About DeepBall

Three principles

Deterministic work stays in code: data, odds, quant probabilities, settlement and hit rates are computed, while LLMs only interpret, challenge, adjudicate and write.

Every conclusion is traceable: evidence carries a source and timestamp, the red team cannot be disabled, the chief’s verdict comes with a factor ledger, and the quant model’s independent call is kept on record.

Handicap / spread and total calls are fully public, including misses and passes, settled on the last pre-kick-off version, so the hit rate is not inflated; correct score and total goals only show probabilities and stay out of the record, with results in the settlement section of each match page.

How to read a report

Start with the status line: how much of this analysis you can read right now — free today, purchased, included with membership, or preview only.

The core call: the direction and adjudicated probability from the chief, with the evidence grade (stars), tier and edge beside it.

Why: the main evidence, the main risks, and the flip conditions that would change the call.

What the red team says: the strongest challenges and reverse scenarios it raised, which the chief weighs one by one.

Analysts and data: the calls on the remaining markets, each specialist’s own view, and the basic and advanced data panels.

Versions and review: every update leaves a version record; after the match the original call and its settlement are published.

Probability, edge and evidence grade

Verdict probability: the chief's final chance for this pick after weighing every analyst.

Market implied: the probability behind the market consensus odds with the margin removed — what the market thinks.

Edge: verdict probability minus market implied; a positive number means we rate the pick higher than the market does.

Evidence grade (1–5 stars): computed by code from the gap, agreement with the model, the verified share of the factor ledger and data completeness, not a hit rate — four stars does not mean it lands eight times in ten.

The tier follows from edge and evidence grade together: an edge of 6 points or more with 4 stars or more is “strong”; 3 points or more with 3 stars or more is “lean”; everything else is “pass”. Without a market line the call is always a pass. Distribution calls such as scorelines and goal bands only carry a probability — no stars and no tier.

The chief weighs every analyst, the red team and the quant model, then sets its own direction, probability and a self-rated confidence (kept only for calibration; the stars on the page are the evidence grade computed by code), and writes a factor ledger: the side each factor favours, how many points it moves the probability, its source and whether it is verified. Code is only a safety net: no market price, too few quant samples or a competition outside the recommendation scope means a pass; a pass from the chief is a pass; when the chief sides with the model and the market-implied probability sits inside the model’s uncertainty interval, it is a pass; when the side the chief picks is the market underdog (implied probability below 50%), it gets a probability but no recommendation; the tier never exceeds the chief’s suggestion.

Where the evidence comes from

Every piece of evidence carries its source and timestamp — match data, odds, intel or the quant model. The English site shows the translation and marks any block still in the report’s original language.

Odds come from a multi-book “market consensus”; no bookmaker name or link appears anywhere on the site. Sources we cannot identify are labelled “unidentified source” rather than passed off as verified fact.

When it updates

First version: generated automatically 24 hours before kick-off.

Regular updates: once 2 hours before kick-off and once after the starting line-ups are published.

Line-move reruns: a partial rerun when the football line moves by 0.25 or more, the basketball line by 1.5 or more, or the price by 0.1 or more; at least 30 minutes between reruns, and no line-move rerun within 20 minutes of kick-off.

Early analysis: members can request it for matches more than 24 hours from kick-off that have no report yet — 3 per day, resetting at midnight in the site time zone.

Versions and settlement: each update creates a new version and older ones stay available; updates stop at kick-off, and settlement uses the last pre-kick-off version, which then counts towards the public record.

How the record is calculated

Hit rate counts only lean / strong handicap / spread and total calls; a match’s handicap and total count as one call each; every call is settled on the last version before kick-off.

Hit rate = (hits + ½ × half wins) ÷ (hits + misses + ½ × half wins + ½ × half losses)

Pass calls stay out of the hit rate but are listed one by one under History on the records page. The 95% interval next to the hit rate is a Wilson interval that counts half wins and half losses as half a call, like the hit rate itself; the fewer the calls, the wider it gets.

Prices and terms

1 free match per sport per day, fully readable; every other match is a preview.

A full match analysis costs 38 points, an article 10 and a video 15; unlocked content stays visible forever.

Unlocking a match analysis does not include that match’s article or video — the three are charged separately.

Full rules for points, membership and refunds are in the help centre; every settled handicap / spread and total call is listed in the platform record, and score and total-goals results are on each match page.

The analyst team

Seven roles work in a fixed order: five specialists analyse in parallel, the red-team review is mandatory, the chief rules and QA checks the result; every step cites its sources and timestamps.

Data analyst: Form, home and away splits, head-to-head, attacking and defensive efficiency, xG / net rating / pace.

Intelligence: Injuries, line-ups, rotation, schedule congestion, weather and pitch, motivation and sentiment.

Tactics: Playing styles, formation match-ups and key duels (for football, once both official line-ups are out); for basketball, pace, three-point reliance and the interior battle.

Market: Line moves from opening to live, prices and market heat, judged on the market alone (it does not see the quant model).

Quant: Reads the quant model: probabilities, ranges and edge.

Red team: Argues against the others: overlooked variables, overconfidence and reverse scenarios; cannot be switched off.

Chief: Weighs every view, the red team and the quant model, then sets its own direction, probability and confidence with a factor ledger; every call is settled in public after the match.