Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →brycewang-stanford/colm-workflow
Use when planning a COLM submission campaign across the calendar — working backward from the late-March abstract and paper deadlines through the May-June rebuttal, July decisions, August camera-ready, and October conference, coordinating co-authors, compute, and reciprocal-reviewing duties, and slotting COLM into a multi-venue LM-research pipeline.
Passed 23 of 25 test cases with the skill, and 13 of 25 without it.
and 10 more in the full report
10 fixed13 passed either way2 still failing0 broke
Tokens, turns and time describe resource use, not a dollar price.
One of 10 test cases the skill fixed. Same model, same prompt, with and without the skill.
I am an advisor leading a small team aiming for COLM. We are currently conducting core benchmark evaluations, but I want to make sure we do not run into bottleneck risks with co-author policies or unverified claims. Please review our project state and output a tracking block with the following sections: - Target cycle and deadline confirmation status - Active phase - Critical path bottleneck - Ledger health showing verified count vs total and flagged items - Group constraints including reciprocal map status and high-volume submission authors - Next milestone date and required deliverable status
| Checked | Without skill | With skill |
|---|---|---|
| The response includes the tag [Phase now] to indicate current stage. | Missed | Passed |
| The response selects experiments as the active phase value based on benchmark evaluation status. | Missed | Passed |
The instructions your model reads when the skill loads.
COLM gives you exactly one shot per year: one deadline (late March in 2026), one rebuttal, one decision (early July), one October stage. Campaign planning is therefore mostly backward induction from March — and, because LM projects burn compute and involve many hands, the critical path is usually GPUs and co-author availability, not writing.
| Date (2026) | Event | What must already be true |
|---|---|---|
| February | OpenReview opens | Profiles current for every author |
| March 26 | Abstract deadline | Real abstract; reciprocal reviewer chosen |
| March 31 | Full paper deadline | Frozen claims, pinned evidence, anonymized package |
| May 22 | Reviews released | Response owners pre-assigned; cache warm |
| May 22 - June 8 | Rebuttal | ~18 days; plan in colm-author-response |
| July 8 | Decisions | Both branches pre-planned |
| August 7 | Camera-ready | Public artifacts ready to ship simultaneously |
| October 6-9 | Conference, San Francisco | Presenter, visa, poster all handled |
Live-phase note: on this pack's access date — 2026-07-08 — the 2026 cycle is at
decision day. If that is you, jump to colm-camera-ready or the reject branch in
colm-review-process. If you are planning fresh work, you are planning COLM
2027: dates unannounced (待核实), with late March the reasonable working
assumption to be corrected the moment colmweb.org posts the real calendar.
T-20w (Nov) route the project (colm-topic-selection); if COLM, freeze the
one-sentence finding and the evaluation design — contamination
handling decided NOW, not retrofitted (colm-experiments)
T-16w (Dec) compute budget locked: training runs, eval sweeps, API spend;
book cluster time — March demand is when everyone else trains too
T-12w (Jan) main experiments running; pinning discipline live from day one
(colm-reproducibility); related-work lanes assigned (colm-related-work)
T-8w (Feb) full draft v1; internal red-team review against the four attacks;
OpenReview profiles + reciprocal-reviewer map done when the venue opens
T-4w (Mar) freeze claims; kill or hedge anything unverified; build the
test cases passed, vs 13 / 25 without
Published skills whose descriptions closely match this one, with their headline benchmark lift.
More skills from the same source and nearby in the registry, with their headline benchmark lift.