Use when auditing a COLT (Conference on Learning Theory) submission for CMT readiness, the 12-page PMLR-format main body, single-PDF assembly with unlimited appendix, double-anonymous formatting, the parallel-submission ban covering journals and proceedings venues, and the ordered final-week checks before the AoE deadline.
NewMaintained
Automatic
◆Research pack
Last verified 8/31/2026
What it changes
Passed 20 of 23 test cases with the skill, and 20 of 23 without it.
Without skill87%
With skill87%
Checks it passed that the model alone missed
Identifies Microsoft CMT as the required platform.
Identifies Microsoft CMT as the submission portal.
Limits the desk-reject risk output to a single item.
and 1 more in the full report
3 fixed17 passed either way0 still failing3 broke
Pass-rate lift
0 pp
Tokens
+65%
Agent turns
0%
Response time
-18%
Model gemini-3.6-flashMeasured 8/31/2026v120 of 23 cases passed with skill
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 3 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 3
Prompt
We are preparing to submit our theoretical machine learning paper to COLT and need a complete pre-submission compliance audit. Evaluate our draft and provide a structured report with these exact fields:
- [COLT readiness]
- [Blocking checks] (addressing portal setup, main text page limits, unified PDF formatting, author blinding, and concurrent submission constraints)
- [Proof-integrity risk] (evaluating potential alignment issues between statements and proofs, unverified lemmas, or improper use of external results)
- [Desk-reject risk]
- [Fix order] (a prioritized action list to complete before the submission cutoff)
What was checked, without and with the skill
Checked
Without skill
With skill
Identifies Microsoft CMT as the required platform.
Missed
Passed
Uses the unnumbered bracketed header format specified in the output schema.
Passed
Passed
Limits the desk-reject risk output to a single item.
Missed
Passed
case-01 · gemini-3.6-flash · 8/31/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md130 linessingle file
COLT Submission
Run this audit before uploading a learning-theory paper to COLT. Every number below was
checked against the COLT 2026 Call for Papers on 2026-07-08; reopen the current CFP at
learningtheory.org before trusting any of it in a later cycle, because COLT republishes
its rules from scratch each year.
What the 2026 cycle actually required
Platform: Microsoft CMT, not OpenReview. The COLT 2026 site pointed authors to
https://cmt3.research.microsoft.com/COLT2026. COLT has used other systems in the
past, so never assume the portal — follow the link in the live CFP.
Deadline: February 4, 2026, Anywhere on Earth, for the 39th edition (June 29 to
July 3, 2026, San Diego). One deadline for the whole manuscript; the 2026 CFP had no
separate abstract-registration step, unlike NeurIPS-style venues. Confirm each year.
Length: main body at most 12 PMLR-formatted pages excluding references, with no page
limit on references and appendices, all submitted as one PDF.
Anonymity: submissions must be suitable for double-anonymous review; the official
LaTeX template's anon option suppresses author names.
Scope: theoretical aspects of machine learning at the intersection of computer
science, statistics, and applied mathematics, explicitly including theory that
explains empirical phenomena.
The single-PDF discipline
Because the appendix rides inside the same PDF as the body, COLT has no separate
supplement portal to forget — but also no second chance to patch a broken proof file.
Treat the appendix as reviewable text, not as an attachment:
Compile body plus appendix together and check the PDF page count where the 12-page
boundary falls; references must start after the body, appendices after references.
Nothing decision-critical may exist only in the appendix. A reviewer who stops at
page 12 should already believe the theorem statements and the proof strategy.
Anonymize the appendix with the same care as the body: acknowledgements, grant
numbers, code URLs pointing at personal repositories, and PDF author metadata all leak.
Illustrative template skeleton
The exact style file name changes with the year (e.g., a colt2026 package for the
2026 cycle). Shape, not filename, is the stable part: