Method
HOW THE SYSTEM FINDS BETS.
A pipeline. Built in Sydney. Runs every twenty minutes, 24/7. Compares live AU bookmaker prices against a no-vig consensus of the whole market and emits only the bets that clear edge, EV and cross-book deviation filters. The maths is boring and that’s why it works.
01
Pulls live odds — every 20 minutes.
The pipeline pulls live odds from major AU bookmakers (via The Odds API) on a fixed 20-minute cycle, 24/7. No human pricing markets in real time. No "lead tipster" reading lines on Saturday morning.
02
Compares to the no-vig consensus.
Every price is compared against a no-vig consensus: the median of what the whole market believes once bookmaker margin is stripped out. No single price source is trusted on its own — including the Betfair Exchange, which now acts only as a veto: when fresh exchange data actively disagrees with a candidate, the candidate is dropped. Betfair was the primary benchmark until 10 June 2026; the ledger showed the consensus did the job better, so it was demoted. That change is logged like every other.
03
Scores edge and deviation.
Each candidate gets an adjusted edge score on a 0–100 scale, weighted by edge percentage and cross-book deviation, adjusted for the consensus implied probability. Cross-book deviation measures whether one book has drifted from an otherwise settled market — a single outlier against a stable consensus is a better signal than a price the whole market has already moved to.
04
Filters.
Claimed edge capped at 12% — an edge above that is treated as a data artifact, not a bargain, and rejected. Adjusted edge ≥ 40 (≥ 50 for tennis). Cross-book deviation < 2.5. Best price > 2.00 (short-priced favourites structurally lost the model money and were gated out 19 May 2026). Tiered EV minimums: 1.81–2.50 odds need ≥ 1.5% edge; 2.51–4.00 need ≥ 2.5% edge. If a price looks too good to be true, the model assumes it is.
05
Fires to Telegram + email.
When a price clears every filter, the bet fires to Telegram and email at the same time, with per-channel deduplication. No batching. No Monday slate. If nothing fires for six hours, that’s the model not seeing edge — not a bug. Most twenty-minute cycles produce zero bets.
06
Settles within the hour.
Agent 5 settles bets at :55 past each hour. Closing lines are pre-captured every hour, so every settled row has a CLV figure attached. Daily summary email the next morning, weekly review Mondays, monthly summary first Monday of each month.
07
Public ledger.
Every settled bet joins /results with sport, market, odds at publish, closing price, CLV, units staked and result. Append-only. Rows are never edited after grading — if a result needs a correction, it’s footnoted, not silent.
08
Where the human work happens.
Built and run by one operator in Sydney who is publicly accountable for the ledger. The methodology, the filters, and every change to them are logged. When something breaks, the pipeline raises the alarm before a member notices. No team photo, no panel of experts — the ledger is the credential.
What it doesn’t do
NO AI HAND-WAVE. NO HOUSE TIPS.
- No “AI” in the marketing sense. No black-box machine-learning predictor. The model is a deterministic edge-detection pipeline against a no-vig consensus of published prices. Same inputs, same output.
- No multis. No specials. No same-game parlays. Singles only. The maths on multis is structurally rigged against the punter; we won’t price-launder by recommending them.
- No bookmaker-supplied tips. Bookmaker-published tips target the markets the book wants liquidity in, not the markets where the price is wrong. Different objective, different output.
- No personality bets. No “monster lock”, no “banker”, no “smash this in early”. The model emits bets when prices qualify and stays quiet when they don’t.
18+ only. Informational only. You place your own bets at your own bookmaker. We never hold your money. Past performance is not a guarantee of future results. Gambler's Help: 1800 858 858.