Venue Bootstrap

Severity Guidelines

• error: Submission will be desk-rejected or flagged as non-compliant if this rule fires. Use for page-limit violations, missing required sections, disallowed content in blind submissions, PDF-integrity failures.
• warning: Formatting deviation that may receive reviewer feedback or a minor reject notice, but is unlikely to cause an outright desk rejection. Use for style deviations, caption formatting, reference formatting.

When in doubt, prefer warning for coverage items that have normative but not mandatory language ("should", "recommended") and error for explicitly mandatory language ("must", "required", "will be rejected").

What Makes a Good Rule Summary

The summary field must describe the concrete requirement, not just the topic area.

Bad (too vague):

• "Abstract formatting requirements"
• "Figures and tables"
• "Citation style"

Good (concrete requirement):

• "Abstract must not exceed 200 words and must appear as a single unnumbered section before the introduction."
• "All figures must include a caption placed directly below the figure; captions must end with a period."
• "References must follow the ACL anthology style; author names must appear as printed in the original publication."
• "Submissions must not exceed 8 pages of content, excluding references. An unlimited number of pages is allowed for references only."
• "Acknowledgements and funding statements must be omitted from the anonymized review submission."

Coverage Status

• agent_review: The topic has explicit requirements in the source text; a rule is checking them at review time. The rule_ids field must list the rule(s) responsible.
• tracked_not_applicable_to_review_pack: A requirement exists in the source text but cannot be checked automatically by this system (e.g., submission-system metadata, pre-submission registration, conflicts-of-interest forms, ethical-review certificates). Use this status conservatively — prefer generating a rule whenever the requirement is detectable in the PDF.

Common Topics to Look for in Venue Sources

Read the normalized source text carefully for the following patterns:

Page and Layout Constraints

• Explicit page limits: main body, appendix, references (different limits for each?)
• Whether appendices are allowed and whether they count against the page limit
• Whether references count against the page limit
• Font size restrictions (e.g., "minimum 10pt", "no font smaller than 9pt in figures")
• Line spacing (single-spaced, 1.5, double — and whether this applies to main body only)

Required and Forbidden Sections

• Whether an abstract is required and if it has a word or line count limit
• Required ethics statement or broader-impact statement
• Required limitations section (NeurIPS, ACL 2023+)
• Required reproducibility checklist or contribution statement
• Acknowledgements: are they allowed in the submission or only in the camera-ready?

Anonymous Submission Restrictions

For double-blind venues, look for restrictions on:

• Self-citations ("we previously showed..." must become "Author et al. showed...")
• Acknowledgements and funding disclosures
• Links to GitHub, project pages, or personal/lab websites that identify the author
• ORCID identifiers or institutional email addresses in the text
• Author metadata embedded in PDF file properties

References and Citations

• Citation style: numbered (IEEE style [1]) vs. author-year (ACL style (Doe et al., 2023))
• Bibliography format: ACL anthology, IEEE, Chicago, APA
• Whether URLs in references must include access dates
• Reference count limits
• Whether self-citations in the reference list are allowed (blind review)

Figures and Tables

• Caption placement: above or below the figure/table?
• Whether all figures must be referenced in the text before they appear
• Resolution requirements for images (e.g., "minimum 300 DPI")
• Restrictions on color figures (e.g., "must be legible in greyscale")
• Whether figure and table numbers must follow a specific convention

PDF and File Format

• PDF-only submissions (no Word, no TeX source)
• File size limit
• Must fonts be embedded? (Type 1, TrueType, OpenType — NOT bitmap)
• PDF/A compliance requirement
• Whether the PDF must be text-searchable (not scanned images)
• Submission-system-specific constraints (anonymized PDF, blind PDF)

What NOT to Generate

• Do not generate rules that duplicate the deterministic hard rules already in the pack (geometry.page_box, geometry.columns, pdf.integrity, semantic.forbidden_headings, semantic.pattern_blacklist).
• Do not generate rules for topics that are mentioned in the source only as background or examples, not as requirements.
• Do not speculate about requirements that are not explicitly stated in the normalized source text.
• Do not generate rules with a check_type other than agent.review.
• Do not include rules whose source_ids cite a source that does not contain relevant evidence for that rule.

Operating Constraints

• Preserve user-supplied official documents under rule_packs/<venue>/sources/official/....
• Create a new versioned pack directory such as rule_packs/<venue>/<year>/.
• When an LLM provider is configured, treat agent review as the primary venue-bootstrap reviewer and rewrite the draft pack in place.
• Use the local deterministic draft as a baseline and fallback, not as the final authority for review-only topics.
• Reuse existing check_type values when the source text clearly matches current execution capabilities.
• If a rule cannot be mapped cleanly, leave it out of rules.json for now and record it in policy_notes.md or formatting_coverage.json.
• Never claim that a new venue is fully automated unless regression cases exist.
• Keep source URLs, hashes, and local paths traceable.

Bootstrap Outputs

• source_manifest.json
• rules.json
• tests.json
• formatting_coverage.json
• policy_notes.md

Minimum Review Pass After Bootstrap

1. Confirm every source URL.
2. Classify each source as policy, checklist, or formatting.
3. Add the first batch of executable rules.
4. Add regression cases before release.