Guide
TMLR explained: 70.6% accepted, a differently shaped bar
TMLR has abolished, in writing, the reject reason that ends most competent conference papers: not novel enough, not state of the art, contribution too modest. It kept the other reason and named it the most important.
Published 14 August 2026
Holding a conference rejection? TMLR removes the reason most of them give, so the next question is where this paper goes now. What to do after a rejection
Read this before you decide
- TMLR does not accept expanded versions of conference papers. Not a longer version, not one with added experiments or restored proofs, and no reuse of text, figures or results from work published, accepted or under review elsewhere.
- That manuscript is the one JMLR asks for. If this describes your paper, you are on the wrong page.
- If the work is new, TMLR judges it on two questions only. Are the claims supported by accurate, convincing and clear evidence, and would some individuals in its audience be interested. The first is named the most important.
- Novelty is not a criterion at all. That is the whole reason this venue exists.
The criteria
Every line of your meta-review, sorted
Conference objections split into the ones TMLR has ruled out of bounds and the ones it holds you to harder.
| What the meta-review said | Where it lands at TMLR | The published wording |
|---|---|---|
| The method is not novel enough | Out of bounds | Not a necessary criteria for acceptance. |
| Incremental, it does not beat the state of the art | Out of bounds | The interest criterion cannot be used to reject work for not reaching a new state of the art. |
| The contribution is modest | Out of bounds | Accept papers that meet the criteria, even if the contribution is modest. |
| Only a narrow set of people would care | Weakly binding | Whether some individuals would be interested, and an unsure reviewer must assume they would. |
| The experiments do not support the claims | The first criterion | The most important criterion, answered yes or no. The cheap fix is allowed: reduce the claims instead. |
| The paper is hard to follow | The first criterion | Criterion one covers the clarity of the narrative and arguments; unclear writing is on the reject list. |
| You are claiming credit for published work | Binding, and asymmetric | Incorrectly claiming novelty over published work is a listed reject reason. Lacking novelty is not. |
| It just re-runs someone else's experiments | Binding | A class report re-running experiments can fail the interest test; a robustness study with actionable lessons passes. |
Paraphrased from TMLR's acceptance criteria and editorial policies, read 14 August 2026. The left column is our wording, not TMLR's; tinted rows still bite. The reviewer guide has the form: each criterion yes or no, requested changes marked critical or merely strengthening, then accept, leaning accept, leaning reject or reject. No score to average.
The sentence worth quoting back at a reviewer
The criteria say it outright: the interest criterion should not be used to reject work that is not considered significant or impactful because it is not achieving a new state of the art, nor to reject a method considered not novel enough. Those three words, significant, impactful and novel, are named as terms TMLR explicitly avoids. If a review leans on one of them, it is departing from the page its author was asked to read.
A claims and evidence gap has two exits, not one
Reviewers are told the gap must be addressed, and that more experiments are not the only way: another is for the authors to reduce their claims. That move barely exists in a conference rebuttal, where the paper is fixed.
The disqualifier
The extended conference paper goes to JMLR
Not a ladder, a fork. The FAQ: send shorter manuscripts to TMLR, extended works that incorporate previously published material to JMLR.
| TMLR | JMLR | |
|---|---|---|
| Extended conference papers | Ineligible, and named as such. | Invited, if it goes into much greater depth and extends the results substantively. |
| What counts as reuse | Text, figures or results from work published, accepted, or under review elsewhere. | Restored proofs, minor variations of previous experiments and extra background do not qualify. |
| Overlap that is allowed | Venues publicly declared non-archival, such as workshops, plus arXiv preprints. | |
| A narrow audience | Accepted by design: an unsure reviewer is told to assume some individuals would be interested. | May be unsuitable if the board finds the audience too narrow, and applications to other domains are out. |
| Surveys | Not considered for review as of 1 September 2026. | By invitation only, never unsolicited. |
| A paper a conference rejected | Not overlap, on our reading: the ban covers published, accepted or parallel work. |
Overlap from the TMLR editorial policies, routing and surveys from the TMLR FAQ, the JMLR column from JMLR's author information, read 14 August 2026. Empty cells are positions JMLR does not state.
Surveys: the door shut on 1 September 2026, and the site disagrees with itself
The FAQ answers it in one line: as of 1 September 2026, TMLR no longer considers survey papers for review. That date has passed, so a survey has nowhere to go here now, and the rest of the site still has not caught up: re-read on 10 September 2026, Survey Certification is documented word for word in both the editorial policies and the reviewer guide.
The real clock
Nine weeks is the target, three months is the measurement
| Figure | Time to decision | Where it comes from |
|---|---|---|
| Official target | about 9 weeks | No guarantee attached, and longer above 12 pages of main body. |
| Measured median, short papers, 2025 | 91 days | The annual report, improved from 97 days in 2024. Short means conference length. |
| Measured median, long papers, 2025 | 104 days | Same report, improved from 113 days. TMLR puts the ideal case at 60 to 70 days, and JMLR at over 200. |
| Decisions available | 3 | Accept as is, accept with minor revision, or reject, with the action editor saying whether a revised version would be considered. No revise and resubmit. |
From the TMLR FAQ and the 2025 TMLR annual report, read 14 August 2026. Reviews are due two weeks from assignment, four above 12 pages of main content, and the discussion phase runs two to four weeks with revisions allowed. A revised paper returns as a new submission linking the rejected one.
Prestige, counted
84.4% of TMLR papers carry no certification at all
Significance did not disappear, it moved into a badge. TMLR says so itself: it relies on certifications such as Featured and Outstanding to carry the more speculative judgements about significance and impact.
| Certification | Papers | Share | What it actually says |
|---|---|---|---|
| None | 3,752 | 84.4% | The ordinary outcome, and not a weak one. |
| Journal to Conference | 282 | 6.3% | New in October 2025, the ticket to the NeurIPS, ICLR and ICML joint track: a poster slot, no further review, 150 places each. Deadlines: NeurIPS 2026, 26 September; ICLR 2027, 18 December. |
| Event | 251 | 5.6% | Not a quality award: TMLR papers presented at other conferences. |
| Survey | 124 | 2.8% | Exceptionally thorough or insightful. Submissions closed on 1 September 2026, the badge is still documented. |
| Featured | 114 | 2.6% | TMLR's comparison: an oral or spotlight at a top-tier conference. |
| Expert | 99 | 2.2% | About the authors, not the paper: written by TMLR's Expert Reviewers. |
| Reproducibility | 72 | 1.6% | Reproduction adding value through extra baselines, ablations or analysis. |
| Outstanding | 4 | 0.09% | Four in the journal's history, called a best paper equivalent. |
Our own count of all 4,444 entries on the official TMLR papers listing on 14 August 2026. A paper can hold more than one badge, so the shares do not sum to 100. Where they overlap the count matches TMLR's 2025 report. Definitions from the editorial policies. The joint track bars anything its authors already published at a conference, and our count finds the badge on 7.6% of the 2026 papers.
The acceptance rate is two numbers, and everyone quotes the one they found
TMLR's 2025 report publishes both: 70.6% if withdrawn and desk-rejected submissions are ignored, 46.3% if they are counted. Only the denominator changed. The second fell from 50.6% in 2024, which TMLR puts down to a much higher desk-rejection rate after an influx of weak submissions.
Featured is getting rarer per paper, not commoner
It went to 26 papers in 2023, 38 in 2024 and 24 in 2025, and 22 of the 1,230 published in 2026 to 14 August. Output climbed steeply across those years, so the badge TMLR compares to a conference oral is scarcer every one of them.
The fine print
Five rules that will bite you
Publishing is free in either direction and papers are CC BY 4.0, copyright retained. The rest can cost you a submission.
| The rule | What it costs |
|---|---|
| Revealing your name on a rejected submission | It precludes submitting a revised version to TMLR, permanently. Rejected submissions stay public on OpenReview and their authors are offered the chance to unmask. Desk rejections stay private. |
| The annual submission budget | The Generalized Harmonic Quota Rule, N1 = 2 and N9 = 9, doubled for reviewers and action editors. Every author needs budget left, desk rejections spend it, and duplicate accounts may draw a lifetime ban. |
| Touching the stylefile | Any change to formatting, font or layout may mean rejection without review. The clear desk rejections: not anonymised, wrong stylefile, or overlapping with published or parallel work. |
| Inappropriate use of LLMs | More than 20 instances identified in 2025, each investigated, with actions up to a one-year submission ban. |
| No DOIs, and no second thoughts | DBLP records none for the venue and Crossref returns nothing for ISSN 2835-8856; JMLR's record reports zero too. Camera-ready versions are final. |
Quota and licence from the editorial policies, with its calculator; desk rejections from the action editor guide; the stylefile from the submission page; DOIs from DBLP. The LLM figure is TMLR's own count, not an audit.
Free tool, no account
Where should this paper go?
Paste the title and the abstract. It ranks the venues that publish this kind of work, with the reach and the article charge OpenAlex has on record.
Whether TMLR is the room is the first question
TMLR will not reject you for being unfashionable, and it will reject you for claiming more than your evidence carries. Which rooms want the paper you actually have is answerable from the abstract alone.
The matcher needs no account. Why papers get rejected sorts the objections by whether they are recoverable, and what AI review cannot do is the limit of any such read.
phdflow reads a finished draft against the guidelines of the venue you are aiming at and returns an accept or reject call with the objections behind it. Scored against 697 real decisions: AUC 0.91 and 96% accuracy on 297 ICLR 2025 papers, one false accept in 149 rejects. The full measurement. Read a finished review before you pay for one. Your first paper gets a free preview; the full report is 60 credits, €9. No account needed. Pricing.