UAI Workflow
Use this as the project-management layer over the other UAI skills. UAI is a conference
run by the AUAI: there is no editor-in-chief and no rolling submission — one deadline
per year, per-edition chairs, and PMLR publication funded by registration rather than
author fees. The 2026 calendar below is the verified anchor (checked 2026-07-08);
substitute the current year's official dates before planning anything.
The 2026 shape of a UAI year
| Phase |
2026 anchor |
Owner-side work |
| Fit decision |
months before |
Confirm uncertainty is the contribution (uai-topic-selection) |
| Submission opens |
Jan 25 |
OpenReview profiles, conflicts, subject areas ready |
| Paper deadline |
Feb 25, 23:59 AoE |
Single PDF (8pp main + appendix), ≤15 MB, ZIP ≤50 MB |
| Bidding |
Mar 2–9 |
Nothing — routing was set by your abstract |
| Reviews written |
Mar 21 – Apr 11 |
Prepare response infrastructure; draft anticipated objections |
| Author response |
Apr 23 – May 2 |
Evidence-anchored replies (uai-author-response) |
| Reviewer–AC discussion |
May 2–8 |
Silence; the response argues for you |
| Notification |
Jun 1 |
Triage outcome; celebrate or re-route |
| Camera-ready |
Jul 27 |
PMLR files, public artifacts (uai-camera-ready) |
| Conference |
Aug 17–21, Amsterdam |
Tutorials Mon, main Tue–Thu, workshops Fri |
Backward plan from the paper deadline
Offsets assume a probabilistic-methods paper with theory and experiments; shift for
your mix, and re-anchor to current official dates:
- T−10 weeks: contribution sentence frozen; identifiability/guarantee target named;
ladder of experiments designed (exact-truth, stress, real).
- T−8: core theorem believed and sketched; assumption set labeled A1…An and no
longer moving.
- T−6: exact-truth experiments done; the theorem's prediction visibly matches or
the paper's story changes now, not at T−1.
- T−4: full draft in the UAI template; appendix skeleton mirrors main-text
numbering; baselines tuned and logged.
- T−3: internal mock review against the four criteria (correctness, novelty,
backing, clarity) by someone outside the project.
- T−2: stress and real-data experiments locked; determinism ledger and manifests in
the artifact; supplementary ZIP assembled from a clean export.
- T−1: anonymity sweep of PDF and ZIP; page/size caps verified mechanically;
OpenReview form fields drafted.
- T−0: upload early in the day AoE-wise; re-download and cold-read.
Role assignments that prevent the classic failures
# uai-cycle-owners.yaml — one name per row, no shared ownership
anonymity_sweep: # PDF + ZIP + metadata; leaks are unfixable after the deadline
evidence_map: # every claim -> exhibit; updated when any result changes
appendix_integrity: # body statements == appendix statements after every edit
artifact_build: # clean-export ZIP, manifest files, size cap
openreview_form: # profiles, conflicts, subject areas, abstract == PDF abstract
response_marshal: # archives submitted PDF; drafts objection bank before reviews
camera_ready: # PMLR forms, de-anonymization, public repo tag
presentation: # poster, spotlight, registration, travel/visa timeline
The two most damaging single points of failure at UAI specifically: an unowned
anonymity sweep (appendices inside the reviewed PDF widen the leak surface), and an
unowned appendix-integrity check (deadline-week edits to body theorems silently orphan
their proofs).
Between decision and conference
- Accepted: the June-to-July window carries camera-ready, artifact publication, and
registration simultaneously — assign them to different people on day one.
- Rejected: run the axis triage from
uai-review-process within a week while context
is fresh; the natural re-route targets (AISTATS, NeurIPS/ICML, CLeaR, or a journal)
have their own calendars, and reviews in hand are an asset that depreciates.
- Either way: at least one author may be called to review for UAI — the volunteer
agreement signed at submission — so protect reviewing bandwidth in March–April.
The cycle archive
One deadline per year means institutional memory is the compounding asset. At each
transition, file into a shared, dated directory:
- The exact uploaded PDF and ZIP (deadline day) — responses must quote these.
- The OpenReview form contents as submitted (screenshot or export).
- All reviews, your responses, and the meta-review (decision day).
- The camera-ready as published, plus the PMLR entry once the volume is live.
- A one-page retrospective: which criterion drew fire, which anticipated objection
materialized, what the next cycle should do differently.
Groups that skip the retrospective re-learn the same reviewer objection annually.
Contingency table
| If this happens |
Then |
Because |
| Theorem breaks at T−3 |
Demote it to a conjecture + empirical study, or pull the paper |
A wrong proof fails correctness; a scoped claim can still pass backing |
| Key experiment unfinished at T−1 |
Cut the claim, not the caveats |
Claims without exhibits fail the backing criterion |
| Coauthor conflict list incomplete at deadline |
Fix profiles before the form, even if tight |
Conflict errors poison assignment invisibly |
| Reviews arrive harsher than expected |
Run axis triage before drafting anything |
Emotion-first responses waste the short window |
| Accepted but presenter cannot travel |
Check the current remote-presentation rules immediately |
2026 said "physically or remotely", but details were 待核实 |
| Rejected with fixable-axis reviews |
Target the next honest-fit deadline within the month |
Review memory and momentum decay fast |
First-cycle notes
For a group's first UAI attempt: budget one extra week for OpenReview profile setup
and conflict entry across all authors; expect the two-column template to shrink an
apparently-fitting draft by half a page; and have the mock reviewer score the four
posted criteria explicitly — first-time submissions most often lose on backing, not
on correctness.
Standing warnings
- Never inherit dates from this file or from memory; the official important-dates page
is the only calendar.
- Registration fees, in-person requirements, and remote-presentation options were not
fully verifiable for 2026 at pack-check time (待核实) — confirm before booking travel
or promising remote attendance.
- One-deadline venues punish schedule slips brutally: a week late here is a year late.
Output format
[Cycle position] fit / build / submitted / response / decided / camera-ready / conference
[Next hard date] <official date + source URL>
[Critical path] <three tasks gating the next transition>
[Owner gaps] <rows in the owner map without a name>
[Re-route plan] <standing answer to "if rejected, then where">
Source: brycewang-stanford/Awesome-Journal-Skills → UAI-Skills/skills/uai-workflow/SKILL.md
1---2name: uai-workflow3description: Use when planning a UAI cycle end to end, from venue-fit confirmation through the winter submission deadline, the spring review and author-response phases, the June decision, the July PMLR camera-ready, and the August conference, as one backward-planned calendar with named owners for anonymity, evidence, and packaging tasks.4---5
6
7# UAI Workflow
8
9Use this as the project-management layer over the other UAI skills. UAI is a conference
10run by the AUAI: there is no editor-in-chief and no rolling submission — one deadline
11per year, per-edition chairs, and PMLR publication funded by registration rather than
12author fees. The 2026 calendar below is the verified anchor (checked 2026-07-08);
13substitute the current year's official dates before planning anything.
14
15## The 2026 shape of a UAI year
16
17| Phase | 2026 anchor | Owner-side work |
18|---|---|---|
19| Fit decision | months before | Confirm uncertainty is the contribution (uai-topic-selection) |
20| Submission opens | Jan 25 | OpenReview profiles, conflicts, subject areas ready |
21| Paper deadline | Feb 25, 23:59 AoE | Single PDF (8pp main + appendix), ≤15 MB, ZIP ≤50 MB |
22| Bidding | Mar 2–9 | Nothing — routing was set by your abstract |
23| Reviews written | Mar 21 – Apr 11 | Prepare response infrastructure; draft anticipated objections |
24| Author response | Apr 23 – May 2 | Evidence-anchored replies (uai-author-response) |
25| Reviewer–AC discussion | May 2–8 | Silence; the response argues for you |
26| Notification | Jun 1 | Triage outcome; celebrate or re-route |
27| Camera-ready | Jul 27 | PMLR files, public artifacts (uai-camera-ready) |
28| Conference | Aug 17–21, Amsterdam | Tutorials Mon, main Tue–Thu, workshops Fri |
29
30## Backward plan from the paper deadline
31
32Offsets assume a probabilistic-methods paper with theory and experiments; shift for
33your mix, and re-anchor to current official dates:
34
35- **T−10 weeks**: contribution sentence frozen; identifiability/guarantee target named;
36 ladder of experiments designed (exact-truth, stress, real).
37- **T−8**: core theorem believed and sketched; assumption set labeled A1…An and no
38 longer moving.
39- **T−6**: exact-truth experiments done; the theorem's prediction visibly matches or
40 the paper's story changes now, not at T−1.
41- **T−4**: full draft in the UAI template; appendix skeleton mirrors main-text
42 numbering; baselines tuned and logged.
43- **T−3**: internal mock review against the four criteria (correctness, novelty,
44 backing, clarity) by someone outside the project.
45- **T−2**: stress and real-data experiments locked; determinism ledger and manifests in
46 the artifact; supplementary ZIP assembled from a clean export.
47- **T−1**: anonymity sweep of PDF and ZIP; page/size caps verified mechanically;
48 OpenReview form fields drafted.
49- **T−0**: upload early in the day AoE-wise; re-download and cold-read.
50
51## Role assignments that prevent the classic failures
52
53```yaml
54# uai-cycle-owners.yaml — one name per row, no shared ownership
55anonymity_sweep: # PDF + ZIP + metadata; leaks are unfixable after the deadline
56evidence_map: # every claim -> exhibit; updated when any result changes
57appendix_integrity: # body statements == appendix statements after every edit
58artifact_build: # clean-export ZIP, manifest files, size cap
59openreview_form: # profiles, conflicts, subject areas, abstract == PDF abstract
60response_marshal: # archives submitted PDF; drafts objection bank before reviews
61camera_ready: # PMLR forms, de-anonymization, public repo tag
62presentation: # poster, spotlight, registration, travel/visa timeline
63```
64
65The two most damaging single points of failure at UAI specifically: an unowned
66anonymity sweep (appendices inside the reviewed PDF widen the leak surface), and an
67unowned appendix-integrity check (deadline-week edits to body theorems silently orphan
68their proofs).
69
70## Between decision and conference
71
72- Accepted: the June-to-July window carries camera-ready, artifact publication, and
73 registration simultaneously — assign them to different people on day one.
74- Rejected: run the axis triage from `uai-review-process` within a week while context
75 is fresh; the natural re-route targets (AISTATS, NeurIPS/ICML, CLeaR, or a journal)
76 have their own calendars, and reviews in hand are an asset that depreciates.
77- Either way: at least one author may be called to review for UAI — the volunteer
78 agreement signed at submission — so protect reviewing bandwidth in March–April.
79
80## The cycle archive
81
82One deadline per year means institutional memory is the compounding asset. At each
83transition, file into a shared, dated directory:
84
85- The exact uploaded PDF and ZIP (deadline day) — responses must quote these.
86- The OpenReview form contents as submitted (screenshot or export).
87- All reviews, your responses, and the meta-review (decision day).
88- The camera-ready as published, plus the PMLR entry once the volume is live.
89- A one-page retrospective: which criterion drew fire, which anticipated objection
90 materialized, what the next cycle should do differently.
91
92Groups that skip the retrospective re-learn the same reviewer objection annually.
93
94## Contingency table
95
96| If this happens | Then | Because |
97|---|---|---|
98| Theorem breaks at T−3 | Demote it to a conjecture + empirical study, or pull the paper | A wrong proof fails correctness; a scoped claim can still pass backing |
99| Key experiment unfinished at T−1 | Cut the claim, not the caveats | Claims without exhibits fail the backing criterion |
100| Coauthor conflict list incomplete at deadline | Fix profiles before the form, even if tight | Conflict errors poison assignment invisibly |
101| Reviews arrive harsher than expected | Run axis triage before drafting anything | Emotion-first responses waste the short window |
102| Accepted but presenter cannot travel | Check the current remote-presentation rules immediately | 2026 said "physically or remotely", but details were 待核实 |
103| Rejected with fixable-axis reviews | Target the next honest-fit deadline within the month | Review memory and momentum decay fast |
104
105## First-cycle notes
106
107For a group's first UAI attempt: budget one extra week for OpenReview profile setup
108and conflict entry across all authors; expect the two-column template to shrink an
109apparently-fitting draft by half a page; and have the mock reviewer score the four
110posted criteria explicitly — first-time submissions most often lose on backing, not
111on correctness.
112
113## Standing warnings
114
115- Never inherit dates from this file or from memory; the official important-dates page
116 is the only calendar.
117- Registration fees, in-person requirements, and remote-presentation options were not
118 fully verifiable for 2026 at pack-check time (待核实) — confirm before booking travel
119 or promising remote attendance.
120- One-deadline venues punish schedule slips brutally: a week late here is a year late.
121
122## Output format
123
124```text
125[Cycle position] fit / build / submitted / response / decided / camera-ready / conference
126[Next hard date] <official date + source URL>
127[Critical path] <three tasks gating the next transition>
128[Owner gaps] <rows in the owner map without a name>
129[Re-route plan] <standing answer to "if rejected, then where">
130```
131
132---
133
134**Source:** [`brycewang-stanford/Awesome-Journal-Skills`](https://github.com/brycewang-stanford/Awesome-Journal-Skills) → `UAI-Skills/skills/uai-workflow/SKILL.md`