J1 2026-27 -- round of 2026-08-21

Match probabilities for Japan's J1 League. 10 fixtures, fixed before kickoff. The market is measurably better than this model — a disagreement below is a question, not an answer. Full caveat & numbers.

Calibrated model probabilities. Kickoff times are UTC, 2026. No market odds are shown here — see Methodology for why.
Kickoff Home Away P(H) P(D) P(A) Result
Aug 21 10:00 Kashiwa Reysol V-Varen Nagasaki 53.3% 24.3% 22.4% not yet played
Aug 21 10:30 FC Tokyo Chiba 63.4% 19.8% 16.8% not yet played
Aug 22 09:00 Kashima Antlers Avispa Fukuoka 51.1% 27.5% 21.3% not yet played
Aug 22 09:30 Okayama Verdy 40.2% 31.1% 28.7% not yet played
Aug 22 10:00 Cerezo Osaka Shimizu S-Pulse 46.3% 25.0% 28.7% not yet played
Aug 22 10:00 Nagoya Grampus Gamba Osaka 37.3% 25.4% 37.3% not yet played
Aug 22 10:00 Kyoto Mito 54.2% 23.5% 22.3% not yet played
Aug 22 10:15 Sanfrecce Hiroshima Kawasaki Frontale 52.8% 22.1% 25.1% not yet played
Aug 22 10:30 Yokohama F. Marinos Vissel Kobe 35.6% 25.7% 38.7% not yet played
Aug 23 10:30 Machida Urawa Reds 51.7% 24.9% 23.4% not yet played

Over 3,882 backtested J1 matches (2014–2025-26), the closing betting market scored 0.0044 RPS points better than this model on average (95% CI [0.0026, 0.0060]) — see Methodology. This model has no information the market lacks: it does not see injuries, lineups, suspensions or travel. The bold figure in each row is only the most likely of three outcomes, not advice — in most football matches the most likely single outcome still loses more often than it wins.

Metadata & verification

Published (UTC)
2026-08-20 10:29 UTC
Trained as of (UTC)
2026-08-20 10:29 UTC
Model version
footy-ev-phase2-dc-tb-v1
params_hash
ea317b1dbde414553b28ff026b02ecdc2ba3b088e5d242b6cff416d721ed3f7f
File SHA-256
c4a1117a9e07f5e67491a2d0b10033e34bf07989e3a06721a663de0ffcc922be
Market snapshot SHA-256
70ad7fab1506a0d19771e547ae556d7b4598edf7159f48e84718343d7644ad4b
Training window
1772 matches, 29 teams, since 2021
Publish gate
PASS

How to verify

This file is committed as predictions/j1_2026_2026-08-21.json in the project's git repository, with the SHA-256 shown above. On its own, a git commit timestamp is not strong third-party evidence that a prediction existed before kickoff — the author can set it to any date. Independent timestamping (OpenTimestamps), Wayback Machine archiving and a GitHub Actions server-time witness are part of this project's published plan for closing that gap (design document, section 3) and are not yet wired into this build. Until they are, verify what you can today: recompute the SHA-256 of the committed file yourself and confirm it matches the value above, and read the raw JSON — nothing on this page is derived from anything else.