This Saturday Rosas Jr. vs Barcelos · Meta APEX Las Vegas, NV United States See this week's picks →
FightIQ · Our record · Calibration
The number, graded

Is FightIQ over- or under-confident?

Across 1,919 graded picks the model predicted 62.1% and hit 66.9%: under-confident by 4.8pp.
The record page grades the direction of every pick. This page grades the number beside it: when we publish 70%, do seven in ten of those picks win? Each chart plots the mean published probability of a bucket against how often those picks actually won; the dotted diagonal is perfect calibration. Above the line the picks won more often than the number said (under-confident); below it, less often (over-confident).
1,919 graded picks · 14 January 2023 to 19 September 2026 · recomputed from the graded archive on 20 September 2026
All years combined2023202420252026

All years combined

2023–2026 · back-tested and live together

Every graded pick this site holds, the back-tested years and the live season together. 1,919 graded picks from 14 January 2023 to 19 September 2026, of which 475 were published as toss-ups and are counted like any other pick.

Reliability diagram, All years combined: predicted probability against actual hit rate, n=1919 30% 30% 40% 40% 50% 50% 60% 60% 70% 70% 80% 80% 90% 90% 100% 100% predicted probability (bucket mean) actual hit rate above the line: under-confident (picks won more often than the number said) below the line: over-confident no FightIQ pick is under 50% sportsbook implied · bucket under 50% · n=153 · predicted 43.7% · actual 47.1% (95% CI 39.3–54.9%) sportsbook implied · bucket 50–55% · n=589 · predicted 50.6% · actual 59.9% (95% CI 55.9–63.8%) sportsbook implied · bucket 55–60% · n=194 · predicted 57.6% · actual 58.2% (95% CI 51.2–65.0%) sportsbook implied · bucket 60–65% · n=232 · predicted 62.7% · actual 72.0% (95% CI 65.9–77.4%) sportsbook implied · bucket 65–70% · n=230 · predicted 67.5% · actual 67.0% (95% CI 60.6–72.7%) sportsbook implied · bucket 70–75% · n=183 · predicted 72.3% · actual 76.5% (95% CI 69.9–82.1%) sportsbook implied · bucket 75–80% · n=170 · predicted 77.3% · actual 82.4% (95% CI 75.9–87.3%) sportsbook implied · bucket 80–85% · n=92 · predicted 82.0% · actual 80.4% (95% CI 71.2–87.3%) sportsbook implied · bucket 85–90% · n=59 · predicted 87.3% · actual 93.2% (95% CI 83.8–97.3%) sportsbook implied · bucket 90–95% · n=16 · predicted 92.2% · actual 87.5% (95% CI 64.0–96.5%) sportsbook implied · bucket 95–100% · n=1 · predicted 95.4% · actual 100.0% (95% CI 20.7–100.0%) FightIQ · bucket 50–55% · n=473 · predicted 52.5% · actual 50.5% (95% CI 46.0–55.0%) FightIQ · bucket 55–60% · n=454 · predicted 57.4% · actual 63.4% (95% CI 58.9–67.7%) FightIQ · bucket 60–65% · n=355 · predicted 62.5% · actual 70.7% (95% CI 65.8–75.2%) FightIQ · bucket 65–70% · n=273 · predicted 67.3% · actual 75.8% (95% CI 70.4–80.5%) FightIQ · bucket 70–75% · n=181 · predicted 72.4% · actual 81.2% (95% CI 74.9–86.2%) FightIQ · bucket 75–80% · n=112 · predicted 77.2% · actual 79.5% (95% CI 71.1–85.9%) FightIQ · bucket 80–85% · n=62 · predicted 82.2% · actual 87.1% (95% CI 76.6–93.3%) FightIQ · bucket 85–90% · n=9 · predicted 86.3% · actual 88.9% (95% CI 56.5–98.0%) FightIQ published number sportsbook implied, same picks
FightIQ published number
Brier 0.2127
ECE 5.79pp
n 1,919 · predicted 62.1% · actual 66.9%
Sportsbook implied, same picks
Brier 0.2135
ECE 5.54pp
n 1,919 · predicted 61.7% · actual 66.9%
BucketnMean predictedActual95% CI (Wilson)Gap (pp)
50–55%47352.5%50.5%46.0–55.0%−2.0
55–60%45457.4%63.4%58.9–67.7%+6.0
60–65%35562.5%70.7%65.8–75.2%+8.2
65–70%27367.3%75.8%70.4–80.5%+8.5
70–75%18172.4%81.2%74.9–86.2%+8.8
75–80%11277.2%79.5%71.1–85.9%+2.3
80–85%6282.2%87.1%76.6–93.3%+4.9
85–90%986.3%88.9%56.5–98.0%+2.6
All buckets1,91962.1%66.9%64.7–68.9%+4.8
Sportsbook implied, same picks, binned on the no-vig sportsbook probability
Sportsbook bucketnMean impliedActual95% CI (Wilson)Gap (pp)
under 50%15343.7%47.1%39.3–54.9%+3.3
50–55%58950.6%59.9%55.9–63.8%+9.4
55–60%19457.6%58.2%51.2–65.0%+0.6
60–65%23262.7%72.0%65.9–77.4%+9.3
65–70%23067.5%67.0%60.6–72.7%−0.6
70–75%18372.3%76.5%69.9–82.1%+4.2
75–80%17077.3%82.4%75.9–87.3%+5.0
80–85%9282.0%80.4%71.2–87.3%−1.6
85–90%5987.3%93.2%83.8–97.3%+5.9
90–95%1692.2%87.5%64.0–96.5%−4.7
95–100%195.4%100.0%20.7–100.0%+4.6
All buckets1,91961.7%66.9%64.7–68.9%+5.2

The sportsbook series bins the same picks by the sportsbook’s own probability for the side we picked. Where that is under 50% the sportsbook had our pick as the underdog; those rows are the “under 50” bucket rather than being dropped.

Why no bucket sits under 50%. A pick is always the favoured side, so the lowest number FightIQ can publish on a pick is 50% and the lowest bucket is the coin-flip one; the sportsbook series can sit under 50% because it is the sportsbook’s view of the side we picked. One caveat on the combined view. The model behind these numbers changed across the years, so the combined line is the calibration of what we published each season, not of one fixed method.

2023

back-tested — scored after the fact

A back-tested year: the numbers were generated on the season's fights after the fact and scored against the results, which is a weaker claim than a live pick and is labelled as one. 504 graded picks from 14 January 2023 to 16 December 2023, of which 131 were published as toss-ups and are counted like any other pick.

Reliability diagram, 2023: predicted probability against actual hit rate, n=504 30% 30% 40% 40% 50% 50% 60% 60% 70% 70% 80% 80% 90% 90% 100% 100% predicted probability (bucket mean) actual hit rate above the line: under-confident (picks won more often than the number said) below the line: over-confident no FightIQ pick is under 50% sportsbook implied · bucket under 50% · n=53 · predicted 42.7% · actual 49.1% (95% CI 36.1–62.1%) sportsbook implied · bucket 50–55% · n=128 · predicted 50.9% · actual 56.2% (95% CI 47.6–64.5%) sportsbook implied · bucket 55–60% · n=63 · predicted 57.5% · actual 58.7% (95% CI 46.4–70.0%) sportsbook implied · bucket 60–65% · n=65 · predicted 62.8% · actual 78.5% (95% CI 67.0–86.7%) sportsbook implied · bucket 65–70% · n=67 · predicted 67.4% · actual 67.2% (95% CI 55.3–77.2%) sportsbook implied · bucket 70–75% · n=46 · predicted 72.4% · actual 73.9% (95% CI 59.7–84.4%) sportsbook implied · bucket 75–80% · n=41 · predicted 76.9% · actual 92.7% (95% CI 80.6–97.5%) sportsbook implied · bucket 80–85% · n=27 · predicted 81.6% · actual 77.8% (95% CI 59.2–89.4%) sportsbook implied · bucket 85–90% · n=14 · predicted 87.3% · actual 100.0% (95% CI 78.5–100.0%) FightIQ · bucket 50–55% · n=130 · predicted 52.6% · actual 51.5% (95% CI 43.0–60.0%) FightIQ · bucket 55–60% · n=120 · predicted 57.3% · actual 60.8% (95% CI 51.9–69.1%) FightIQ · bucket 60–65% · n=99 · predicted 62.5% · actual 75.8% (95% CI 66.5–83.1%) FightIQ · bucket 65–70% · n=70 · predicted 67.3% · actual 78.6% (95% CI 67.6–86.6%) FightIQ · bucket 70–75% · n=45 · predicted 72.2% · actual 73.3% (95% CI 59.0–84.0%) FightIQ · bucket 75–80% · n=27 · predicted 77.1% · actual 85.2% (95% CI 67.5–94.1%) FightIQ · bucket 80–85% · n=11 · predicted 81.9% · actual 90.9% (95% CI 62.3–98.4%) FightIQ · bucket 85–90% · n=2 · predicted 85.7% · actual 100.0% (95% CI 34.2–100.0%) FightIQ published number sportsbook implied, same picks
FightIQ published number
Brier 0.2130
ECE 6.05pp
n 504 · predicted 61.5% · actual 67.1%
Sportsbook implied, same picks
Brier 0.2091
ECE 6.21pp
n 504 · predicted 61.3% · actual 67.1%
BucketnMean predictedActual95% CI (Wilson)Gap (pp)
50–55%13052.6%51.5%43.0–60.0%−1.0
55–60%12057.3%60.8%51.9–69.1%+3.5
60–65%9962.5%75.8%66.5–83.1%+13.2
65–70%7067.3%78.6%67.6–86.6%+11.2
70–75%4572.2%73.3%59.0–84.0%+1.2
75–80%2777.1%85.2%67.5–94.1%+8.1
80–85%1181.9%90.9%62.3–98.4%+9.0
85–90%285.7%100.0%34.2–100.0%+14.3
All buckets50461.5%67.1%62.8–71.0%+5.5
Sportsbook implied, same picks, binned on the no-vig sportsbook probability
Sportsbook bucketnMean impliedActual95% CI (Wilson)Gap (pp)
under 50%5342.7%49.1%36.1–62.1%+6.3
50–55%12850.9%56.2%47.6–64.5%+5.4
55–60%6357.5%58.7%46.4–70.0%+1.2
60–65%6562.8%78.5%67.0–86.7%+15.7
65–70%6767.4%67.2%55.3–77.2%−0.2
70–75%4672.4%73.9%59.7–84.4%+1.5
75–80%4176.9%92.7%80.6–97.5%+15.8
80–85%2781.6%77.8%59.2–89.4%−3.8
85–90%1487.3%100.0%78.5–100.0%+12.7
All buckets50461.3%67.1%62.8–71.0%+5.7

The sportsbook series bins the same picks by the sportsbook’s own probability for the side we picked. Where that is under 50% the sportsbook had our pick as the underdog; those rows are the “under 50” bucket rather than being dropped.

2024

back-tested — scored after the fact

A back-tested year: the numbers were generated on the season's fights after the fact and scored against the results, which is a weaker claim than a live pick and is labelled as one. 513 graded picks from 13 January 2024 to 14 December 2024, of which 127 were published as toss-ups and are counted like any other pick.

Reliability diagram, 2024: predicted probability against actual hit rate, n=513 30% 30% 40% 40% 50% 50% 60% 60% 70% 70% 80% 80% 90% 90% 100% 100% predicted probability (bucket mean) actual hit rate above the line: under-confident (picks won more often than the number said) below the line: over-confident no FightIQ pick is under 50% sportsbook implied · bucket under 50% · n=38 · predicted 44.1% · actual 36.8% (95% CI 23.4–52.7%) sportsbook implied · bucket 50–55% · n=173 · predicted 50.4% · actual 63.6% (95% CI 56.2–70.4%) sportsbook implied · bucket 55–60% · n=50 · predicted 57.7% · actual 56.0% (95% CI 42.3–68.8%) sportsbook implied · bucket 60–65% · n=50 · predicted 62.5% · actual 70.0% (95% CI 56.2–80.9%) sportsbook implied · bucket 65–70% · n=69 · predicted 67.5% · actual 66.7% (95% CI 54.9–76.6%) sportsbook implied · bucket 70–75% · n=51 · predicted 72.1% · actual 76.5% (95% CI 63.2–86.0%) sportsbook implied · bucket 75–80% · n=41 · predicted 77.2% · actual 80.5% (95% CI 66.0–89.8%) sportsbook implied · bucket 80–85% · n=19 · predicted 81.5% · actual 78.9% (95% CI 56.7–91.5%) sportsbook implied · bucket 85–90% · n=20 · predicted 87.5% · actual 90.0% (95% CI 69.9–97.2%) sportsbook implied · bucket 90–95% · n=2 · predicted 91.2% · actual 100.0% (95% CI 34.2–100.0%) FightIQ · bucket 50–55% · n=127 · predicted 52.7% · actual 54.3% (95% CI 45.7–62.7%) FightIQ · bucket 55–60% · n=133 · predicted 57.3% · actual 62.4% (95% CI 53.9–70.2%) FightIQ · bucket 60–65% · n=93 · predicted 62.5% · actual 67.7% (95% CI 57.7–76.4%) FightIQ · bucket 65–70% · n=77 · predicted 67.3% · actual 72.7% (95% CI 61.9–81.4%) FightIQ · bucket 70–75% · n=44 · predicted 72.4% · actual 86.4% (95% CI 73.3–93.6%) FightIQ · bucket 75–80% · n=28 · predicted 77.1% · actual 75.0% (95% CI 56.6–87.3%) FightIQ · bucket 80–85% · n=10 · predicted 81.9% · actual 90.0% (95% CI 59.6–98.2%) FightIQ · bucket 85–90% · n=1 · predicted 86.4% · actual 100.0% (95% CI 20.7–100.0%) FightIQ published number sportsbook implied, same picks
FightIQ published number
Brier 0.2166
ECE 4.99pp
n 513 · predicted 61.5% · actual 66.3%
Sportsbook implied, same picks
Brier 0.2156
ECE 6.92pp
n 513 · predicted 61.2% · actual 66.3%
BucketnMean predictedActual95% CI (Wilson)Gap (pp)
50–55%12752.7%54.3%45.7–62.7%+1.6
55–60%13357.3%62.4%53.9–70.2%+5.1
60–65%9362.5%67.7%57.7–76.4%+5.3
65–70%7767.3%72.7%61.9–81.4%+5.4
70–75%4472.4%86.4%73.3–93.6%+14.0
75–80%2877.1%75.0%56.6–87.3%−2.1
80–85%1081.9%90.0%59.6–98.2%+8.1
85–90%186.4%100.0%20.7–100.0%+13.6
All buckets51361.5%66.3%62.1–70.2%+4.8
Sportsbook implied, same picks, binned on the no-vig sportsbook probability
Sportsbook bucketnMean impliedActual95% CI (Wilson)Gap (pp)
under 50%3844.1%36.8%23.4–52.7%−7.3
50–55%17350.4%63.6%56.2–70.4%+13.2
55–60%5057.7%56.0%42.3–68.8%−1.7
60–65%5062.5%70.0%56.2–80.9%+7.5
65–70%6967.5%66.7%54.9–76.6%−0.9
70–75%5172.1%76.5%63.2–86.0%+4.4
75–80%4177.2%80.5%66.0–89.8%+3.3
80–85%1981.5%78.9%56.7–91.5%−2.6
85–90%2087.5%90.0%69.9–97.2%+2.5
90–95%291.2%100.0%34.2–100.0%+8.8
All buckets51361.2%66.3%62.1–70.2%+5.1

The sportsbook series bins the same picks by the sportsbook’s own probability for the side we picked. Where that is under 50% the sportsbook had our pick as the underdog; those rows are the “under 50” bucket rather than being dropped.

2025

back-tested — scored after the fact

A back-tested year: the numbers were generated on the season's fights after the fact and scored against the results, which is a weaker claim than a live pick and is labelled as one. 515 graded picks from 11 January 2025 to 13 December 2025, of which 135 were published as toss-ups and are counted like any other pick.

Reliability diagram, 2025: predicted probability against actual hit rate, n=515 30% 30% 40% 40% 50% 50% 60% 60% 70% 70% 80% 80% 90% 90% 100% 100% predicted probability (bucket mean) actual hit rate above the line: under-confident (picks won more often than the number said) below the line: over-confident no FightIQ pick is under 50% sportsbook implied · bucket under 50% · n=30 · predicted 44.3% · actual 63.3% (95% CI 45.5–78.1%) sportsbook implied · bucket 50–55% · n=211 · predicted 50.3% · actual 58.8% (95% CI 52.0–65.2%) sportsbook implied · bucket 55–60% · n=47 · predicted 57.5% · actual 57.4% (95% CI 43.3–70.5%) sportsbook implied · bucket 60–65% · n=61 · predicted 62.5% · actual 75.4% (95% CI 63.3–84.5%) sportsbook implied · bucket 65–70% · n=54 · predicted 67.6% · actual 68.5% (95% CI 55.3–79.3%) sportsbook implied · bucket 70–75% · n=40 · predicted 72.3% · actual 75.0% (95% CI 59.8–85.8%) sportsbook implied · bucket 75–80% · n=44 · predicted 77.6% · actual 70.5% (95% CI 55.8–81.8%) sportsbook implied · bucket 80–85% · n=22 · predicted 82.6% · actual 77.3% (95% CI 56.6–89.9%) sportsbook implied · bucket 85–90% · n=5 · predicted 86.9% · actual 100.0% (95% CI 56.6–100.0%) sportsbook implied · bucket 90–95% · n=1 · predicted 90.8% · actual 100.0% (95% CI 20.7–100.0%) FightIQ · bucket 50–55% · n=135 · predicted 52.4% · actual 47.4% (95% CI 39.2–55.8%) FightIQ · bucket 55–60% · n=129 · predicted 57.5% · actual 66.7% (95% CI 58.2–74.2%) FightIQ · bucket 60–65% · n=105 · predicted 62.4% · actual 67.6% (95% CI 58.2–75.8%) FightIQ · bucket 65–70% · n=73 · predicted 67.3% · actual 72.6% (95% CI 61.4–81.5%) FightIQ · bucket 70–75% · n=38 · predicted 72.4% · actual 84.2% (95% CI 69.6–92.6%) FightIQ · bucket 75–80% · n=21 · predicted 77.2% · actual 90.5% (95% CI 71.1–97.3%) FightIQ · bucket 80–85% · n=11 · predicted 82.3% · actual 81.8% (95% CI 52.3–94.9%) FightIQ · bucket 85–90% · n=3 · predicted 85.8% · actual 100.0% (95% CI 43.9–100.0%) FightIQ published number sportsbook implied, same picks
FightIQ published number
Brier 0.2156
ECE 6.93pp
n 515 · predicted 61.2% · actual 65.4%
Sportsbook implied, same picks
Brier 0.2275
ECE 7.39pp
n 515 · predicted 59.7% · actual 65.4%
BucketnMean predictedActual95% CI (Wilson)Gap (pp)
50–55%13552.4%47.4%39.2–55.8%−5.0
55–60%12957.5%66.7%58.2–74.2%+9.2
60–65%10562.4%67.6%58.2–75.8%+5.2
65–70%7367.3%72.6%61.4–81.5%+5.3
70–75%3872.4%84.2%69.6–92.6%+11.8
75–80%2177.2%90.5%71.1–97.3%+13.2
80–85%1182.3%81.8%52.3–94.9%−0.5
85–90%385.8%100.0%43.9–100.0%+14.2
All buckets51561.2%65.4%61.2–69.4%+4.3
Sportsbook implied, same picks, binned on the no-vig sportsbook probability
Sportsbook bucketnMean impliedActual95% CI (Wilson)Gap (pp)
under 50%3044.3%63.3%45.5–78.1%+19.0
50–55%21150.3%58.8%52.0–65.2%+8.4
55–60%4757.5%57.4%43.3–70.5%0.0
60–65%6162.5%75.4%63.3–84.5%+12.9
65–70%5467.6%68.5%55.3–79.3%+0.9
70–75%4072.3%75.0%59.8–85.8%+2.7
75–80%4477.6%70.5%55.8–81.8%−7.1
80–85%2282.6%77.3%56.6–89.9%−5.3
85–90%586.9%100.0%56.6–100.0%+13.1
90–95%190.8%100.0%20.7–100.0%+9.2
All buckets51559.7%65.4%61.2–69.4%+5.7

The sportsbook series bins the same picks by the sportsbook’s own probability for the side we picked. Where that is under 50% the sportsbook had our pick as the underdog; those rows are the “under 50” bucket rather than being dropped.

2026

served live — published before the card

The live season: each number was published before the card and graded after it. 387 graded picks from 24 January 2026 to 19 September 2026, of which 82 were published as toss-ups and are counted like any other pick.

Reliability diagram, 2026: predicted probability against actual hit rate, n=387 30% 30% 40% 40% 50% 50% 60% 60% 70% 70% 80% 80% 90% 90% 100% 100% predicted probability (bucket mean) actual hit rate above the line: under-confident (picks won more often than the number said) below the line: over-confident no FightIQ pick is under 50% sportsbook implied · bucket under 50% · n=32 · predicted 44.4% · actual 40.6% (95% CI 25.5–57.7%) sportsbook implied · bucket 50–55% · n=77 · predicted 51.0% · actual 61.0% (95% CI 49.9–71.2%) sportsbook implied · bucket 55–60% · n=34 · predicted 58.0% · actual 61.8% (95% CI 45.0–76.1%) sportsbook implied · bucket 60–65% · n=56 · predicted 62.9% · actual 62.5% (95% CI 49.4–74.0%) sportsbook implied · bucket 65–70% · n=40 · predicted 67.6% · actual 65.0% (95% CI 49.5–77.9%) sportsbook implied · bucket 70–75% · n=46 · predicted 72.5% · actual 80.4% (95% CI 66.8–89.3%) sportsbook implied · bucket 75–80% · n=44 · predicted 77.5% · actual 86.4% (95% CI 73.3–93.6%) sportsbook implied · bucket 80–85% · n=24 · predicted 82.3% · actual 87.5% (95% CI 69.0–95.7%) sportsbook implied · bucket 85–90% · n=20 · predicted 87.3% · actual 90.0% (95% CI 69.9–97.2%) sportsbook implied · bucket 90–95% · n=13 · predicted 92.4% · actual 84.6% (95% CI 57.8–95.7%) sportsbook implied · bucket 95–100% · n=1 · predicted 95.4% · actual 100.0% (95% CI 20.7–100.0%) FightIQ · bucket 50–55% · n=81 · predicted 52.4% · actual 48.1% (95% CI 37.6–58.9%) FightIQ · bucket 55–60% · n=72 · predicted 57.5% · actual 63.9% (95% CI 52.4–74.0%) FightIQ · bucket 60–65% · n=58 · predicted 62.5% · actual 72.4% (95% CI 59.8–82.2%) FightIQ · bucket 65–70% · n=53 · predicted 67.4% · actual 81.1% (95% CI 68.6–89.4%) FightIQ · bucket 70–75% · n=54 · predicted 72.6% · actual 81.5% (95% CI 69.2–89.6%) FightIQ · bucket 75–80% · n=36 · predicted 77.3% · actual 72.2% (95% CI 56.0–84.2%) FightIQ · bucket 80–85% · n=30 · predicted 82.3% · actual 86.7% (95% CI 70.3–94.7%) FightIQ · bucket 85–90% · n=3 · predicted 87.1% · actual 66.7% (95% CI 20.8–93.9%) FightIQ published number sportsbook implied, same picks
FightIQ published number
Brier 0.2031
ECE 7.64pp
n 387 · predicted 64.7% · actual 69.3%
Sportsbook implied, same picks
Brier 0.1978
ECE 5.67pp
n 387 · predicted 65.4% · actual 69.3%
BucketnMean predictedActual95% CI (Wilson)Gap (pp)
50–55%8152.4%48.1%37.6–58.9%−4.3
one voice7852.4%50.0%39.2–60.8%−2.4
four-voice stack351.8%0.0%0.0–56.1%−51.8
55–60%7257.5%63.9%52.4–74.0%+6.4
one voice7057.5%64.3%52.6–74.5%+6.8
four-voice stack257.8%50.0%9.5–90.5%−7.8
60–65%5862.5%72.4%59.8–82.2%+9.9
one voice5462.5%72.2%59.1–82.4%+9.7
four-voice stack462.5%75.0%30.1–95.4%+12.5
65–70%5367.4%81.1%68.6–89.4%+13.7
one voice4867.4%85.4%72.8–92.8%+18.0
four-voice stack567.3%40.0%11.8–76.9%−27.3
70–75%5472.6%81.5%69.2–89.6%+8.9
one voice4872.6%85.4%72.8–92.8%+12.8
four-voice stack672.5%50.0%18.8–81.2%−22.5
75–80%3677.3%72.2%56.0–84.2%−5.0
one voice3377.3%72.7%55.8–84.9%−4.6
four-voice stack377.0%66.7%20.8–93.9%−10.4
80–85%3082.3%86.7%70.3–94.7%+4.3
one voice1882.2%83.3%60.8–94.2%+1.1
four-voice stack1282.5%91.7%64.6–98.5%+9.2
85–90%387.1%66.7%20.8–93.9%−20.4
one voice186.7%100.0%20.7–100.0%+13.3
four-voice stack287.3%50.0%9.5–90.5%−37.3
All buckets38764.7%69.3%64.5–73.6%+4.6

Two served models inside 2026

Served 24 January – 29 August 2026: one voice's number
Brier 0.2016
ECE 8.14pp
n 350 · predicted 63.8% · actual 70.0%
Served from 5 September 2026: the four-voice stack
Brier 0.2170
ECE 19.16pp
n 37 · predicted 72.7% · actual 62.2%

The served model changed on 5 September 2026. Up to and including the 29 August 2026 card every published probability came from a single voice; from the 5 September 2026 card it comes from all four together. The two are graded as two series, and each bucket in the table above splits the same way, because a single curve across both would describe a model this site no longer serves. The later series is 37 picks, so its intervals are wide and its bucket rows are indicative at best; it will firm up card by card.

Which of the two served a given pick is derived from its date: the graded archive carries no per-pick record of it, so the boundary used here is the day the serving path changed rather than a stamp on the row. All 387 of the 2026 picks are assigned that way.

Sportsbook implied, same picks, binned on the no-vig sportsbook probability
Sportsbook bucketnMean impliedActual95% CI (Wilson)Gap (pp)
under 50%3244.4%40.6%25.5–57.7%−3.8
50–55%7751.0%61.0%49.9–71.2%+10.1
55–60%3458.0%61.8%45.0–76.1%+3.8
60–65%5662.9%62.5%49.4–74.0%−0.4
65–70%4067.6%65.0%49.5–77.9%−2.6
70–75%4672.5%80.4%66.8–89.3%+8.0
75–80%4477.5%86.4%73.3–93.6%+8.9
80–85%2482.3%87.5%69.0–95.7%+5.2
85–90%2087.3%90.0%69.9–97.2%+2.7
90–95%1392.4%84.6%57.8–95.7%−7.8
95–100%195.4%100.0%20.7–100.0%+4.6
All buckets38765.4%69.3%64.5–73.6%+3.9

The sportsbook series bins the same picks by the sportsbook’s own probability for the side we picked. Where that is under 50% the sportsbook had our pick as the underdog; those rows are the “under 50” bucket rather than being dropped.

How to read this

The rows. Every graded pick on the record page’s loader, and only those: wins, losses and the toss-ups together, 1,919 in all. A toss-up is a graded pick with a published number and is counted like any other; leaving out the rows we were least sure of would flatter exactly the thing this page measures. The only rows that are absent are the ones the record page withholds because their source export is corrupt (none at this build).

The buckets. Ten equal-width probability buckets from 50% to 100%. Each dot is one bucket, sized by how many picks it holds, at the bucket’s mean published probability across and its actual hit rate up; the whisker is a Wilson 95% interval on that hit rate. A bucket of seven picks that all won is not “100%”, it is “somewhere above 65%”, and the whisker says so.

Brier is the mean squared error of the published probability against the 0/1 result: lower is better, and publishing 50% on everything would score 0.2500. ECE (expected calibration error) is the n-weighted mean of the bucket gaps: 0 means every bucket sat on the diagonal. Both are recomputed from the archive at every build; nothing here is typed in.

Back-tested against live. 2023 to 2025 were scored after the fact against a placeholder freeze, which is a weaker claim than a live pick and is labelled that way on each panel. 2026 is the live season, published before each card. The methodology page sets out how the number is made and where it is weak.