How the day's pick is chosen

A day is sold as one selection. That pick, and only that pick, decides whether the day counts: it wins and one day comes off your balance; it loses or voids and your balance does not move.

Side picks go out with it. They settle on their own and never touch the counter — an extra that loses costs you nothing, and one that wins does not use a day up. They are part of what you receive and no part of what you are charged for.

That rule is not ours to reinterpret later. It is a constant in the settlement code, and the version that judged each day is recorded against the day itself. See the guarantee.

The picks come from one analyst's feed. We do not re-derive them, overrule them, or blend them with a model of our own. The rule below decides which of that feed's picks becomes the day; nothing about the bet itself is changed on the way through.

That is a smaller claim than this site used to make. An earlier rule required three independent analysts on the identical bet before a pick could ship. It is the better rule and it is not available to us: there is one analyst, and measured on live data before the rule was replaced, no two picks ever landed on the same game — the threshold could not be met at three, at two, or at all.

A rule that can never fire is not caution. It is a product that never ships a pick. So it was replaced with one that can, and this page is where that is said out loud rather than left implied.

Only games that have not started, and that start within twenty-four hours, are eligible at all.

The pick has to be actionable when it lands. A fixture two days out invites the line to move underneath you between the moment we commit and the moment you can act, and the price you were shown would not be the price you got.

Inside that window, a pick qualifies only if it is actually placeable. All five of these have to be there:

• a market
• a side
• a price
• a kick-off time
• two named teams

The count of qualifying plays on the picks pages is this same test applied to today's board, run through the same code the selector runs. It is not a marketing number: a page cannot promise a count the product would not produce.

Among the picks that pass, the day takes the one the feed itself marked as the day's. Failing that, the earliest kick-off — the one you have the most time left to act on.

Parlays. The feed sends multi-leg tickets with no market, no side and no league. A day is sold as one selection whose result decides it; an eight-leg ticket is a different product wearing the same word.

Incomplete rows. A pick missing its market, side, price or kick-off is not placeable, and publishing one sells you a sentence you cannot act on.

Diagnostics. Test rows never reach a buyer.

And when nothing qualifies, the day goes out empty. It does not carry the least-bad option instead. An empty day costs you nothing — the counter does not move — so of the two mistakes available, shipping a pick we would not take is the expensive one.

Once chosen, the pick is committed with a hash before kick-off: published to you and to everyone else at the same moment, and impossible to alter afterwards without the hash changing.

You do not have to take our word for it. The recipe for recomputing the hash yourself, the salt released with the result, and a worked example are on the verification page, and every settled day in the archive carries its own hash to check against.

From the league's official result, at official confirmation. Where a case is ambiguous it resolves in your favour — written into the rules rather than decided case by case, because a discretion exercised by the party being paid is not a rule.

A day voids rather than losing when the match is postponed or abandoned, when the bet pushes, when the price at release was below the published minimum odds, when the source returns no result, and when nothing qualified in the window at all. A void never moves the counter.

The full list, and which version of the rules judged which day, is on the guarantee page.

Counts, and nothing computed from them. The record carries no win rate, no yield, no ROI and no units. Not pending — absent, and structurally so: the code that produces it reads a single column and never loads a price or a stake, so there is no number in scope for a later edit to divide by another.

The reason is not modesty. A percentage drawn from a sample this size says far more than it knows, and a figure about your future money is a claim we cannot evidence and you cannot check.

The record counts every day delivered, including the days sent by Telegram and by hand. When it is larger than the archive list, that is why: only the days published here have a page and a hash. See the record. The archive alone is priced on /returns, past tense, losses included.

Rules move forward only. Each published day records the version that judged it, so tightening something later can never reach back and take a day from somebody who already bought.

The date at the top of this page is the day the selection rule last moved. It is a constant, changed by hand when the rule changes — not today's date rendered fresh, which would make the page impossible to cite and would be a lie on every day but one.