NBA ACADEMY GAMES · HISTORICAL DATABASE

Methodology

Edition separation

Every player, team and game record carries an edition year. The 2017 edition is shown in coral, the 2019 edition in amber, the 2022 edition in periwinkle, the 2023 edition in ivory, the 2024 edition in sky blue, the 2025 edition in hot pink and the 2026 edition in mint. Filters and CSV exports preserve that year field, and identically named teams from different editions remain separate records.

2017 source coverage and event structure

The 2017 edition is integrated from the final corrected Step 3 package and its source-level workbook. It contains 41 roster identities: 40 active statistical participants and Guo Kuo as a source-only 0-GP row. All 15 official games are preserved, including the mixed field of Academy and external teams. The four Academy teams each played five games, so Academy-team GP totals cannot be divided by two to infer the full event game count.

Six documented guest appearances are attributed to the team represented in each box score while each player retains his primary Academy affiliation. DNP rows remain source evidence but are not counted as appearances and are not included in ABI, advanced-rate, comparison or outcome-rate denominators.

2017 corrected downstream models

The displayed 2017 ABI+ and advanced statistics are the verified corrected outputs: team-edition aggregate minutes are normalized to 200 player-minutes per team-game, guest possession exposure follows the represented team, and primary-team opportunity denominators carry the documented guest mismatch caveat. Age Context was rebuilt across all 404 eligible rows with leave-one-participant-out identity handling. The separate Kish effective-sample-size diagnostic is retained only as a diagnostic; legacy Age Context effectiveN is unresolved and remains null for all 2017 rows. Comparison Model 2.1 is recalibrated to 509 eligible targets and 5,090 displayed top-10 pairs.

2019 source coverage

The first-edition files provide 85 player records and eight team records. Player totals and per-game values are usable, but every player is listed with zero minutes and the supplied player per-40 and per-100 exports consequently contain zero rate counts. Those rate fields are displayed as unavailable rather than as genuine zero production. The 2019 team per-game, per-40 and per-100 exports use incomplete denominators and are not presented as comparable rate data. The Teams view instead calculates clearly labeled averages from the 17 available box scores.

The database includes 17 of the 24 2019 game box scores—the complete set that was available. Seven games remain represented only through the aggregate tables. Every available game includes quarter scoring, official team totals and corrected player logs. The PDFs contain several players duplicated onto both teams; those cross-listings are assigned only to the confirmed team, while the official score and source team-total row remain unchanged and the resulting discrepancy is labeled inside the box score.

The first-edition player rows do not always reconcile exactly with the listed team totals, even apart from cross-listings, so each source layer is preserved rather than silently forced to match. The listed team records sum to 18 wins and 30 losses despite every team being shown with six games, another sign that the first-edition export is incomplete.

2022 source coverage

The 2022 edition contains 93 player records, eight active team records and all 24 tournament box scores. Totals, per-game statistics and team per-40/per-100-possession tables are preserved from the supplied aggregate exports. Every player and game PDF lists player minutes as 0:00, so the published player per-40, per-100 and other minute-dependent opportunity rates remain unavailable rather than being presented as genuine zeroes. ABI is still available through clearly labeled model-estimated minutes.

One Day 1 Africa Blue box score includes six rows that do not belong to its official aggregate roster. Those rows are labeled as source cross-listings and excluded from player game logs; the official team-total row and final score remain unchanged. Several source game-duration fields are visibly corrupted and are retained only as source metadata, not used in calculations.

2023 source coverage

The 2023 edition contains the complete supplied player and team aggregate tables—totals, per game, per 40 minutes and per 100 possessions—plus all 24 tournament box scores. The edition is color-coded ivory. Source-listed anomalies, including isolated 0:00 minute entries, implausible player-minute values, a 3:25 listed game duration and partial score-by-period data, are preserved as source metadata rather than silently corrected. One team lacks a complete quarter breakdown in each game; one TSF row contains only two period values, and one Africa Red row does not sum to the published final. Missing period values display as em dashes rather than being reconstructed. All 54 active 2023 player rows now have a listed height: ten retain event-roster values, while the 44 previously blank rows use clearly labeled current RealGM listings. Conflicting RealGM heights are preserved as alternate values rather than silently reconciled. Darian Bailey’s two active aggregate rows remain preserved in the source layer, while derived ABI standardization represents the linked identity once.

2024 source coverage

The 2024 edition contains 82 active nonzero player records, eight active team records and all 24 tournament box scores, with the supplied totals, per-game, per-40 and per-100 tables preserved as separate source views. Exact duplicate team rows and zero-game legacy rows are excluded from active counts but documented as source anomalies. Page-break duplicate player rows are counted once.

All source-listed box-score values remain unchanged, including non-reconciling team minute sums, an unnamed “Player Name” appearance, isolated 0:00 rows with recorded statistics, a 40:45 player-minute entry and a Howard Pulley technical foul for which the source does not identify an individual player.

2025 source coverage

The 2025 edition contains 86 active player records, eight active team records and all 24 tournament box scores. The supplied totals, per-game, per-40 and per-100 tables are retained as distinct source views, with OREB and DREB displayed separately throughout. The identical duplicate NBA Academy Africa team row is counted once, the zero-game NBA Academy Select Blue team row and three zero-game player rows are excluded from active displays, and all nonzero partial-event player rows remain active.

The source-listed 0:00 Zion Green appearances and two unnamed TSF technical-foul listings are preserved rather than silently corrected. The supplied 2025 player per-100 PDF duplicates the per-40 values; the database preserves that published view exactly and documents the duplication.

The aggregate player minute totals are systematically higher than the sums of the listed box-score minute fields, while four team aggregate rows differ from the box-score sums by one to three turnovers or one personal foul. Aggregate and game-level source layers therefore remain separate and are not silently forced to reconcile.

2026 source coverage

The 2026 section retains the final player and team totals, per-game, per-40 and per-100 values, all 24 complete game PDFs, expandable box scores and player game logs. The NBA-published Africa and Select Blue team aggregates remain preserved as source records, while all 2026 derived team context and dependent analysis use reconciled game-level sums from the complete 24-game box-score set with the documented possession fallback applied consistently across the edition.

Repeat-participant links

Repeat participants are connected through a stable participant ID while every edition-specific statistical row remains separate. Links use exact full-name matches plus a small manually reviewed alias registry for documented spelling, middle-name and abbreviation variants. Ambiguous initial-only matches are deliberately left unlinked. The linked profile displays one primary appearance per edition and team, including per-game production, steals, blocks and shooting splits. Its career-average row is weighted by games and combines total shooting makes and attempts across linked appearances. A Development Snapshot compares the earliest and latest recorded appearances across role, production, efficiency, ABI+ and age-adjusted ABI; it describes statistical change and does not infer the cause of that change.

Player comparisons

The player drawer lists the ten closest distinct historical NBA Academy Games profiles below the Rate Snapshot. Comparison Model 2.1 separates the profile into three fixed components: Style (50%), Role (35%) and Impact (15%). Style covers shot distribution, efficiency, creation tendency, ball security, rebounding distribution and defensive events. Role uses each player’s share of his team’s points, field-goal attempts, free-throw attempts, assists, offensive rebounds, defensive rebounds and offensive involvement. Impact uses ABI+ for rate impact and ABI Tournament Impact for unit-free tournament volume.

Every input is converted to an edition-adjusted percentile before cross-edition comparison. Related inputs sit inside fixed weighted buckets so overlapping shooting, creation and rebounding statistics cannot accumulate unlimited influence. Two- to five-game samples are shrunk toward the edition average, and ABI inputs are additionally adjusted by their published reliability and minute-source quality. Candidates and selected players must have at least two recorded games, repeat appearances by the selected player are excluded, and only the closest edition is retained for any comparison player.

The displayed Overall, Style, Role and Impact scores are empirical match-quality percentiles calibrated against the 5,090 top-10 nearest-neighbor comparison pairings generated across all eligible player records. A 50% score is typical among displayed historical comparisons, while a 90% score is closer than approximately 90% of other top-10 comps. This local calibration avoids the misleading 98–99% clustering that occurs when every nearest-neighbor match is compared with the much larger pool of mostly unrelated player pairings. Same-player appearances are excluded from the displayed comparison pool. These comparisons describe tournament statistical profiles; they are not prospect rankings or career projections.

Data-audit rules

The Data Audit view is diagnostic only. It calculates expected game coverage from active team GP totals, compares supplied team aggregates with official game-total rows, checks player aggregate GP against matched box-score appearances and surfaces source-minute, period-scoring, player-row, roster and duplicate-row issues. It does not rewrite stored source values.

Advanced-stat calculations

For 2026, GmSc, USG%, AST%, TOV%, TREB%, DREB%, OREB%, STL%, BLK%, TS%, A/TO, eFG%, 3PAr and FTr use the reconciled game-level team context described above; the NBA-published aggregate source rows remain preserved separately for source fidelity. For 2017, 2019, 2022, 2023, 2024 and 2025, game-level GmSc is available for every supplied box score, while aggregate advanced metrics remain limited to those calculable without player minutes or complete on-court opportunities: GmSc, TOV%, TS%, A/TO, eFG%, 3PAr and FTr. The remaining opportunity-based percentages are left blank.

Academy Box Impact (ABI)

ABI is the frozen FLOOR and CEILING box-score model built specifically for the NBA Academy Games database. It estimates a player’s rate impact per 100 on-court possessions relative to the player-model-minute-weighted average in that edition. It is not true plus-minus or an official external metric, and it summarizes short-tournament performance rather than stable long-term player quality.

The model normalizes every team to 200 player-minutes per game. Published minute shares are used when available; 2017 uses the corrected normalized team-edition minute architecture, while 2019 and 2022 use estimated model minutes based on games played and aggregate role activity. Published team possessions are used when plausible. Otherwise, possessions are estimated as FGA − OREB + TOV + 0.44 × FTA, with the edition environment used as a fallback.

ABI combines five edition-standardized components. Scoring and Efficiency carries 40% of the final model and is built from 60% Points/100 plus 40% Shooting Value/100. Shooting Value compares the player’s actual points with the points an edition-average shooter would be expected to produce from the same number of field-goal attempts and weighted free-throw attempts. In formula form: 100 × [PTS − edition points per true-shot attempt × (FGA + 0.44 × FTA)] ÷ estimated on-court possessions. A positive value means the player produced more points than the edition baseline from the shooting opportunities he used; a negative value means he produced fewer. Points/100 captures scoring volume, while Shooting Value/100 captures the efficiency of that volume.

Creation and Ball Security carries 20% and is 65% Assists/100 minus 35% Turnovers/100. Rebounding carries 15% and is 55% OREB/100 plus 45% DREB/100. Defense and Foul Discipline carries 20% and is 50% Steals/100, 35% Blocks/100 and 15% foul avoidance, creating final model weights of 10% steals, 7% blocks and 3% foul avoidance. Team Context carries 5% and uses the same edition-standardized team scoring margin per game for every player on that team. Short samples and estimated-minute rows retain the existing reliability shrinkage before the results are recentered and scaled.

The formula was built from box-score logic. Known NBA outcomes were used only for retrospective, in-sample stress tests; they were not score inputs and were not used to fit the frozen weights.

ABI is the rate estimate. ABI+ maps each edition to the same within-edition scale: 100 is the player-model-minute-weighted average, 115 is one player-model-minute-weighted standard deviation above it and 130 is two. ABI Tournament Impact multiplies ABI by estimated on-court tournament possessions and divides by 100; it is a unit-free modeled total, not points or observed plus-minus.

Edition ABI+ Rank is calculated against every rated performance in that player’s own edition. Its denominator therefore reflects the complete rated edition field—for example, “1 of 89”—and it does not change when the ABI table is sorted, searched or filtered. The qualification controls determine which performances are visible in the working table; they do not redefine the historical rank.

Physical-profile pilot

The database includes roster position and height records for all 86 active players in the 2025 edition, preserving the user-supplied listings exactly. The three source-only zero-game roster rows—Sebastian Haubro, Flynn Pavely and Casey Williams—are retained in physical-profile metadata but remain excluded from active player displays under the existing 2025 coverage rules. The database also includes official roster-height coverage for 2019, 2022, 2023 and 2024, official 2017 all-tournament position/height listings where published, plus the full 2026 physical-profile pilot. Weight and wingspan appear only where a separate credible source publishes them.

The matched-measurement system now contains 81 active player rows with wingspans and 70 with weights. It combines the official 2023, 2024 and 2025 Basketball Without Borders Global tables with later NBA Draft Combine, NBA G League Elite Camp and Portsmouth Invitational packages, plus a small number of clearly labeled official or reputable listed profiles. For every formal package, the displayed primary height, wingspan and weight come from the same source and date; conflicting Academy-roster or RealGM heights remain alternate listings. Standing reach, hand size and athletic testing are retained when the same source publishes them. Values without a published protocol remain labeled listed or reported, not measured. Apparent source errors—including Parby Musongela’s printed BWB wingspan and Taran Armstrong’s printed G League standing reach—are documented but excluded rather than silently corrected. Later measurements remain explicitly dated and are not presented as a player’s size at an earlier Academy event.

Statistical scouting profiles

The Statistical Scouting Profile classifies each qualifying player through transparent, edition-adjusted Style, Role and Impact percentiles. Primary and secondary archetypes describe tournament production rather than position or projected outcome. Standout indicators identify the clearest statistical signals, while Questions for Film surface concerns and contextual issues that require tape review. Profiles require at least two recorded games and are labeled by confidence.

Career-outcomes registry: 2017, 2019, 2022, 2023, 2024, 2025 and 2026

The Outcomes tab follows every active nonzero-GP player from each included edition, beginning with all 40 active players from the founding 2017 edition and extending through the immediate 2026 pathway cohort. Guo Kuo’s 0-GP row remains in the roster/outcome archive but outside performance-derived denominators. Each record separates current or last verified status from Highest Level Reached and preserves intermediate NCAA Division I, Division II, Division III, NAIA, junior-college, national-team, NBA G League and international professional stops. NBA regular-season appearances are kept separate from NBA or two-way contracts without a verified appearance. “No recent verified record” is a research status, not evidence that a player stopped playing.

The 2017 and 2019 ABI checks are deliberately within-edition. The 2017 cohort uses the complete corrected event package; 2019 has only 17 of 24 available box scores and uses estimated model minutes. The possibility that the 2019 field contained lower overall prospect quality than other included tournaments remains a hypothesis until later cohorts have equally mature career windows. The COVID-19 pandemic overlapped the first two post-event years; the editions currently included in this database jump from 2019 to 2022, but this database does not attribute that coverage gap or any individual outcome solely to COVID without a direct official source. Outcome tiers are a sorting convenience rather than a claim that every international league can be placed on one universal ladder.

The 2022 registry follows all 91 active nonzero-GP players and explicitly separates Highest Level Reached from Current Development Stage. This is necessary because many players remain active in NCAA Division I, Division II, junior college, prep basketball or early professional careers. A prior Basketball Africa League or professional appearance is retained even when the player later enters U.S. college basketball. The 2022 ABI analysis is within-edition and uses model-estimated minutes because all supplied player-minute fields are 0:00.

The 2023 registry follows all 54 active player rows, representing 53 distinct identities because Darian Bailey has two active aggregate rows. It uses the same Highest Level versus Current Development Stage distinction. The cohort is only approximately three years removed from the event, so NCAA, JUCO, prep, academy and early professional placements are treated as developing pathways rather than final career judgments. The 2023 ABI analysis remains within-edition and uses the published player minutes while retaining the edition’s documented source anomalies; Bailey’s two source rows are represented once in derived ABI standardization and ranking.

The 2024 registry covers all 82 active source rows but reports identity-level rates for 81 named players. The anonymous “Player Name” box-score row remains visible for source fidelity and is excluded from all outcome counts and ABI quartiles. The generic analysis layer now deduplicates identities automatically, which also prevents the two Darian Bailey rows from inflating 2023 rates. Because 2024 is only approximately two years removed from the event, prep, academy, JUCO and active college placements are treated as legitimate development stages rather than final outcomes. Ambiguous name matches are left unlinked.

The 2025 registry covers all 86 active player rows and is explicitly a one-year pathway tracker. Most players remain in prep, Academy, junior-college, NCAA or early professional settings, so these placements are not treated as final career outcomes. Prior professional experience—including the Australian NBL, Spanish ACB, Finnish Korisliiga and other senior leagues—is retained when a player subsequently enters U.S. college basketball. Low-confidence records identify research gaps rather than presumed basketball inactivity.

The 2026 registry covers 89 active statistical records. AJ Bowman appeared in two complementary supplied aggregate rows—one carrying minutes and one carrying production—which are consolidated into one displayed performance with the two-row provenance preserved in Data Audit. Because the registry was updated immediately after the July 2026 event, it measures current program placement, confirmed 2026–27 commitments and any verified senior experience that predates the tournament; it is not yet a meaningful test of long-term career outcomes.

Display precision

Totals remain whole numbers. Per-game, per-40, per-100 and advanced values display exactly one decimal, including trailing zeroes. OREB and DREB are displayed separately alongside total rebounds in every basic player view and in the team and player snapshots.

Sorting, stat queries and export

Click a column header to sort. Edition, search, team, minutes and multi-condition stat filters work together. Stat queries use AND logic: every active condition must be met. Games and minutes always use tournament totals; counting stats use the selected query basis; ratios, percentages and advanced metrics use their stored values. The CSV export reflects the visible, filtered table, including its current sorting order.