Decision model design
Architectures, training objectives, distillation, constrained outputs, and alternatives to autoregressive decision making.
Online conference · 20 December 2026
International
Conference
on JeV.
A research meeting for JeV and System One decision models. Fast decisions, rigorous evidence, and an open scientific conversation.
ICJV 2026 / FIRST EDITION / ONLINE
THE RESEARCH SPACE01 / DECISIONS
INPUTState + structured questions
System OneSTRUCTURED
DECISION MODELS
DECISION MODELS
choice
score
probability
A conceptual view of the field, not a performance claim.
Submission window begins11 October 2026
Paper deadline · UTC+814 November 2026
Meet online20 December 2026
01 / About the conference
Better decisions.
Stronger evidence.
ICJV brings together researchers studying JeV and related models that turn context into structured decisions for software.
We welcome work on model design, calibrated probabilities, efficient inference, and real-world decision systems. Our focus is on understanding when these models work, when they fail, and how to evaluate them fairly.
Original methods, careful replications, benchmark contributions, and well-supported negative results all belong in the conversation.
ICJV is an independently organized conference. No affiliation with or endorsement by TypeSafe AI or ICLR is implied.
02 / Call for papers
From model to decision.
We invite research on JeV and the broader System One ecosystem, including the following themes.
Calibration & abstention
Probability quality, uncertainty estimates, distribution shift, selective prediction, and knowing when to defer.
Efficient inference
Latency, throughput, serving cost, edge deployment, and reproducible comparisons under realistic workloads.
Routing & agent control
Semantic routing, tool selection, workflow decisions, and coordination between fast decision models and reasoning models.
Safety & robustness
Adversarial evaluation, failure analysis, human oversight, and reliable decision boundaries in deployed systems.
Benchmarks & replication
Datasets, evaluation protocols, open implementations, reproducibility studies, and informative negative results.
03 / Important dates
Mark your calendar.
All times are Beijing time
Asia/Shanghai · UTC+8
| Milestone | Date / time |
|---|---|
| Submissions open | |
| Paper submission deadline | |
| Initial reviews due | 28 Nov 2026 |
| Author response | 29 Nov – 3 Dec 2026 |
| Review discussion | 4 – 7 Dec 2026 |
| Decisions | 8 Dec 2026 |
| Final papers | 15 Dec 2026 |
| Online conference |
All dates refer to the 2026 conference cycle. Review milestones are the organizing schedule; consult OpenReview for the current workflow configuration. Conference sessions and access details will be announced here.
04 / Open scientific exchange
Public review.
Independent judgment.
Our proposed review process follows an ICLR-inspired approach: a public scientific record with double-anonymous reviewing.
Share the evidence
Anonymized papers and revisions become publicly readable after the submission deadline.
Independent assessment
We plan three human reviews per paper, with expertise matching and conflict-of-interest checks.
Respond and refine
Anonymous reviews, author responses, and moderated public discussion support a transparent exchange.
Keep the record open
Meta-reviews and decisions are public. Both accepted and rejected papers retain their public review records.
Openness with clear boundaries. Reviewer identities remain anonymous; author identities are protected during double-anonymous review. Private contact details, conflicts, and administrative communications are not part of the public record. AI assistance must be disclosed and protect confidentiality; human chairs make final decisions. The venue is deployed. Public-release settings and author guidelines are being finalized; the live OpenReview permissions determine current visibility.
05 / For authors
Make your claims
testable.
Strong submissions explain their assumptions, document their methods, and show the limits of their conclusions.
- Report decision quality and probability calibration.
- Measure latency and cost under stated conditions.
- Include appropriate baselines and failure cases.
- Document data, model versions, and evaluation settings.
- Provide reproducibility materials where possible.
OpenReview venue created
Visit ICJV on OpenReview.
The conference is hosted on OpenReview. Use the conference page for submission access and current author instructions. Paper deadline: November 14, 2026, at 23:59 Beijing time.
Open conference portalThe existing OpenReview address contains 2027; this conference takes place in 2026. Consult the submission form for required fields and license terms.
What kinds of contributions are welcome?
Original research, replications, benchmark and dataset studies, and well-supported negative results relevant to JeV and System One decision models.
Do I need to use the commercial JeV model?
No. Work on related System One decision models and open implementations is within scope, provided its connection to the conference themes is clear.
How will I attend the conference?
The conference will take place online on December 20, 2026. The program, access details, and registration information will be announced later.
Is the conference affiliated with ICLR or TypeSafe AI?
No affiliation or endorsement is claimed. The review proposal is inspired by the open-review model used by ICLR; JeV is the research topic.
06 / Get in touch
Questions about ICJV?
Program chair & conference contact: Yesian Rohn
yesianrohn@gmail.com
For submissions, reviewing, and conference enquiries.