Modern player-seasons with unavailable historical inputs hidden
Audit updated Jul 22, 2026
Career Ranking Audit · 76 seasons
Keep the career model. Treat historical defense as experimental. Show the uncertainty.
The core 10-to-1 Career Dominance rule survives reasonable alternatives. Neither historical-defense candidate cleared its validation gates. A small, position-neutral roster-event adjustment is nevertheless included experimentally by product decision and labeled wherever it appears.
Keep Career Dominance definitionLabel historical defense as experimentalPublish 80% rank ranges
Without the proposed defense residual
Only 0.0004 better—below the gate
point-anchored 80% simulation range (point estimate plus 10th–90th percentile)
Finding 01
The historical defense candidate failed.
The original backtest accidentally used modern Defensive Win Shares, which already contains steals, blocks, and defensive rebounds—the very inputs missing before 1973–74. Recomputing the exact historical DWS allocation removed that advantage.
The pre-registered gate required at least 0.010.
Improvement was too small to clear the subgroup gate.
The separate V7 roster-event adjustment is included only as an explicitly disclosed product experiment.
Finding 02
The broad career order is robust; exact places are not.
Every alternative retains at least 20 of the current top 25 and the median alternative has a median top-25 shift no greater than five places. The audit passes that population-level rule, while still showing meaningful movement for some individuals.
Finding 03
Older careers need a range, not false precision.
Simulation varies only reconstructed seasons and keeps exact seasons fixed. This table preserves the pre-experiment audit baseline; current ranks may move slightly under the separately labeled V7 adjustment.
| Rank | Player | 80% range | Points | #1 seasons | Top-10 seasons | Top-25 chance | Rule sensitivity |
|---|---|---|---|---|---|---|---|
| 1 | LeBron Jamesexact | #1–#1 | 143 | 7 | 20 | 100% | #1–#3 |
| 2 | Kareem Abdul-Jabbarbridge | #2–#3 | 127 | 9 | 14 | 100% | #1–#3 |
| 3 | Wilt Chamberlainlegacy | #2–#11 | 121 | 5 | 14 | 100% | #3–#13 |
| 4 | Michael Jordanexact | #3–#4 | 117 | 9 | 12 | 100% | #2–#7 |
| 5 | Jerry Westlegacy | #3–#9 | 117 | 3 | 14 | 100% | #4–#12 |
| 6 | Magic Johnsonexact | #5–#6 | 93 | 3 | 12 | 100% | #6–#10 |
| 8 | Nikola Jokićexact | #8–#9 | 81 | 5 | 10 | 100% | #4–#14 |
| 10 | Kevin Durantexact | #10–#12 | 79 | 1 | 13 | 100% | #5–#16 |
| 11 | Oscar Robertsonbridge | #8–#16 | 79 | 1 | 10 | 100% | #8–#24 |
| 12 | Giannis Antetokounmpoexact | #11–#13 | 75 | 1 | 10 | 100% | #8–#22 |
| 14 | Elgin Baylorlegacy | #8–#36 | 73 | 2 | 10 | 72% | #14–#27 |
| 15 | Bob Pettitlegacy | #8–#58 | 72 | 1 | 11 | 61% | #12–#29 |
| 16 | Shaquille O'Nealexact | #14–#16 | 71 | 2 | 10 | 100% | #14–#17 |
| 17 | Stephen Curryexact | #15–#17 | 69 | 1 | 12 | 100% | #8–#21 |
| 19 | Larry Birdexact | #17–#19 | 66 | 2 | 8 | 100% | #10–#25 |
| 22 | Julius Ervingexact | #22–#25 | 51 | 1 | 7 | 98% | #22–#34 |
| 25 | Bill Russelllegacy | #11–#84 | 49 | 0 | 12 | 40% | #10–#58 |
| 27 | Dolph Schayeslegacy | #17–#113 | 48 | 0 | 8 | 34% | #27–#45 |
| 28 | Bob Cousylegacy | #20–#86 | 47 | 1 | 7 | 28% | #26–#49 |
| 29 | Karl Maloneexact | #26–#29 | 47 | 0 | 10 | 8% | #15–#29 |
| 31 | Kobe Bryantexact | #27–#31 | 44 | 0 | 8 | 0% | #21–#41 |
| 41 | Tim Duncanexact | #38–#42 | 33 | 0 | 7 | 0% | #26–#62 |
| 43 | Dirk Nowitzkiexact | #39–#43 | 32 | 0 | 6 | 0% | #25–#51 |
| 45 | Hakeem Olajuwonexact | #42–#46 | 30 | 0 | 8 | 0% | #28–#61 |
| 65 | George Mikanlegacy | #44–#216 | 20 | 0 | 4 | 0% | #65–#97 |
| 91 | John Havlicekbridge | #67–#137 | 11 | 0 | 3 | 0% | #80–#115 |
Finding 04
A top-ten finish did not always mean the same thing.
A fixed top-ten cutoff is less selective in smaller historical leagues; the field-adjusted sensitivity model tests the size of that effect rather than silently changing the live definition. This is a definition choice, not an automatic correction: smaller leagues can also contain concentrated elite talent.
1950s85 playerstop 10 = 11.8%
1960s103 playerstop 10 = 9.7%
1970s200 playerstop 10 = 5.0%
1980s258 playerstop 10 = 3.9%
1990s304 playerstop 10 = 3.3%
2000s331 playerstop 10 = 3.0%
2010s358 playerstop 10 = 2.8%
2020s381 playerstop 10 = 2.6%
Definitions and limits
What was tested.
The named legends were case studies, never fitting targets. Awards and championships were not used to tune the order.
Career Dominance
Ten points for a #1 season through one for #10, summed across all qualified seasons. Ties use #1 seasons, standardized margin, top-five seasons, then Career Value.
Masking test
On 1978–1990 data, steals, blocks, and defensive rebounds were hidden. Pre-1974 DWS was rebuilt from minutes, total rebounds, and assists only.
Simulation range
Persistent player error and season-specific error are drawn for modeled seasons. Exact seasons do not move. The display spans the point estimate and the simulation's 10th–90th percentiles.
What remains missing
Structural uncertainty, pre-shot-clock context, direct early defensive attribution, ABA production, playoffs, awards, and championships remain outside the ranking.
Data quality: 25,093 player-seasons across 76 seasons; 18,677 clear the 450-minute floor. Duplicate keys: 0. Missing core fields: 0. Sources: Basketball-Reference Win Shares methodology ↗ and BPM 2.0 methodology ↗.