Watch a weighted draw happen

The simulator below runs the eight pacers of the Kurrajong Mile at the prices that card published, over 1609m, and takes about a minute a race. The mechanism is the point: the finishing order is settled before the first frame renders, and the animation is fitted to it afterwards.

The bundle runs a working copy of the mechanism. Each pacer holds a share of the draw in proportion to the reciprocal of its price. One share is drawn, the runner it belongs to takes first placing and leaves the pool, and what is left is drawn again for second and the rest.

Night Shift at 3.22 holds 27.1% of that draw. Salt Pan at 19.83 holds 4.4%. The favourite wins most often because it holds the largest share, not because it is the better horse. There are no better horses: there is a block of numbers and a draw.

Two consequences follow, the same two every card here carries. A runner cannot be drawn twice, so there is no dead heat and no unfilled placing. And the order is complete before anything moves: each sprite is handed the animation of a placing it already has.

Nothing else is an input — not the credits staked, not the counter beside them, not the bundle's own float. A race with nothing on it behaves like a race with ten credits on the favourite. How a virtual race is decided has the arithmetic, worked on a published weighting table.

1. Past the menu is the bet panel: eight runners, win, place and show prices, and a forecast grid a forecast grid for first and second. Every figure is in credits.
2. Stake a credit on a runner, or stake nothing. The race runs either way, which is the quickest way to see that the draw does not care.
3. Press Run the race. The pacers already carry their placings, so what follows is a render: the lengths and the finish flash fit an order that exists already.
4. The arrivals panel fills as each pacer crosses; the end panel reads out the placings and the credits.

The number beside a runner on screen is its saddlecloth, 1 to 8, in the order the bundle stacks its sprites. The card numbers the same eight by barrier draw, so the two disagree; the table below prints both.

Run the demo, write down the first three, then open the Kurrajong Mile and read the weighting table under the field. The card prints what no commercial virtual product prints: the numbers each runner held out of 1,000, the implied percentage that follows, and the book at 114.6%, which is 14.6% over a fair hundred.

The two sets of figures line up. Night Shift holds 271 of the 1,000 numbers on the card, which is 27.1%; its share of the demo draw is 27.1%, because the book shades every price in the same proportion. Against Salt Pan on 44 numbers, that is 6.2 wins to one. Twenty runs will lean that way. Four will not.

The comparison will not give the same result. The demo publishes no seed; the card is settled once, with its seed printed. Same shape of draw, not the same draw.

As it arrived, the bundle was a compensated game. It decided whether the player should collect — a roll against a win rate of 40 in 100, with an automatic loss whenever the payout would have exceeded its own float — then re-drew finishing orders until one matched. The order depended on the stake, which Britain's RTS 7A does not permit in a licensed product. One weighted draw replaced it. The changed file is play/game/js/CGame.js; the original sits beside it as CGame.js.orig.