Fantasy basketball draft model · H2H points · re-runs for your league in the browser

NBA Fantasy Basketball Points League Database

A projection model you can bend to your own league. Set your scoring and your playoff weeks, switch the optional factors on or off, and both the draft board and 24 seasons of backtest re-rank in front of you.

Your league

Scoring presets

PTSFGMFGAFTMFTAFG3MREBOREBASTSTLBLKTOV

DDTDQD40+ pts50+ pts Bonuses stack

Teams Roster spots

Last scoring week Playoffs start Playoffs end

Playoff weight

Weeks are the fantasy calendar for 2026-27 (24 weeks; the NBA Cup and All-Star weeks are merged). Replacement level is teams × roster spots + 1. The board ranks by total value (FP/G × projected games over the full season); the weeks and playoff weight apply only when the League weeks chip is on. Bonuses stack: with Yes a triple-double also earns the double-double bonus; with No it earns only the highest bonus reached.

Model factors — click to turn on or off

Games-played weighting

Player evaluation

Columns

Possession load team usage crowding — players on over-stuffed offenses get docked, under-used offenses get a bump; corrects a measured bias: on the 1,307 players it moves 1+ FP/G it cuts projection error 0.11 FP/G (p=0.008, 24 seasons); neutral on draft capture because it moves whole teams together. On.

Tank discount −15% on playoff weeks for age-30+ players on bottom-6 Vegas win-total teams — late-season rest and shutdown risk. Thin evidence, measured before the 2026-27 lottery reform; not significant. Off for 2026-27 (owner’s call).

Experience yr2-4 small development bump for players in NBA years 2–4; the age curve already carries most of it (+0.2pp, p=0.21). Off.

Roster churn flat discount on players who changed teams; failed every test (best p=0.52). Off.

Efficient heir when a star (32+ FP/G) left, his team’s best per-minute player projected under 30 minutes gets +2 minutes. Tested on 2013-26 after fitting on 1999-2012: +0.09pp, better in 5 of 11 seasons — leans positive, NOT significant. Off.

Backup C + contracts backup centres −0.5 min and −3.9 games; a pay cut −3.9 min, a raise +1.3, a big raise +1.8 (salary vs last season). Tested the same way: +0.09pp, better in 8 of 11 — leans positive, NOT significant. Off.

Destination adjusts movers by the scoring their new team must absorb vs their old one; a real effect (p<0.0001) that still costs capture on the board. Off.

Rookies valued by the college rookie model — per-minute production from draft slot + college stats, minutes from slot + the minutes his new team has open, games from slot; each part significant on its own target (FP/36 p=.003, minutes p=.0003, 12 draft classes). Click to cycle to ESPN-rank placement or off. Rows marked RK. On (college model).

Depth minutes nudge where lineups.com’s preseason depth chart disagrees with projected minutes by 8+, the projection blends 25% toward the chart’s slot (30/21/14/10 by string); nudged rows carry a dot on Depth. On the 102 players it nudged (2019-25) minutes error fell 4.47 → 3.86 (p=0.012); draft capture +0.13pp, not significant. On.

Games-played weighting how much recent seasons count when projecting games played; 7/2/1 is the shipped setting.

Player evaluation who sets the player order: this model, or the FantasyPros consensus re-scaled by your league’s math (schedule, playoff weeks, replacement).

Position

Team

Sort by

Draft mode — click a row when a player is taken

TANK −15% on playoff weeks — age 30+ on a bottom-6 Vegas team (the only tag that changes value)

RK rookie — valued by the college rookie model (stat line = his college per-40 shape scaled to projected minutes and FP/G); rookies it doesn't cover sit at ESPN’s rank with “-” stat columns

NEW new team — info only, zero effect (priced via team crowding instead)

SHARE incumbent absorbing 60+ FP/g of new teammates — info only

INJ not good to go for the season start, a legitimate injury risk entering it, or projected under 52 games by the model — from ESPN’s live injury list, the owner’s researched notes, or the projection itself; hover for the specifics. Cleared offseason surgeries are NOT tagged. Info only

• depth minutes nudge applied — hover the Depth cell for the FP/G multiplier

green ESPN/ADP/Sleeper 15+ picks cheaper than our rank (pale green 7–14) · red ESPN/ADP/Sleeper 15+ dearer (pale red 7–14)

#	PR	ESPN	ADP	FPros	Sleeper	Player	Tm	Pos	Depth	Min	Usg	Load	Pts	Reb	Ast	Stl	Blk	3PM	TO	FP/G	Total	eFP/G	ΔFP	LY FP/G	LY Tot	GP	Avail	Gms	PO	Δ

Total is projected points inside your scoring window, at the games he is expected to play. Δ is his move from the shipped defaults. PR is his projected rank and does not change when you sort.

#	Player	Tm	Pos	Comb	ESPN	ADP	FPros	College	ESPN line	Pts	Reb	Ast	Stl	Blk	3PM	FP/G	GP

8-team H2H points · 2026-27 · verified against ESPN league settings

How the board is built

One formula, built up in five stages, then a documented layer of case-by-case judgment on top. Every number below is live from the current board, and one player — Tyrese Maxey — is carried through every stage so you can watch his rank get made.

The whole thing in one line

The formula

A player’s value is the fantasy points he puts on your scoreboard, week by week, across your season. The key idea: when he sits out, you don’t score zero — you pick someone up off waivers. In an 8-team league only 104 players are rostered, so the waiver wire always holds a real NBA starter. A missed game costs you the gap to that player, not the whole game.

for each fantasy week: contribution = games_that_week × leverage × tank_discount × [ availability × fp_per_game + (1 − availability) × recovery × replacement_fp ] value = sum of every week’s contribution vorp = value − value of the 105th-best player → sort by value = the board

1fp_per_game — what he scores in a game he plays

2fp_per_game, adjusted — his team’s context

3availability, recovery, replacement_fp — the games he misses

4games_that_week, leverage, tank_discount — your calendar

5value, vorp — add it up, subtract what’s free, rank

Every control on this page changes one term. Your scoring settings change stage 1, the factor toggles change stage 2, waiver recovery changes stage 3, playoff weeks and weight change stage 4, and teams × roster spots change stage 5. The browser re-runs all five stages for every player whenever you touch anything.

Every parameter was chosen by walk-forward testing: build the board for a past season using only what was knowable before it, then score it against what actually happened. Twenty-four seasons, 1999-00 to 2025-26 — the last one held out of all tuning and spent once, blind, at the end. The Proof tab shows every season.

Stage 1 · fp_per_game

What he scores in a game he plays

Built in two separate pieces, because they behave differently: how productive he is per minute, and how many minutes he plays. Each comes from his last three seasons, with last season counting most. Multiply them and you get a full per-game stat line (points, rebounds, assists, steals, blocks, threes, turnovers…), which is then scored under your settings — including the double-, triple- and quadruple-double bonuses.

MAXEY — stage 1 per-minute production, last three seasons 2023-24 70 gm 37.5 min 43.5 FP per 36 2024-25 52 gm 37.7 min 43.8 FP per 36 2025-26 70 gm 38.0 min 49.7 FP per 36 <- weighted 7x weighted 7 / 2 / 1 ~47.8 FP per 36 age 25 -> still improving x 1.032 minutes, weighted the same way 36.8 per game rate x minutes -> stat line -> your scoring = 50.0 FP/game

How it was derived

Weights 7 / 2 / 1 Last season counts seven times as much as three seasons ago. The textbook weighting (5/4/3) was the worst of every configuration tested; 7/2/1 beat it in every test season on both rank accuracy and value captured. Recent form matters more than the textbook assumes.

Per minute, not per game Across 289 measured in-season trades, a player’s per-minute production did not change when he moved (p = 0.39) — what changed was his minutes. So skill and playing time are projected separately, and a player whose minutes swung isn’t punished for it. Minutes are the single largest source of error in the model, which is why most manual adjustments (below) are minutes calls.

Age curve About +1.6% a year below age 27, −2.4% a year above it. These slopes were doubled in September 2026 after an error audit of 7,011 player-seasons found the gentler curve under-projected players 24 and younger by 2.0 FP/g and over-projected 33+ by 1.5. Doubling won 16 of 23 seasons (+0.43 points of value captured, p = 0.015), with a broad sweet spot from 1.75× to 2.75×. One carve-out: the old-age miss lives entirely in aging role players. Players 33+ still projecting above 35 FP/g were actually under-projected, so they keep the original gentle decline (1.2%/yr).

Rookies No NBA minutes means nothing to project, so rookies are seated at their market rank (ESPN’s, with the expert consensus as fallback) and marked RK. Base rate worth knowing: over eight seasons a top-5 pick’s median finish is 101st — right at replacement level.

Honest limit This is the stage where people beat the model. Over six seasons with a frozen FantasyPros consensus, the experts’ player order captured 2.4 points more value (p = 0.034) — they read camp reports and medical news; box scores can’t. The Player evaluation switch on the Board tab lets the experts set stage 1–2 order while stages 3–5 still apply your league’s math. Walk-forward, that hybrid tied the experts and beat this model by 2.3 points.

Stage 2 · fp_per_game, adjusted

His team’s context

Stage 1 projects every player in the role he held last year. That breaks when a team collects several players who were all the first option somewhere else — there are only so many shots to go around. This stage docks players on over-crowded offenses and bumps players on thin ones.

MAXEY — stage 2 Philadelphia's roster, as projected, would use 136% of the possessions that actually exist (Load column; 1.00 = it adds up) crowding correction for every Sixer -2.0 FP/game 50.0 -> 47.97 FP/game

How it was derived

Possession load The model’s over-projection rises in a straight line with team load — −0.8, +0.7 and +1.4 FP/g across low, middle and high load bands, over 3,249 player-seasons. The size of the correction is fitted to that measured bias (7.09 FP/g per unit of load), learned only from prior seasons. Simply forcing each team’s possessions to add up was also tried: it over-corrects about 2.5× and lost at all four strengths tested.

Depth-chart nudge Where the lineups.com preseason depth chart disagrees with projected minutes by 8 or more, minutes blend 25% toward the chart slot (30 / 21 / 14 / 10 minutes by string). Six-season test: +0.13 points, not significant. On for 2026-27 only by the owner’s call, because this is an outlier year for injury returns and moves; nudged rows carry a dot in the Depth column.

Tried and dropped A discount for changing teams (p = 0.52 — about 38% of players are new or returning every year, so it moves everyone together); a destination-quality adjustment (a real effect, p < 0.0001, that still cost accuracy on the board); a bump for NBA years 2–4 (the age curve already carries it). All three remain as toggles, off by default, so you can see what they would do.

Stage 3 · availability, recovery, replacement_fp

The games he misses

Availability is the share of his team’s games he is expected to play: projected games ÷ 82. For the games he misses, you get partial credit — a recovery share of what a replacement player off waivers would score. Recovery is high in the regular season and low in the fantasy playoffs, when you can’t afford to guess.

MAXEY — stage 3 games played, last three seasons 70 / 52 / 70 -> weighted 66.4 blend: 1/3 his record + 2/3 league average (59.7) (66.4 + 2 x 59.7) / 3 = 62 games availability 62 / 82 = 0.756 a missed game is worth, from waivers: regular season 0.85 x 29.3 FP = 24.9 FP fantasy playoffs 0.35 x 29.3 FP = 10.2 FP

How it was derived

Why blend with league average Games missed last year explain only 11% of games missed next year (r² = 0.11) — mostly noise. So his own record is trusted one-third. The most accurate setting was one-fifth; one-third was chosen deliberately, costing about a tenth of a point of accuracy for roughly 70% more separation between fragile and durable players. It cuts both ways: the fragile lose more, the durable gain more.

Injury type 98,108 injury-report entries were classified by body part to test whether chronic injuries predict future absence better than acute ones. They don’t in any way that improved the board, so the model only counts games missed. Known surgery timelines are handled as manual adjustments instead (below).

Replacement 29.3 FP/g The median scoring rate of the next 16 players past #104 — the realistic waiver pool. A median rather than the single #105 player, because one fragile high-scorer sitting on the boundary once dragged it from ~30 to 37.6 and over-rewarded injury-prone players.

Recovery 0.85 / 0.35 Regular season: how often an absence is known before lineups lock, given unlimited adds (structural range 0.77–0.95). Playoffs: absences bunch up 1.23× in playoff weeks and there’s no time to recover a bad week; walk-forward, raising it cost accuracy (−0.3 points, p = 0.033). Lower the regular-season number if your league limits adds or IR.

Stage 4 · games_that_week, leverage, tank_discount

Your calendar

Everything is priced week by week on the published 2026-27 NBA schedule, so a team with more games in your window is worth more. Only weeks 1–22 count (every team plays 72–74 games in that window). Your playoff weeks, 19–22, count 1.4× because they decide the title. And one narrow group — veterans on teams likely to tank — loses 15% in those playoff weeks.

MAXEY — stages 1-4 put together Philadelphia: 59 regular-season games + 15 in playoff weeks no tank discount (PHI Vegas line 50.5) regular games he PLAYS 59 x 0.756 x 47.97 = 2,139 regular games he MISSES 59 x 0.244 x 0.85 x 29.3 = 359 playoff games he PLAYS 15 x 1.4 x 0.756 x 47.97 = 761 playoff games he MISSES 15 x 1.4 x 0.244 x 0.35 x 29.3 = 53 value = 3,311

How it was derived

Schedule Taken straight from the published fixtures, so it is knowable on draft day — not hindsight. Playoff-week games range from 12 (Cleveland) to 15 across the league: a real but modest 4–6% of value.

Playoff weight 1.4 Started at 2.0. Testing showed the playoff-week schedule barely predicts playoff-week scoring (r = 0.075), so it came down. A sweep from 1.0 to 2.0 across ten seasons was flat (no setting differed by more than 0.16 points), so 1.4 is a stated preference — playoff weeks matter more — that can’t be mis-set enough to hurt.

Tank discount −15% in playoff weeks for players aged 30+ on the bottom fifth of preseason Vegas win totals (currently Sacramento 21.5, Brooklyn 24.5, Milwaukee 25.5, New Orleans 27.5, Memphis and Chicago 29.5 — 11 players flagged TANK). Being on a bad team in general does not cut production — tested against three measures of team quality, every one failed. The effect exists only for veterans on genuinely tanking teams, who kept 80% of their playoff-week minutes versus 95% for younger teammates. Thin evidence (55 veteran-seasons), stated plainly.

Stage 5 · value, vorp

Add it up, subtract what’s free, rank

Eight teams × thirteen roster spots = 104 players rostered, so the 105th-best player is free — he’s on waivers all season. VORP (value over replacement) is how much more a player is worth than him. Players are ranked by value, then grouped into tiers.

MAXEY — stage 5 Maxey's value 3,311 the #105 player's value (free on waivers) - 2,157 VORP 1,154 -> RANK 7

How it was derived

Replacement #105 Straight arithmetic from the verified league settings. This is the single fact that separates this league from the 12-team default public rankings are built for: here replacement is a genuine NBA starter, which is why durable high-floor players rank higher than public boards have them. Checked against daily lineups (only 10 of 13 start each day): across 24 seasons of simulated drafts just 0.6% of drafted players’ points land on the bench, and a starters-only replacement (#81) scored worse on every test — #105 holds.

VORP vs order Subtracting the same number from everyone can’t reshuffle anyone — the order comes from value. VORP tells you what a gap is worth: whether a reach is justified, or how to weigh a trade.

Tiers, not a strict order The board’s rank correlation with real outcomes is about 0.72, and roughly a quarter of picks won’t finish top-105 no matter how good the ranking. It can tell tier 3 from tier 6; it cannot tell pick 31 from pick 34. Inside a tier, draft for fit.

Position scarcity Tested and left out: re-ordering by position moves players an average of 2.3 places and changes who is draftable by nobody.

On top of the formula

Manual adjustments

The formula reads three years of box scores. It cannot see news: a surgery announced in September, a trade, a new starter, a star leaving a team. So on top of it sits a documented judgment layer, set player by player in the September 2026 review of every top-130 player. No formula parameter is ever changed to fit an opinion — judgment lives only in these rows, each with its reasoning written down, and each is deleted when the news resolves.

Games · switched off in this version

Replace projected games (stage 3) where a draft-day fact contradicts history: a team-announced surgery timeline (Butler’s ACL → 28 games), or a past absence that wasn’t injury, or an owner durability call.

Role · switched off in this version

Scale a player’s whole stat line (stage 1) for a role change history can’t see — set his minutes (Giannis 32, Naz Reid 30 as a new starter) or his per-minute rate (a usage squeeze or a promotion to first option).

Rosters · 0 rows

Trades and signings the NBA’s data hasn’t caught up with yet, so each player is priced on his real team. Removed when the official data catches up — so far every handover has moved players by a place at most.

Injury notes · 21 rows

Display only: the INJ tag and its hover text for players not ready for opening night or carrying real risk. They inform your judgment; they don’t change value.

Every games and role row — what changed, from what to what, and why — is listed in rank order in the Case-by-case adjustments ledger at the foot of the Board tab. Two rules keep the layer honest: a claimed injury is checked against actual games played and the article’s date before it ships (seasonal news gets recycled), and when a check overturns a call, the row changes and the page says so.

Season-by-season scorecard — the full formula vs every other method

Every method drafts the same players and is scored the same way, one season at a time, using only what was knowable before that season. Total points = share of the best possible full-season fantasy points captured by drafting the top 104 (what the total-value board is built to predict). Standard = the same with this league’s fantasy weeks and 1.4× playoff weeks. Realistic = an 8-team snake draft played out day by day with 10 starters and a waiver pickup whenever a drafted player sits out — the only test that credits the waiver wire. The model row is today’s board, ranked by total value (FP/G × projected games). League default settings; the interactive table further down re-scores under YOUR settings with the simpler base formula.

Test

Did it work? Twenty-four seasons, rebuilt as if it were draft night

For each season below the board was built using only what was knowable before that season started — three prior years of box scores, the published schedule, the age curve — and then scored against what actually happened, under your scoring system and playoff weighting. Where ESPN’s own pre-draft rank exists it is scored on the identical test, so the comparison is like for like.

ESPN column

ESPN’s published H2H points rankings as they stood before each season, recovered from dated Wayback Machine captures (16 Oct 2025, 7 Oct 2024, 3 Oct 2023, 23 Oct 2022, 20 Oct 2021). Corrected 1 Oct 2026: 2025-26 previously used an API rank list that turned out to be frozen in June 2025 (it still ranked Haliburton 20 after his Achilles tear); the genuine October article replaces it, and on it ESPN beat this model for 2025-26. The fantasy API’s stored rank for a past season is refreshed during the year and carries hindsight, so it is not used.

A human editor with every roster move, camp report and injury update on the day is a genuinely strong benchmark, and this model reads only box scores and a schedule — read the comparison with that in mind.

FPros column & the hybrid

FPros is the FantasyPros expert consensus for Yahoo points scoring, frozen at each tip-off and verified the same way. It is a multi-expert panel (6–12 experts per season), so a stronger benchmark than any single writer, though built for a different points format.

Experts × your league math re-ranks that consensus with this model’s magnitudes and your schedule, scoring and playoff weighting — the hybrid described on the Method tab.

What gets scored

Every season’s board is today’s full formula, built only from what was known before that season: projection with the late-season role blend, team load, availability with the injury-news, ended-hurt and returner rules, and rookies from the college rookie model (from 2012-13, the first class with three earlier college classes to learn from; rows marked RK). The optional factors on the Board tab stay off. Ranked by total value (FP/G × projected games) and scored on real full-season points unless League weeks is on.

Compared with the bare projection these boards used before 1 Oct 2026, scored against the same perfect draft: straight-down capture 89.2% vs 89.3% (a tie, 13 of 24 seasons); the realistic daily-lineup draft 95.3% vs 95.2% (+0.14pp, 15 of 24, not significant). Each added part earned its place on its own target (projection accuracy); together they move draft capture only a little.

The playoff window on this tab uses the league’s default weeks; custom windows apply to the live board.

Seasons left out, and why

2019-20 was suspended in March 2020 and finished in the Orlando bubble with 22 of 30 teams — 63–75 games and no comparable late-season calendar (where your playoff weeks live). 2020-21 was a delayed, compressed 72-game season with heavy COVID-protocol absences. 2011-12 was the 66-game lockout. Scoring a board against those measures the distortion, not the board.

Run as an exhibition anyway (capture self-prorates): 2019-20 board 90.6% vs FantasyPros 93.2%; 2020-21 board 87.1% vs 88.9%. Same direction as the official seasons — excluded for calendar comparability, not because they flatter anyone.

Season

Sort by

#	Player	Tm	Proj	Proj pts	Actual	Actual pts	GP	Miss	ESPN	ESPN miss	FPros	FPros miss

Value captured is the share of the best-attainable points a team would have banked by drafting straight down that ranking to fill every roster in the league. Miss is actual rank minus projected rank: positive means he finished lower than projected. A model that only ever had to beat last year’s totals would sit around 0.87; random drafting sits near 0.60.

The pre-draft opinion archive

Every outside pre-draft opinion this project has recovered and verified, season by season, as raw rows. These are the benchmarks the Proof tab scores — here you can browse what each source actually said before each draft. ESPN projections are full stat lines, so they re-score under your scoring settings above.

Source

Season

NBA Fantasy Basketball Points League Database

Rookie board

Case-by-case adjustments

My team (0/13)

How the board is built

The formula

What he scores in a game he plays

How it was derived

His team’s context

How it was derived

The games he misses

How it was derived

Your calendar

How it was derived

Add it up, subtract what’s free, rank

How it was derived

Manual adjustments

Season-by-season scorecard — the full formula vs every other method

Did it work? Twenty-four seasons, rebuilt as if it were draft night

The pre-draft opinion archive