The BetBuddy football betting robot

It means a scheduled program, not a person and not a tipster. The same code runs on every fixture, in the same order, with the same thresholds. Nobody overrides a selection because it looks good, and nobody removes one after it loses. The rules are written down in the full methodology and every bet the robot has ever taken stays in the public results history, win or lose.

The models are statistical: a Poisson goal model, an Elo rating and a recent-form component, blended into a single probability. That is machine analysis, not a large language model and not a self-learning neural network — and we do not describe it as one.

The robot's starting point is market activity rather than a fixture list. It reads a dropping-odds feed for football 1X2 markets and stores every price observation as a new timestamped snapshot, never overwriting the previous one. That is what makes a price movement a measured change between two real observations rather than an assertion.

A match enters consideration when its price has moved meaningfully on genuine traded volume. Thin markets move on a single large ticket, so a minimum-volume requirement is applied before any movement is treated as informative at all. Matches whose competition, kick-off time or team pairing cannot be identified cleanly are discarded before any modelling happens.

No selection may ever be taken below decimal odds of 1.80. That floor is written into the code as a hard constant — configuration can raise it but cannot lower it. Prices are also checked for freshness: a stale snapshot disqualifies the candidate outright rather than being used with a caveat.

Qualification is a set of gates, not a score. Volume, price movement, snapshot freshness, data quality and a positive edge against the margin-free fair probability must all pass. Failing any single gate ends the candidate; the gates are never traded off against one another. On most days most fixtures qualify for nothing, and that is the intended output of a system with real thresholds.

Nothing is locked more than 120 minutes before kick-off. Inside that window a qualifying candidate becomes provisionally locked and is then re-checked continuously: if the price falls below the minimum, if the market data goes stale, or if the candidate stops qualifying, it is unlocked and dropped. Only a bet that survives to execution is recorded as taken, at the price recorded at that moment.

Each taken bet is staked in parallel by four independent staking strategies — fixed stake, percentage bankroll, fractional Kelly and capped recovery — on the identical selection at the identical price. Each keeps its own isolated bankroll, so the comparison between them isolates the staking plan alone. The exact production formulas, caps and rounding rules are documented on the strategy comparison hub.

Final scores are read from several independent public scoreboards, so grading does not depend on any single paid provider. A bet is graded only once a match is confirmed final. Grading is idempotent: a match cannot be settled twice, and settlement writes the result, the profit or loss and the new bankroll into a permanent ledger.

• Paper mode only — no bookmaker account is connected and no real money is staked.
• A hard minimum price of 1.80 on every selection and every market.
• Per-bet stake caps as a share of bankroll and as a multiple of the base stake.
• Strictly sequential ledgers: a strategy cannot open its next bet until the previous one has settled.
• Implausibly large edges are treated as a data-quality signal, not as an opportunity.

It is not guaranteed-profit software, and no output here should be read as a prediction of a certain outcome. It estimates probabilities, and probability estimates are wrong a predictable share of the time by design. It does not place real bets, hold funds or connect to a bookmaker. If you are new to any of this, start with the plain-English overview and read the responsible gambling guidance before acting on anything.