How history becomes a playable decision.
Full Court Games turns historical regular-season records into fictional strategy games. This page documents the source, transformations, ratings, lineup model, simulation behavior, and important limitations.
Season production
Points, rebounds, assists, field-goal percentage, team, season, and games played are aggregated from regular-season player game logs into a comparable player-season record.
Role + position
Listed positions determine where a season can play naturally. Moving one positional step costs 3 fit points; moving two costs 6; larger gaps cost 9 before lineup power is calculated.
Game-design ratings
Overall and category ratings are original design values. They organize the game and are not official NBA ratings, scouting grades, predictions, or claims about a player’s career rank.
Historical inputs and attribution
The primary archive is Eoin A. Moore's Historical NBA Data and Player Box Scores dataset. The local generation pipeline reads player game logs, player identities, and team histories; keeps NBA and BAA regular-season records; reconciles franchise abbreviations and name variants; and aggregates game rows into season averages.
The playable catalog contains thousands of player-seasons. A small group of landmark stars is manually reviewed so familiar peak seasons remain available, while the larger historical universe is generated from the same normalized schema. All-Star and Hall of Fame lists are curated separately to build focused game pools.
The dataset is evidence, not an authority beyond correction. Older records can be incomplete, team abbreviations can change, and source rows can contain naming or classification errors. Suspect seasons are filtered instead of silently promoted into the game, and visitors can report a correction through the contact page.
From a season line to game impact
Every season starts with an overall rating, then receives bounded adjustments for scoring volume, rebounding, playmaking, field-goal efficiency, and All-Star pedigree. The adjustments are intentionally small: the overall rating remains the center of the model, while the box score distinguishes seasons that arrive at a similar rating through different kinds of production.
Six category grades—scoring, shooting, passing, rebounding, defense, and ball handling—support the player and team reports. When a manually reviewed category profile is unavailable, the fallback model scales points, field-goal percentage, assists, and rebounds, then blends position-aware priors for defense and handling. Values are bounded from 55 to 99 so the letter grades remain readable across eras.
What the grades do not mean: they do not reproduce an official rating system and should not be used as a scouting evaluation. Box-score inputs cannot fully measure spacing, off-ball movement, defensive assignment difficulty, era-specific rules, or playoff translation. The grades are deliberately transparent game mechanics.
Fit, versatility and five-player power
A lineup has one PG, SG, SF, PF, and C slot. A listed position is a natural fit. An adjacent mismatch receives a 3-point penalty, a two-slot mismatch receives 6, and a larger gap receives 9. The computer-controlled teams test every assignment and choose the arrangement with the smallest total fit penalty.
Lineup power is the average season impact of all five players, plus a small reward for multi-position versatility and natural balance, minus 42% of the accumulated fit penalty. This makes roster construction matter without erasing the quality of the seasons selected. A versatile wing can solve two jobs; five centers cannot avoid paying for the experiment.
The team report separately averages the six category grades, identifies the two strongest and two weakest traits, and provides the player-level evidence behind the summary. That report is descriptive; the core simulation uses season impact and positional fit rather than secretly rewarding the displayed letter grade.
How the fictional simulations behave
Auction Draft compares each completed lineup with the four-team field and a normal NBA-quality schedule. Absolute lineup strength establishes the baseline, relative strength separates the drafted teams, and a small roster-derived form adjustment prevents every close field from producing identical records. The result is clamped between 48 and 79 wins.
Beat the Champ is a single-game model. Lineup power shifts each side's scoring baseline, both teams receive a narrow random scoring range, and consecutive wins add a small momentum bonus. Better lineups are favored but not guaranteed to survive a one-game elimination format.
These outputs exist for entertainment and replayability. They are not predictions of real games, betting advice, official player evaluations, or historical counterfactual claims. No fictional budget, bid, score, or result has cash value.
Who maintains this?
Full Court Games is an independently maintained fan project. Its interface, rules, transformations, ratings, and written analysis are original to the project.
Can the process be inspected?
Yes. The source repository contains the generation pipeline, game formulas, tests, and issue tracker.
Why publish the model?
The decisions are more interesting when players can understand the tradeoffs, challenge assumptions, and report data or design problems.
What happens when data is wrong?
A report is checked against the source archive and the normalization rules. Confirmed issues can be corrected in the catalog or generation pipeline.