VAR was supposed to fix bad calls. Instead it left fans everywhere asking the same question: how can the same incident be a penalty in one country and a yellow card in another?
This isn't a Premier League problem, or a Bundesliga problem. It's a global one — Champions League, World Cup, qualifiers, lower divisions. Same IFAB Laws of the Game, wildly different applications.
OURVAR.AI runs every clip against the full IFAB rulebook — uniformly, without bias, fatigue, or favouritism. Every verdict shows the exact law clause cited and the full reasoning.
An AI applies the same rules to the same situation every single time. No two referees interpret 'unnatural arm position' the same way. AI doesn't.
VAR checks routinely take 3–5 minutes, killing match momentum. AI analysis of key frames takes seconds — fast enough to be invisible to fans.
Currently, VAR publishes minimal reasoning. With AI, every decision comes with a full explanation: which law was applied, which frame was decisive, and why.
Every AI decision is logged, auditable, and comparable. Referee error rates become measurable facts rather than debatable opinions.
VAR exists in many leagues, but implementation quality varies wildly. AI could bring genuinely consistent, high-quality review to every competition — from the Bundesliga to non-league football — at a fraction of the cost.
Football's legitimacy depends on fans believing the rules are applied fairly. A consistent, transparent, explainable system builds more trust than the current black box.
Every clip becomes a precedent. The verdict isn't the end — it's the start of a community review. Up/down votes, comments, and "I was there" notes turn every case into a discussion the AI listens to.
Some calls aren't just about the letter of the law. A handball that's technically by-the-book can still feel wrong to 90% of fans — and that's exactly the kind of common-sense signal that's hardest for an AI alone. Community sentiment tells the system when context overrides the strict reading.
More cases analysed → more community feedback → a tighter precedent library → better verdicts on the next ambiguous play. The platform improves with every clip you upload.
Every verdict on OURVAR.AI is generated by Anthropic's Claude Sonnet 4.6 — a large language model — guided by a structured system prompt that contains the full IFAB Laws of the Game (2024/25 edition), the relevant interpretation guidance from IFAB Circulars, and a growing library of community-verified precedents drawn from past cases on this site.
We don't ask the model to "have an opinion". We give it the same rulebook a professional referee uses, the same frames a VAR official sees, and the same question — was the on-field call consistent with the Laws? The output is a structured verdict, not a hot take.
When a clip is submitted, the system downloads it (or accepts a direct upload) and extracts the key frames — typically 12 stills sampled around the decisive moment using motion-peak detection via OpenCV. Each frame is downscaled to ~960×540 and encoded as a base64 JPEG for the model. We never store the raw video file beyond the analysis window: only the extracted frames and the final verdict are persisted.
Before the main analysis runs, a smaller Claude Haiku 4.5 pass screens three early frames to confirm the clip actually contains football and isn't explicit or off-topic. This catches mis-submitted media before we pay full analysis cost, and prevents the verdict surface from being abused as a generic image analyser.
The main analysis call sends the full Laws system prompt (cached on Anthropic's side via ephemeral cache_control to keep costs reasonable), the match context (teams, competition, referee, claimed actual decision), the relevant precedents for that incident type, and all extracted frames. Claude Sonnet 4.6 returns a JSON-structured verdict: CORRECT / WRONG / INSUFFICIENT FOOTAGE (with no-call variants), a confidence band, the exact law clause cited, a recommended action, and a frame-referenced reasoning trail.
Every verdict is open to community voting: agree, disagree, or mark debatable. Once five verified votes accumulate, a consensus rating attaches to the case. Comments add context the AI couldn't see — "the camera angle is misleading," "the referee blew before contact," "this exact incident was talked about in IFAB's last circular." Verdicts users mark as correct feed back into the precedent library, tightening future analyses on the same incident type.
On a rolling basis, we sample a blind subset of cases against expert reviewers to measure agreement between AI verdict and human-referee consensus. When agreement on a specific incident type drops below a threshold, we tighten the precedent set fed into future calls for that category, and audit the system prompt section that governs it. The model itself doesn't "learn" between calls — calibration happens via the precedent library and prompt iteration, not via fine-tuning.
OURVAR.AI exists to make football refereeing decisions transparent, comparable, and accountable. That goal only works if our process is visible. The standards below describe what we cite, how we handle errors, where our independence sits, and where our limits are.
The AI's primary source is the IFAB Laws of the Game (2024/25 edition), specifically Law 11 (Offside), Law 12 (Fouls and Misconduct), and Law 14 (The Penalty Kick), together with the supporting Practical Guidelines section. We also reference IFAB's published Circulars when they clarify ambiguities (e.g. the 2024 handball-deflection guidance). Match metadata (teams, scores, competition) comes from the submitter and is cross-checked against publicly-visible match results where possible. We do not scrape broadcaster footage at rest — clips are processed once, frames extracted, and the source file is discarded.
When a verdict turns out to be wrong — either because the AI misread a frame, an incident type was miscategorised, or new footage from a different angle changes the analysis — we re-run the case and append a correction note to the original verdict rather than silently overwriting it. The community vote history is preserved. If you believe a verdict on this site is materially incorrect, the comments section on every case is the fastest route to flag it; the admin queue reviews community-flagged corrections daily.
OURVAR.AI is independent. We have no affiliation with FIFA, UEFA, IFAB, any national federation, broadcaster, or club. We do not accept paid placements that influence verdict outcomes. Subscription revenue covers Anthropic API costs and infrastructure; advertising (where shown to non-premium visitors) is served by Google AdSense and does not see the analysis content. Editorial decisions about which incidents become daily-featured cases are made by the operator alone.
Be aware of the limits. (1) The AI only sees the frames it's given — angles the broadcaster didn't include can change the call. (2) Some incidents are genuinely interpretive: two trained referees can apply Law 12 differently and both be defensible; on those, the AI's confidence drops to LOW and the verdict reads "INSUFFICIENT FOOTAGE" or "DEBATABLE". (3) The AI doesn't know broadcast context (commentary, replay graphics, the referee's mic feed). (4) For matches involving more than 22 visible players, occluded contact, or VAR-only camera angles, the system may flag the analysis as INSUFFICIENT FOOTAGE rather than guess. We'd rather say "we can't tell" than fabricate a confident call.
OURVAR.AI is a community experiment. It's not affiliated with FIFA, UEFA, or any football authority. Think of it as a parallel VAR run by fans, for fans — where every call can be scrutinized, debated, and rated. The more cases we analyze and verify, the better the precedent library gets, and the more useful the verdicts become.
If you've had a match ruined by a terrible call, submit the clip. Let's build the evidence base that shows whether AI can be trusted with the whistle.
OURVAR.AI processes video clips submitted by users for the sole purpose of AI-assisted referee decision analysis. Uploaded footage is used only for frame extraction and analysis — raw video files are not stored after processing. Users are responsible for ensuring they have the right to upload any content they submit.
Analysis performed on this platform is transformative in nature: it generates new data (referee accuracy reports and structured verdicts) rather than reproducing the original broadcast. Footage is not publicly displayed. We believe this use falls within the fair use doctrine (US) and equivalent provisions in other jurisdictions. However, OURVAR.AI makes no legal guarantee and is not a law firm.
If you are a rights holder and believe content on this platform infringes your copyright, please contact us at **hello@ourvar.ai** with: (1) identification of the work, (2) the infringing content, (3) your contact details, and (4) a statement of good-faith belief. We will respond promptly and remove content where required.
OURVAR.AI has no affiliation with FIFA, UEFA, any national federation, broadcaster, or club. All verdicts are AI-generated opinions for discussion and research purposes only.
For legal enquiries: hello@ourvar.ai