How accurate the forecast has been

Before play, the Desk's forecast was right on 62.8% of 417 doubles lines, no more than the higher Desk level, a free player rating.

417 doubles lines logged before play, October 8 to October 21, 2026, in the Combo and Mixed 55 & Over leagues. The last two rows count all 567 lines logged since October 3, 2026, 150 of them by an earlier version of the forecast.

Way of picking the winner	Lines	Picked the winner (95% range)
The Desk's forecast	417	62.8% (58.1 to 67.3)
The higher Desk level wins	417	62.8% (58.1 to 67.3)
The higher NTRP rating wins	417	53.2% (48.4 to 58.0)
The Desk's forecast, every line logged	567	61.2% (57.1 to 65.1)
The higher NTRP rating wins, every line logged	567	55.4% (51.3 to 59.4)

The Desk's forecast and the higher Desk level chose the same side on 415 of the 417 doubles lines. So the forecast doesn't yet pick more winners than a guess anyone can make from the free Desk levels. The forecast's win chances sat a little closer to the results than the higher Desk level's, by too little to be sure.

Logged before play, the forecast picked the winner less often than in a replay of past lines, which knew who played each line. Knowing who played, the higher Desk level picked the winner on 74.3% (69.9 to 78.3) of the same 417 doubles lines. So most of the gap is the guess of who will play.

The replay of past lines

The Desk's line forecast picked the winner on about two lines in three: 67.2% of 2,175 lines, replayed with the real lineups. A line is one singles or doubles match inside a team match.

Picking the side with the higher NTRP rating, or else the home side, got 56.1% of the 2,175 replayed lines right. Anyone can make that guess for free, because the rosters show each player's NTRP rating.

Most lines are even on paper. On about three lines in four of the replay, both sides carried the same NTRP rating, so the guess fell back to the home side. The home side won about one line in two: 51.7% of the 2,175 replayed lines. Where the NTRP ratings differ, on 504 replayed lines, the higher rating was right 71.8% of the time, about seven lines in ten, while the Desk's forecast got 74.8%.

Picking the side with the higher Desk level got 68.0% of 2,172 replayed lines right, while the Desk's forecast got 67.2% of them. So in the replay the forecast picked no more winners than the free Desk level.

Apart from the live record, the forecast's figures come from a replay of past lines. A forecast before a match also has to guess who will play, which the replay already knew. The forecast was tuned on earlier matches and replayed only on lines played after them.

The forecast's figures are all for single lines. The Desk also logs a chance to win the whole team match, but doesn't show or sell it.

If you want the Desk's line forecast before your next match, your first each week is free with an email sign-in, and the Desk Pass gives you more.

Method and full figures

How the replay was run, and every figure with its count and dates

The forecast gets each line's win chance from the Desk's line model. The replay scores that win chance on the Desk's 2026 lines, with the players who actually played.

Two guesses from the free pages are scored too, the higher NTRP rating and the home side, because a forecast is only worth paying for if it beats them.

All the main figures come from the same 2,175 lines, played July 6 to September 27, 2026.

The Desk tried a line forecast for 217 team matches in the leagues it covers, scheduled before October 24, 2026. The Desk's last try logged one for 217 of them.

The Brier score and log loss both measure how far each chance fell from what happened, so lower is better. Each percentage's 95% range comes from the Wilson method, a standard way to put a range on a share of right picks. Each score's range comes from drawing the team matches again 2,000 times, keeping each match's lines together.

A free guess's chance is the share of earlier logged lines it got right, as in the replay.

Doubles lines logged before play and scored before October 24, 2026. An even chance on every line scores 0.2500 on the Brier score and 0.6931 on log loss.

Way of picking the winner	Lines	Brier score (95% range)	Log loss (95% range)
The Desk's forecast	417	0.2253 (0.2089 to 0.2421)	0.6432 (0.6066 to 0.6814)
The higher Desk level wins	417	0.2365 (0.2249 to 0.2487)	0.6661 (0.6422 to 0.6911)
The higher NTRP rating wins	417	0.2533 (0.2467 to 0.2606)	0.7003 (0.6865 to 0.7152)
The Desk's forecast, every line logged	567	0.2363 (0.2212 to 0.2518)	0.6690 (0.6343 to 0.7038)
The higher NTRP rating wins, every line logged	567	0.2481 (0.2421 to 0.2545)	0.6895 (0.6771 to 0.7023)

The replay

• The Desk ran this replay on September 29, 2026, on the data the live site uses, without changing it. No earlier replay exists to compare it with.
• The replay ran in date order through all 55,087 league lines the Desk holds for the 2022 to 2026 seasons.
• Each line was predicted from what the Desk knew at the end of the day before it was played.
• The replay counts home court on every line, as a forecast for one line between named players does. A forecast for a whole team match leaves home court out, so its win chance for a close line can differ a little from the replay's.
• The replay left out the 493 lines of the 2026 season that had a recorded winner and a side with no players named.
• The Desk scored the higher Desk level on October 24, 2026, by running the same replay again on the scorecards it held before September 29, 2026. Scorecards changed since then leave that run 2,172 lines, and the Desk's forecast is scored again on them.

How often each one picked the winner

2,175 lines played July 6 to September 27, 2026, and the 504 of them where the two sides' NTRP ratings differ. The NTRP ratings were equal on 1,670 of the 2,175 lines, and a player had none on 1, so on those lines the rating guess picked the home side.

Way of picking the winner	Lines	Picked the winner (%)
The line model	2,175	67.17
The higher NTRP rating	2,175	56.14
The home side	2,175	51.72
The higher Desk level, in the later run	2,172	67.96
The line model, in the later run	2,172	67.17
The line model, where NTRP ratings differ	504	74.80
The higher NTRP rating, where NTRP ratings differ	504	71.83

How close each chance was

Log loss and the Brier score both measure how far each chance was from what happened, so lower is better for both.

Log loss weighs a confident miss most heavily. The Brier score is the average squared gap between each chance and the result.

The higher NTRP rating's chance is the share of earlier lines the higher-rated side won, and the home side's is the share the home side won.

2,175 lines played July 6 to September 27, 2026.

Way of picking the winner	Log loss	Brier score
The line model	0.6063	0.2089
The higher NTRP rating	0.6706	0.2391
The home side	0.6931	0.2500
An even chance on every line	0.6931	0.2500

How often the line model's chances came true

A well-judged chance comes true about as often as it says, so the two percentages in each row should sit close together.

2,175 lines, grouped by the line model's chance for the home side.

The line model's chance for the home side	Lines	Average chance (%)	Home side won (%)
0.0 to under 0.1	28	6.43	17.86
0.1 to under 0.2	76	15.83	15.79
0.2 to under 0.3	201	25.50	22.39
0.3 to under 0.4	260	35.34	34.62
0.4 to under 0.5	355	44.85	39.44
0.5 to under 0.6	365	55.00	49.32
0.6 to under 0.7	335	64.87	64.18
0.7 to under 0.8	296	74.92	75.34
0.8 to under 0.9	184	84.10	79.89
0.9 to 1.0	75	93.38	90.67

The whole season

Over all 11,206 lines of the 2026 season, the line model was right on 69.97% and the higher NTRP rating on 58.98%.

The whole-season figures include lines the line model's settings saw: 9,031 of the 2026 season's lines were played on or before the last day of its tuning data.

Way of picking the winner	Picked the winner (%)	Log loss	Brier score
The line model	69.97	0.5799	0.1978
The higher NTRP rating	58.98	0.6608	0.2344
The home side	53.02	0.6914	0.2491

Against the free guess

The line model did better than the free guess of the higher NTRP rating on both counts. On 2,175 replayed lines, it was right 67.17% of the time and the higher NTRP rating 56.14%. On log loss, where lower is better, it scored 0.6063 against 0.6706.

What the replay cannot rule out

• The Desk tuned the line model's settings on matches up to July 5, 2026, so the replay's main figures use only lines played after that date, and the figures for the whole 2026 season include lines the settings saw.
• The replay uses the NTRP rating the roster shows now for the line's season, never a later one, so a rating that changed mid-season counts from the season's first line. The rating guess and the line model both read that rating, so they share the error.
• The replay scores the chance for the players who actually played each line. A forecast before a match also has to guess who will play, and that adds error this replay does not measure.
• Lines where one side named no players were not played, and are left out for all three predictors alike.

The chance to win a whole team match

The Desk logs a team's chance to win a whole team match before lineups, and scores it after the result, but does not show or sell it.

A team match logged for both of its teams counts twice in the figures for the whole-team-match chance, once for each team's chance.

The whole-team-match chance had a Brier score of 0.2489 and a log loss of 0.6911, over 705 chances logged for team matches played July 18 to September 27, 2026. An even chance scores 0.2500 on the Brier score and 0.6931 on log loss.

705 whole-team-match chances logged for team matches played July 18 to September 27, 2026, counted from the scorecards the Desk held on September 29, 2026.

Way of picking the team match winner	Chances	Picked the winner (%)
The whole-team-match chance	705	52.62
The home side	705	52.06

How the replay was checked

As a check, the replay rebuilt each player's level before each line. Those levels matched the ones the Desk had stored on 40,074 of 40,074 player spots in the 2026 lines, one spot for each player on each line.