The Daily Lob

Where the tournament stands

The value board, after its worst day

Yesterday this page put the backtest at the top and said the number would run there every day, whichever way it pointed. It pointed down.

Eight cards, eight bets, two winners and minus 4.62 units — the worst single day since the brief started keeping this score. Worse than the count: the two biggest edges on the page, the ones the new ordering promoted to the top, both lost. The two that won were a mid-sized flip and a 2.9-point edge at odds of 1.08, which returned eight pence in the pound.

Across the whole archive the signal now stands at 91 bets, -7.01 units, an ROI of -7.7%. That is down from −2.9% yesterday. One day moved it nearly five points, which is the clearest possible illustration of why the sample size has to be quoted alongside the number every time.

The shape is worse than it was, and in an instructive way. Yesterday the table was positive at 2 points, negative at 5 and 8, then strongly positive at 10 and 15. Today it is negative at every threshold except the top two, and the 10-point band has collapsed from +33.0% to +6.4% on the addition of three bets. A band that swings twenty-seven points on three results was never evidence of anything. The honest reading remains: no edge has been demonstrated, and the run is now meaningfully behind the market rather than level with it.

The stale ratings file is the leading suspect and it got worse, not better. Both Elo files are still stamped 31 August — the sixth consecutive day, and they have still seen no match of this tournament. Every disagreement this page prints is computed from a world in which the US Open has not begun.

Yesterday's brief, scored

Eight cards, eight results, the favourite won seven — and the model had its worst day too, closer on just two.

Both flips failed and one flip won. The page led on Bergs at +184 because Elo made him level with van de Zandschulp; van de Zandschulp won in five, and the card had said the edge was probably just staleness — the ratings had not seen him beat Medvedev, Hurkacz and de Minaur. That was the right diagnosis and it should have been a reason not to card it at the top. Elo flipped Jović v Eala to Eala by fourteen points; Jović won 7-5 3-6 7-5, and the head-to-head that said she had beaten Eala twice this year was the better guide. The flip that did come off was Arthur Géa, who beat Michael Zheng in five and is a lucky loser in the last sixteen of a Grand Slam.

The Rybakina card was the one that read correctly. The brief said a three-point edge at 1.08 was not a bet but an illustration of what the model is blind to — the heel tendon, which Andy Roddick had described on air. She won 7-6(6) 6-3, taking the first set on a tiebreak. The ankle held, the market's three-point discount looked about right, and nobody should have staked anything on it.

Elise Mertens lost, and the deep-run flag scores its first clean hit. Flagged at 28.2% coming off the Monterrey final seven days earlier, she took a set off Naomi Osaka and lost 6-3 3-6 7-5. That takes the experiment to six settled cases — and it is still running against its own thesis, because the other five were priced such that three wins is roughly what the market expected.

Sixteen days. Selected matches played to a finish have gone to the favourite 76 times in 126.

Elo v Vegas

Elo was closer on 2 of eight — its worst day of the run. Across fifteen days it stands at 41 of 92.

Two of eight, and the two it won were the two where it was the more timid line — Géa and Rybakina. On the six it lost it was the bolder line every time. That is the same pattern as 2 September, reversed from 4 September, and across ninety-two cards it now sits at 45%. A model that is closer than the market on fewer than half its cards, and behind on money, is not yet earning its place on the page. It stays because the point of publishing the number is to find that out, not to prove it.

The deep-run flag

Be wary of a player arriving off a semifinal or final at their previous tournament, within about ten days. Run as a tracked experiment, not an opinion. Six settled cases; they have won three, and their own prices implied 2.8. The flag is still running slightly AGAINST its own thesis — flagged players have done marginally better than the market expected, not worse. Six is far too few to conclude anything.

Nothing is flagged today, and that is a decision rather than an absence. Two players on this page contested the Cincinnati final on 23 August — Jessica Pegula, who lost it, and Frances Tiafoe, who lost it having had hand surgery nine days earlier. Both would be interesting flags. Both are fourteen days past that final, and the rule says about ten. The window is not being widened to catch them.

This is the second time the question has come up. On 3 September the flag missed Ignacio Buse's collapse by one round and the brief declined to extend the rule to capture it, on the grounds that widening a window after seeing a result that would benefit from a wider window is how an experiment stops being one. The same applies to widening it after seeing two cases you would like to include. If the ten-day window is wrong, it should be changed on argument before the fact, not on convenience after it.