDRESSED, NEW.

Mechanical regressions are reported alongside reviewer-finding regressions but keep their [Script] provenance so the user can distinguish style drift from substantive reviewer concerns.