IRRATIONAL CONFIDENCE: a single number (0–100) for a player's willingness to resolve the possession himself — iso volume, clutch takes, and the fadeaway. Who's putting it up — whether or not they should.
Read the formula →“The Bureau records the decision. It passes that decision through the math to determine how irrational or rational, and how confident or unconfident the decision was.”
The Bureau presents the full register of seasons in which a player has demonstrably preferred his own counsel, in rejection of any and all available evidence. Waving off the coach, his teammates, perhaps even his loved ones, he seeks individual glory. His track record is a factor, weighing how “rational” a shot this is, while the tendency to take shots is the “confidence.” Put together: this is Irrational Confidence. Each entry is a single year of recorded confidence-over-evidence.
What follows is the top of the register — ten decisions featured in full. The remainder, two hundred deep, runs in the abbreviated continuation list, and the methodology by which each is scored is set out in the notes that close the file. The Bureau reminds the reader that irrational_confidence measures the decision, not its outcome — willingness is the watchword.
Each season placed on two axes: how readily the player took the shot upon himself (vertical — confidence), against how well, by his own total record, he merited it (horizontal — merit, per-shot value over the league baseline). Both measured against the player's own era. The dashed diagonal is perfect calibration; the oxblood line is the league's actual average confidence/merit line.
Each dot is a player-season, sample size (and therefore confidence) shown by dot size — the well-documented loom large, the noisier shrink to a speck. Showing the 284 most-documented seasons, 1997–2025. Narrow to a single season below to see its full field.
Nos. 13 through 42, in brief; the full register of two hundred is available in the appendix.
Irrational Confidence is a compound measure — a single number assembled from the two axes of the field above. The first, confidence, asks did he shoot it himself? — how readily the player took the possession onto his own shoulders. The second, merit, asks should he have? — whether his established shot-making, on the exact diet he took, beat the league's baselines. Both are measured against the player's own season, so eras compare directly.
The trait the Bureau is after is the gap between the two: confidence that outruns merit. A season scores high only when the player both asserted himself and did so past what the record warranted — the upper-left corner of the field. Warranted volume (high confidence, positive merit — the league's efficient stars) and deference (low confidence) both fall away toward the floor. The result is bounded 0–100; the median season sits near 4, because most players are not, in fact, irrationally confident.
confidence_z = 0.50 × trigger_z
+ 0.30 × self_creation_z
+ 0.20 × difficulty_zmerit = Σ_zone diet_share × trust × (skill_pps − baseline_pps)
where trust = eff_shots / (eff_shots + 40) # smooth, not a gate
merit_z = z-score of merit vs the season's qualified fieldThe two axes combine into a single assertion-over-merit term, tilted toward confidence so a high-volume gunner outranks a modest one at the same gap, then smoothly restricted to the irrational corner by two logistic factors — one that fades the score as confidence drops below the field, one that fades it as merit climbs above zero. No hard gates: a warranted or a deferential season slides continuously to the floor rather than falling off a cliff. The corner term is mapped to 0–100 by a final logistic.
# σ(x) = 1 / (1 + e^−x) · conf_z, merit_z are the two axes above
raw = (conf_z − 0.6 × merit_z) # assertion, tilted toward confidence
× σ(conf_z / 0.4) # fade out as confidence drops
× σ(−merit_z / 0.4) # fade out as the shot becomes warranted
irrational_confidence = 100 × σ( 2.27 × (raw − 1.40) ) # → 0–100Anchors — Iverson ’05 ≈ 99 · Westbrook ’11 ≈ 99 · Kobe ’16 ≈ 96 · Harden ’19 ≈ 38 (warranted) · Kobe ’13 ≈ 9.5 · DeAndre ’15 ≈ 4 (deferential).
Sign up to receive information on experiments, features, releases, and other news from the Institute. No spam. Cancel any time.