Every point where this tooling (the guide, the Internal Review Board) differs from BOLDiquette § ML Conference Cycle. This is the list to work through when requesting changes to BOLDiquette.
Provenance: BOLDiquette § ML Conference Cycle (Pre-submission / Post-submission / Arxiv / Publicity / Rebuttal / On Acceptance / Pre-Conference) · text read 2026-09-10 · also as plain DIFF.md
New Process or tooling BOLDiquette doesn’t cover at all.
Divergence Our pages currently say something different from BOLDiquette.
Proposed A change we want made to BOLDiquette; until it lands, the implementation follows BOLDiquette.
New
1. Internal Review
BOLDiquette
Pre-submission says only “Submit a first full draft to the compulsory BOLD PI on your paper 10 days ahead of the final deadline.” No structured runway, no review, no roles, no checklist.
Implementation
The pre-submission run is a fixed timeline, and the first full draft goes through a simulated full peer review (venues are all similar).
8 wk before the deadline — register the paper (abstract, outline, author list, target venue, compute estimate) on the Internal Review Board.
6 wk — Pitch Day: every paper is pitched to the group.
4 wk — first full draft → Internal Review, worked through by a junior and a senior reviewer from the pool.
2 wk — final draft, every review comment incorporated.
1 wk — the PI reads the final draft and approves it for submission. That is the PI’s only involvement.
The reviewer pool = every paper’s authors plus volunteers, with at least one senior.
12-item checklist (6 Format, 6 Science; each item folds together several related points to keep the tick count manageable), on every board card’s detail page. A junior and a senior reviewer are assigned per submission on the board; each fills in the checklist. A paper can’t be marked Approved until both passes are complete.
Proposed BOLDiquette change
Add a new subsection between Pre-submission and Post-submission (drop-in prose in proposed-boldiquette-addition.md). Supersedes the old “first draft to the PI 10 days ahead” line.
New
2. Internal Review Board
BOLDiquette
Nothing.
Implementation
The Internal Review Board, one board per venue + year. Columns mirror the guide’s 9 phases, named past tense for the checkpoint already met to be sitting there: Registered → Pitched → Drafted → Reviewed → PI Approved → Submitted → Posted → Reviews Out → Accepted — each column is an attestation of a real-world checkpoint being met, not a work-in-progress bucket. Within Drafted a card is colour- and badge-coded In review / Changes requested / Approved (derived from the checklist, plus a reviewer-settable flag); the outcomes Accepted / Rejected / Withdrawn are a separate badge, not to be confused with the Accepted column (the final draft actually reaching the venue’s own decision). Per-card fields: compute estimate; post-submission — submission link (OpenReview or other), rebuttal deadline, rebuttal doc link, review notes. The venue carries the call for papers and submission-page links, the paper deadline, an optional explicit Pitch Day, and its own dates — reviews released and rebuttal deadline as standard fields, abstract / notification / camera-ready as optional — plus page limit and anonymity. From the paper deadline the board derives and clearly displays one chronological deadline list — register (−8 wk), Pitch Day (explicit, or −6 wk), first draft (−4 wk), final draft (−2 wk), PI approval (−1 wk), then the venue’s own dates — and exports the whole list as a downloadable .ics.
Proposed BOLDiquette change
Covered by the same draft addition plus its “Addition to Post-submission” block.
Known limitation (not a BOLDiquette question): status changes are a free-form dropdown, except the Approved gate (both checklist passes must be complete). Who may move a card / tick which pass is unenforced — see TODO.md Permissions.
Divergence
3. Where the project is logged
BOLDiquette (Pre-submission)
“Add your project 4 weeks ahead of the deadline to the [project tracker] spreadsheet.”
Implementation (Phase 1)
Register the paper on the Internal Review Board 8 weeks ahead (title, abstract, authors, venue, compute estimate); the card starts in Registered.
Proposed BOLDiquette change
Replace the project-tracker instruction with the Internal Review Board, and move the timing from 4 to 8 weeks.
Proposed
4. Pre-submission timeline
Implementation is ahead of BOLDiquette.
BOLDiquette
Register 4 weeks ahead; first full draft to the PI 10 days ahead; clear your calendar for the 4 weeks before the deadline.
Implementation
The real timeline (per Mattie, who ran this before) — 8 wk abstract + outline + compute → registered · 6 wk Pitch Day · 4 wk first full draft → Internal Review · 2 wk final draft with all comments in · 1 wk PI approval · 0 submit.
Proposed BOLDiquette change
Replace the Pre-submission timing lines with this runway.
Divergence
5. Who reviews the draft
BOLDiquette
“The compulsory BOLD PI on your paper” reviews the first full draft.
Implementation
The first full draft is reviewed by a junior and a senior reviewer from the pool (peers). The PI is involved only in the week-1 polish.
Proposed BOLDiquette change
Fold into the new Internal Review subsection (§1).
New
6. Compute estimate at registration
BOLDiquette
No mention of compute at the pre-submission stage.
Implementation
An estimate of the compute the paper will need is required at registration (8 weeks out) and stored on the board card.
Proposed BOLDiquette change
Add “a compute estimate” to what’s required at registration.
New
7. Pitch Day
BOLDiquette
No pitch requirement.
Implementation
Every paper is pitched to the group at a fixed Pitch Day, at least 6 weeks before the deadline; a paper that hasn’t been pitched doesn’t go forward.
Proposed BOLDiquette change
Add Pitch Day to Pre-submission.
Divergence
8. “On Acceptance” step dropped
BOLDiquette (On Acceptance)
“Submit the form to add the paper to our website.”
Implementation
The guide has no “On Acceptance” phase. Once BOLD OS is live the project (and its publications) already exist there, and an accepted publication is published to the website automatically — see TODO.md.
Proposed BOLDiquette change
Replace the manual form step with “BOLD OS publishes the accepted publication automatically” once that exists; until then the manual step still applies and the guide is ahead of reality here.
Divergence
9. Presentation format
BOLDiquette (Pre-Conference)
“Please use Google Slides for your presentation to facilitate collaboration.”
Implementation (Phase 9)
Slides can be Google Slides or an HTML deck — whatever is easiest to build (often with an agent) and share.
Proposed BOLDiquette change
Loosen to “a format co-authors can collaborate on and exchange easily (Google Slides, an HTML deck, …)”.
New
10. Pre-Conference networking
BOLDiquette (Pre-Conference)
Poster, practice talk, slides format, example posters, “posters are not papers”. No networking guidance.
Implementation (Phase 9)
Adds two items — email people you want to meet at the conference at least 3 weeks beforehand to set up a meeting; and attend as many of the conference’s social events as you can (mandatory), since they’re usually the best networking.
Proposed BOLDiquette change
Add both to Pre-Conference, including the mandatory framing on social events.
Reconciled — no longer a difference
arXiv affiliation. Guide now uses “BOLD, {University of Oxford / Imperial College / UCL}”, matching BOLDiquette exactly (was Oxford-only). Fixed 2026-09-10.
Roger Grosse attribution. The guide credits “Roger Grosse” for the loose-ends practice; BOLDiquette has a typo, “Rober Grosse”. Editorial only — worth fixing in BOLDiquette when the section is next edited.