Draft Science · White paper
How does Pikola decide who to draft? By the weeks each player wins you.
The short answer
Draft for the weeks a player wins your team, not for a place on a list. Every time you pick or bid, Pikola asks one question: how much would this player raise your chance of winning your weeks, with the team you have, against the managers in your draft? In simulated auctions, drafting that way won about three times the titles of a manager sticking to a good cheat sheet.
- Not every category wins you weeks equally. A cheat sheet counts all nine the same. Pikola counts each by how much it moves a week. Steals and the shooting percentages swing so much from week to week that an edge in them counts for less: in a 14-team league, an edge in steals counts about half as much as the same edge in blocks, and an edge in either percentage not much more.
- What a player adds depends on your team, so Pikola re-ranks after every pick. Take three centres and you'll win rebounds and field-goal percentage most weeks, while threes and points slip away, so players who help in the races still open move up. Against a league of cheat-sheet drafters, re-ranking alone won about 1.4 more categories out of every 100 played, and Pikola, which also counts turnovers at half, about 2.7: one extra category every four weeks.
- The other managers set the price. In an auction, Pikola works out what a player will cost from the money left and how real drafts spend, and sets your top bid. In a snake draft, it works out who will still be there at your next pick. When we switched parts of Pikola off one at a time, knowing how real drafts spend was one of the two that clearly mattered.
Most managers draft from a list
Most managers draft from a list: Yahoo's draft order, an expert's ranking, or a cheat sheet. The best cheat sheets are built from projections. They add up how far above average each player is in each of the nine categories, and rank players by the total. It's a sound way to sort players, and it's what Pikola has to beat.
But a list can't see your draft
A list gives every manager the same ranking, and it never changes during the draft. Three things it can't see change what a player is worth to you.
- Some categories decide weeks more than others. A big edge in steals is often wiped out by one quiet week. An edge in blocks holds up better.
- Your team changes what you need. If you already win rebounds most weeks, another rebounder adds little. If you've all but lost threes, another shooter adds little too.
- The other managers decide the price. In an auction, what a player costs depends on the money left and how the other managers spend. In a snake draft, a player who'll still be there next round isn't worth your pick now.
Pikola is built to see all three. Below, one real pick shows how, step by step; then the tests show whether drafting that way wins.
How Pikola thinks about a pick
Every time you're on the clock or a player is up for bid, Pikola runs four steps. Exhibit 1 follows one real pick through them. It's pick 50 of a 14-team snake draft, you pick seventh, and you've taken three centres: Scottie Barnes, Domantas Sabonis and Jalen Duren. Every other team takes the best player left on Yahoo's list. Your next pick is 63.
Exhibit 1How Pikola works, in an auction and in a snake draft
Pick 50, through Pikola's four steps. Steps 1 and 2 work the same in any draft. Then the path splits: green follows this snake draft to the pick; blue shows what Pikola does instead in an auction. Every number is Pikola's own.
Source: Pikola, run on this draft at pick 50, with the other 13 teams taking the best player left on Yahoo's list. Step 1's chances are Pikola's model of a week against those teams. Step 2 ranks the 381 players you could still add by what each would add to an average team's chances, and to yours. Step 3 simulates the next 12 picks 200 times. The auction path describes its two steps rather than computing them, because this example is a snake draft.
The sections below take the steps in turn, coming back to pick 50 at each. Then the tests: does drafting this way win?
Step 1: Picture your weeks. Some categories win more of them than others
Pikola starts by picturing a week in your league: who you'd start, how many games they'd play, and how each category would go against each team. That gives your chance of winning each category. Back at pick 50, your three centres win rebounds about three weeks in four, but threes only about one in seven (Exhibit 1, step 1).
Here Pikola parts ways with a cheat sheet. A cheat sheet counts an edge in any category the same. Pikola counts it by how many more weeks it wins you, and that differs a lot from one category to the next (Exhibit 2).
Exhibit 2A typical edge, and how many more weeks it wins
Two evenly matched teams; one gets a typical edge in a category, about the gap between a good and an average drafted player that week. The faint bar is how much the category swings from week to week; the dark bar is the edge. The bigger the edge against the swing, the more weeks it wins. A 14-team league.
How much it swings week to weekA typical edge
Source: computed by Pikola, not measured. A 14-team Yahoo league, nine standard categories, Yahoo's default roster, this season's projections. The swing is the standard deviation of the weekly margin between two average teams; the edge is one standard deviation of drafted players' weekly output, the unit a z-score cheat sheet counts in; ties count for the whole-number categories.
Read it like this: a cheat sheet scores a typical edge the same in every category. In blocks, that edge is a fifth of the weekly swing, and lifts your chance of winning blocks from 50% to about 58%. In steals it's a tenth of the swing, and lifts your chance only to about 54%. Equal on a cheat sheet, about half the payoff in weeks won.
Pikola works the exact numbers out for your league. In every league size we checked, from 10 to 16 teams, steals counted least apart from turnovers, and both percentages counted below average. Turnovers count half by design: of the rules we tried, it was best or tied for best in every kind of league, most likely because the busy players who commit turnovers also win points, threes and assists.
So what: don't pay list price for a player whose value is mostly steals or shooting percentages. On its own, that won't beat a good cheat sheet. It pays off once you also ask what your team needs, which is step 2.
Step 2: Rank the players left for your team, after every pick
Next, Pikola ranks every player left by how much they'd raise your chance of winning the week, with the team you have. Your team changes with every pick, so the ranking does too. Start with what three centres did to your weeks (Exhibit 3).
Exhibit 3Three centres in: where each race now stands
Your chance of winning each category at pick 50: hollow if you'd taken the best player left on Yahoo's list three times, filled with the three centres instead. Over 70%, a category is mostly yours; under 30%, Pikola gives it up; in between, it's still in play. "Now worth" is how much more or less an edge in that category is now worth to Pikola.
Best player left, three timesThree centres
Source: two drafts run in Pikola from the seventh seat of a 14-team league, the other teams taking the best player left on Yahoo's list. In one, the seventh seat does the same for three rounds; in the other, it takes the best centre left each time: Scottie Barnes, Domantas Sabonis and Jalen Duren. The chances are Pikola's model of a week against the other 13 teams, with open roster spots filled by average picks. "Now worth" is the change in Pikola's weight on each category between the two drafts.
With three centres, rebounds and field-goal percentage became mostly yours, threes and points slipped under 30%, and assists climbed back into play.
So Pikola now counts threes and rebounds about a quarter less: threes are slipping away and rebounds are all but won, so an extra three or rebound changes your week less. Assists, back in play, count a third more. Nothing switches a category off: under 30%, Pikola calls it given up, but still counts what an extra unit would win you. Points, at 29%, even count a little more, because the total weight is fixed and what threes and rebounds lose goes to the rest. That's how a punt finds you instead of being planned.
Pikola lets at most two categories slip under 30% like this (three in a one-win league, where the week goes to whoever wins more categories). If a third slips under, Pikola pushes back, ranking players who help it higher. In our punt study, no punt planned before pick one won more titles than staying flexible, and most clearly cost some.
Back at pick 50, that reorders the players left (Exhibit 1, step 2). For an average team, Pikola's best player left would be Matas Buzelis, with Desmond Bane second and Darius Garland fourth. Buzelis brings blocks, threes and rebounds; your centres already win you rebounds, and threes are slipping. So for your team Buzelis drops to fourth. Bane gives back a little on threes but gains on assists, and moves from second to first. The point guards, who bring the most assists, move up most: Garland from fourth to second, Dejounte Murray from ninth to third.
Does re-ranking after every pick matter? We tested it in simulated snake drafts against a league of cheat-sheet drafters (Exhibit 4).
Exhibit 4Extra categories won per 100 played, against a cheat-sheet drafter
Simulated 14-team snake drafts in a league of cheat-sheet drafters, 336 seasons for each way of drafting, measured on 28 September 2026. A hollow bar is so close to zero that it could just be luck.
Extra categories won per 100 playedHollow: could be luck
Source: simulated 14-team snake drafts, 14 seats × 3 league draws × 8 seasons = 336 seasons per way of drafting, the other 13 teams drafting by a z-score cheat sheet. Each way of drafting took the same seat in the same league, with the same luck. Method review of 28 September 2026. The hollow bar is within two standard errors of zero: it could be luck.
Read it like this: in a league of cheat-sheet drafters, counting categories by the weeks they win, but the same way for every team, did no better than a cheat sheet. Re-ranking for your team after every pick is what won: about 1.4 more categories out of every 100 played, and 2.7 with turnovers counted at half, as Pikola does. That's about one extra category every four weeks, from the draft alone.
So what: re-rank after every pick. The best player on the list is often not the best player for your team, and the gap grows as your team fills up.
Steps 3 and 4: Price in the other managers
The last two steps bring in the other managers. They work differently in the two kinds of draft: first the snake draft, where pick 50 left off, then the auction.
In a snake draft, step 3 asks who'll still be there at your next pick. Pikola expects each player to go near their usual spot on Yahoo's list: give or take a pick or two at the very top, and about 21 picks either way around pick 100. It simulates the other managers' picks before your turn 200 times and counts how often each player is still there.
Step 4 makes the pick: the best player now, with your next pick in view. Back at pick 50, your next pick is 13 picks away, at 63 (Exhibit 5). Bane, top of your ranking and usually taken around pick 53, is still there at 63 in only 39% of the simulations. Garland, usually taken around 64, is still there in 65%; Murray, usually around 70, in 79%. So Pikola takes Bane now and plans on Garland or Murray at 63.
Here, looking ahead agrees with the plain ranking, because the player who adds most is also the one least likely to come back. If instead the top player would likely still be there at 63 and the runner-up wouldn't, Pikola would take the runner-up now and wait. At the turn, where your two picks come back to back, Pikola weighs them together.
Exhibit 5Who'll still be there at your next pick
Pick 50, with your next pick at 63. Each curve shows where the other managers are likely to take that player: centred near their usual pick, and wider later in the draft. The faint part, before pick 50, is already past. The shaded part, after 63, is the chance they're still there for you. The percentages come from Pikola's 200 simulations.
Desmond Bane, usually around pick 53Darius Garland, usually around pick 64
Source: Pikola's model of where each player goes, at its default spread (1 + 0.2 × the player's usual pick, in picks): a bell curve around Yahoo's order, starting from the current pick. The percentages are the share of 200 simulations of the next 12 picks in which each player is still there at 63. The curves alone give 33% and 64%; the simulations come out a little higher because exactly 12 players must go. Computed, not measured; the spread was fitted to one real league's draft.
If your league runs an auction instead, step 3 works out what each player will go for. Pikola starts from what real auctions pay: four seasons of a real Yahoo league, scaled to your league's size and budget, blended with Yahoo's prices. After every sale it re-prices everyone, sharing out the money the other managers have left: if they overspend on stars, everyone else gets cheaper. It also learns how your league bids: whether stars go for extra, how fast the money goes, and how hard each manager pushes.
Step 4 sets your top bid: what the player is worth to your team, checked against what the same money could buy instead. Early on that check is rough, so it gets little say; it counts for more as your team fills up. So your top bid can sit above the going price when you need what a player does, or below it when you don't.
To see where prices start, Exhibit 6 shows a 14-team, $200 league before the first sale: the top player at about $81, the 14th-best at about $42, the top 42 picks (three per team) taking about 60% of the league's $2,800, and 34 players going for $1. Every sale re-deals this curve over the money that's left.
Exhibit 6What each player is expected to go for, best first
Pikola's expected price for each of the 182 players a 14-team, $200 league drafts, best first, before the first sale.
Source: computed by Pikola's engine for a 14-team, $200 Yahoo league: four seasons of a real league's auction results, each season's prices by rank scaled to this league's size and budget and averaged, blended with Yahoo's projected and average prices, and adding up to the league's $2,800. Prices by rank only; no player is named.
Which of these steps matter most? When we switched parts of Pikola off one at a time in simulated auctions, two clearly mattered: knowing how real drafts spend, worth about two extra categories out of every 100 played, and picturing the ten players you actually start each week, worth about the same.
So what: in an auction, know how your league spends, not just what a player is worth. In a snake draft, when two players are close, take the one who won't make it back. Our simulated leagues couldn't measure a gain from that, so treat it as a tiebreak.
Does it work? It wins more titles than the common ways of drafting
We tested each kind of draft against the way people in it most often draft: snake drafts in 16-team leagues against a manager following Yahoo's order, and auctions in 14-team leagues against a manager sticking to a good cheat sheet. Each test is paired: Pikola and the other manager draft in the same league, with the same players and the same luck, so any gap comes from how they draft. The full method for the auctions is in How Pikola is tested.
- +18 to +29titles per 100 seasons over a manager drafting by Yahoo's order, in snake drafts
- 3×the titles of a manager sticking to a good cheat sheet, in auctions
- 7 of 7kinds of simulated league, snake and auction, where Pikola came out ahead
Exhibit 7Extra titles per 100 seasons for Pikola
Snake: 16-team leagues, against a manager drafting by Yahoo's order in the same seat, 128 drafts of 8 seasons in each kind of league. Auction: 14-team leagues, against a manager sticking to a good cheat sheet in the same auction, 40 auctions of 32 seasons in each.
Snake draftsAuctions
Source: snake, the pre-registered 16-team test: 16 seats × 8 league draws = 128 paired drafts, 8 seasons each, seasons played on the projections, Pikola against a manager drafting by Yahoo's order in the same seat. Auction: 40 fourteen-team auctions × 32 seasons per kind of league, Pikola and a cheat-sheet drafter in the same auction, 2 October 2026. Every bar is more than two standard errors above zero.
Read it like this: in snake leagues, Pikola won 18 to 29 more titles per 100 seasons than a manager following Yahoo's order, even in a league where every other team counted categories the way Pikola does. In auctions it won about three times the titles of a manager sticking to a good cheat sheet, about 26 against 9. For scale, an average team wins about 7 titles per 100 seasons in a 14-team league, and about 6 in a 16-team one. When the same snake drafts were scored on projections 15% off, or in leagues whose managers knew things the projections got wrong, the gap ranged from about 5 to 31 titles per 100 seasons, and never closed.
So what: don't draft from a fixed list. Whether it was Yahoo's order or a cheat sheet, it left titles on the table in every kind of league we built.
What this means on draft day
- Count weeks, not rank. Ask what a player does to your weekly races, not where they sit on a list.
- Discount the streaky categories. Steals and shooting percentages win fewer weeks than their rankings suggest.
- Re-rank after every pick. Let races you've won or lost go, and chase the ones still in play. That's how a punt should find you.
- Price in the other managers. In an auction, how your league spends and the money left matter as much as the player. In a snake draft, when two players are close, take the one who won't make it back.
The methods, for readers who want them
Each step above rests on a method chosen for a reason. Here they are, named, with the reason, for readers who want them. The full model follows, folded.
- A week's totals. Method: each category's weekly total is a normal distribution, built from per-game projections, expected games played, and the ten lineup slots your league starts, so a bench player counts only in the weeks they'd start. Percentages come from makes and attempts. Why: the question is how often your total beats theirs, which needs the spread of a week as well as its average. A sum of thirteen players' season totals answers a different question.
- Winning a category. Method: the chance one team's total beats the other's, with ties counted for the whole-number categories. Why: steals, blocks and threes tie often in a real week. Ignoring ties overstates how often an edge wins.
- Winning the week. Method: in a categories league, where each category is its own win or loss, the expected share of the nine categories won. In a one-win league, the chance of winning five or more, combining all nine chances at once (a Poisson-binomial distribution). Each opponent is scored separately and the results averaged. Why: averaging the opponents first would say you're about even against everyone. In fact you're favoured against weak teams and not against strong ones, and the week's result depends on which you're playing.
- The playoffs. Method: playoff weeks get their own weight, 35% of the total, as single matchups against the strongest teams in the league. Why: a title is three knockout weeks against good teams. A model of the regular season alone would undervalue what wins those. This weight is one we've marked for re-testing.
- What a player adds. Method: the slope of your chance of winning with respect to each category (the gradient), times the player's weekly contribution in each. Your open roster spots, and every opponent's, are stood in by the average player of each rank band of what's left. Why: one player is a small change to a team, and a slope prices small changes in the right currency, your chance of winning. Filling open spots with an average pick keeps the answer from jumping when one assumed player changes.
- Giving up a category. Method: a threshold and a cap. A category under a 30% chance is treated as given up, and the model may give up two (three in a one-win league) before it protects the next. Why: the cap was measured, and raised titles. The 30% line is a judgement, not a measurement, and the limits above say so.
- Auction prices. Method: a price ladder that always adds up to the money the managers have left. Its starting point is a price-by-rank curve from past drafts, scaled to your league, blended with Yahoo's prices. From the sales, a Bayesian update learns the league's star premium, its pace and each manager's aggressiveness, under tight priors. Why: prices in a draft must add up to the money in it. That one rule produces late-draft inflation and the pressure of unspent cash without anything being learned. Tight priors mean one odd sale moves little.
- Your top bid. Method: what the player adds, checked by a beam search over market cells (player types × price tiers × positions) that compares the best rosters you could finish with and without the player. The price scenarios are drawn once and shared across every comparison (common random numbers, in a Latin hypercube). The search's say grows as your roster fills. Why: searching over named players overfits to players who'll be sniped; cells don't. Sharing the price draws keeps the difference between two searches from being noise. With little bought, the search is noisy, so it's heard less.
- Who survives to your next pick. Method: each player's pick is a normal around the league's usual order, with a spread of 1 + 0.2 × position, fitted by maximum likelihood to one real league's draft. The picks before your turn are simulated 200 times, with the same draws shared across every candidate, and the spread's scale learns from this draft's reaches under a prior worth 40 picks. Why: shared draws make "with this player" and "without" comparable. The prior stops one early reach from rewriting the model.
- Testing. Method: paired simulation experiments. Both drafters take the same seat in the same league with the same random luck, so only the drafting rule differs. Standard errors are taken across the paired drafts, a gap counts only when it's more than twice its standard error, and the main snake test's sample size and rule were set before it ran. Parts are switched off one at a time, and the same drafts are rescored on projections 15% off and on a league that knows more than the projections. Why: pairing removes most of the luck between the two drafters. The two-error rule and the pre-set sample stop us reading noise as a result.
The model in full
Step 1: a week
Each team's week is modelled from its players' per-game projections: ten starters by the league's roster slots, with the bench and the waiver wire filling in for missed games where they can. A projection's games are read halfway toward 70 (66 as 68, 82 as 76). A projection of 30 games or fewer is a known long absence and is read as itself; the halfway rule fades in between 30 and 50, so 40 is read as about 48 and 50 as 60. Every projection then loses three games, and none is read below 20 unless it already was. Missed games are split into the kind a bench covers and the kind that is simply lost. Each category's weekly total is treated as roughly normal, with a variance that includes who happens to start; the percentages are modelled from makes and attempts, so a high-volume shooter moves them more.
Against each opponent, the chance of winning a category comes from the gap between the two weekly totals and its spread, with ties for the counting categories. In a categories league a week is worth the expected share of the nine won; in a one-win league, the chance of winning five or more, from all nine chances at once. Each opponent is scored separately and the results averaged, never the other way round: against a weak team you're favoured almost everywhere, against a strong one almost nowhere, and averaging the opponents first would lose that. The playoff weeks, single matchups against the strongest teams, get their own weight in the total. Turnovers count half: their chance of being won is pulled halfway toward a coin flip before it's counted.
Step 2: what a player adds
A player's worth to your team is their weekly contribution in each category times how much one more unit of that category raises your chance of winning (the gradient), at your team as it stands. Your open spots are filled with the average player of each band of what's left, the way an average draft would fill them, and the other teams' rosters are filled the same way. With nothing drafted, every team looks the same and the weights are the league's; as your picks fill the roster, they outweigh the fill and the weights bend to them. A category is treated as given up when you name it, or when your chance of winning it falls under 30%. Left to itself, the model may give up two categories, three in a one-win league, before the next one is protected.
Steps 3 and 4 in an auction
Worth becomes dollars on a ladder: the money the managers have left, less a dollar for each open spot, shared out in proportion to worth above the last player a team would keep (the two spots each team streams from the wire go for a dollar). The expected prices always add up to the money left, which is what drives late-draft inflation and the pressure of cash with nowhere to go. The prices start from past drafts (four seasons of a real Yahoo league, two at 14 teams and two at 16), the k-th best player at the k-th highest price, scaled to your league's size and budget, blended with Yahoo's projected and average prices, and they learn this draft's star premium, pace and each manager's aggressiveness from its sales, under tight priors so one strange sale moves little. Your top bid starts from what the player adds to your team. As your roster fills, a search through the rest of the draft, comparing the best rosters you could finish with and without that player, is given a growing share of the say.
Steps 3 and 4 in a snake draft
The pick is the best legal player by what they add to your team, with your next pick in view. Each player's draft position is a normal around the league's usual order (Yahoo's expert ranking where the draft shows it, then average draft position) with a spread of 1 + 0.2 times that position in picks, fitted to a real 14-team league's picks (the best fit was 1.0 + 0.18 times). The picks before your next turn are drawn 200 times and shared across every candidate, so a candidate's worth with the next pick in view is their own plus the best of who survives. The spread's scale learns from how far this draft's managers reach, under a tight prior. At the turn, the two picks are weighed together.
The numbers behind the exhibits, and the parts switched off
| Where | What was measured | Result | Sample | When |
|---|---|---|---|---|
| Exhibit 4 | Same for every team (never re-ranked), extra categories won per 100 against a cheat sheet | −0.7 ± 0.4 | 14-team snake, 336 seasons | 28 Sep 2026 |
| Exhibit 4 | Re-ranked for your team (after every pick), extra categories won per 100 against a cheat sheet | +1.4 ± 0.3 | 14-team snake, 336 seasons | 28 Sep 2026 |
| Exhibit 4 | Pikola (re-ranked, turnovers at half), extra categories won per 100 against a cheat sheet | +2.7 ± 0.4 | 14-team snake, 336 seasons | 28 Sep 2026 |
| Parts switched off | How real drafts spend (past drafts' prices, by rank), what it adds to the categories Pikola wins per 100 | +2.2 ± 0.6 | 16-team auctions, 640 seasons | 28 Sep 2026, earlier engine |
| Parts switched off | Weekly lineups (the ten you start each week), what it adds to the categories Pikola wins per 100 | +1.8 ± 0.6 | 16-team auctions, 640 seasons | 28 Sep 2026, earlier engine |
| Parts switched off | Prices add up (to the money left in the room), what it adds to the categories Pikola wins per 100 | −0.5 ± 0.6 | 16-team auctions, 640 seasons | 28 Sep 2026, earlier engine |
| Parts switched off | Each team scored apart (not one average opponent), what it adds to the categories Pikola wins per 100 | −0.2 ± 0.5 | 16-team auctions, 640 seasons | 28 Sep 2026, earlier engine |
| Parts switched off | Playoff weighting (extra weight on playoff weeks), what it adds to the categories Pikola wins per 100 | −0.2 ± 0.5 | 16-team auctions, 640 seasons | 28 Sep 2026, earlier engine |
| Parts switched off | How real rooms spend, what it adds to titles per 100 seasons over a cheat sheet | +7.3 ± 3.1 | 16-team auctions, 640 seasons | 28 Sep 2026, earlier engine |
| Exhibit 7 | Snake, League by Yahoo's order: titles per 100 seasons over a manager drafting by Yahoo's order (weekly category wins +9.2) | +28.9 ± 2.1 | 16-team snake, 1,024 seasons | recorded 2 Oct 2026 |
| Exhibit 7 | Snake, Yahoo's order, plus needs: titles per 100 seasons over a manager drafting by Yahoo's order (weekly category wins +9.2) | +24.6 ± 1.8 | 16-team snake, 1,024 seasons | recorded 2 Oct 2026 |
| Exhibit 7 | Snake, League counting like Pikola: titles per 100 seasons over a manager drafting by Yahoo's order (weekly category wins +5.1) | +17.8 ± 2.0 | 16-team snake, 1,024 seasons | recorded 2 Oct 2026 |
| Exhibit 7 | Snake, A mix of all three: titles per 100 seasons over a manager drafting by Yahoo's order (weekly category wins +8.0) | +23.6 ± 1.8 | 16-team snake, 1,024 seasons | recorded 2 Oct 2026 |
| Exhibit 7 | Auction, Last year's prices: titles per 100 seasons, Pikola 26.4 against a cheat sheet 9.4 | +17.0 ± 2.0 | 14-team auctions, 1,280 seasons | 2 Oct 2026 |
| Exhibit 7 | Auction, A mix of both: titles per 100 seasons, Pikola 25.7 against a cheat sheet 9.6 | +16.1 ± 1.6 | 14-team auctions, 1,280 seasons | 2 Oct 2026 |
| Exhibit 7 | Auction, Yahoo's numbers: titles per 100 seasons, Pikola 26.6 against a cheat sheet 9.0 | +17.6 ± 2.2 | 14-team auctions, 1,280 seasons | 2 Oct 2026 |
| Snake model | Players expected gone before pick 70: share predicted still there, against the share that was | 7.0% against 7.2% | one real 14-team league, 69 picks | recorded 2 Oct 2026 |
Exhibits 1, 2, 3, 5 and 6 are computed by Pikola's engine for a standard 14-team Yahoo league (PG, SG, G, SF, PF, F, C, three Util and three bench; $200 in the auction) at its default settings, from this season's projections, as each source line says; the auction path of Exhibit 1 describes its two steps rather than computing them. Exhibits 4 and 7 and the parts switched off are the measurements listed above, each from the run named. The auction tests use the engine of 2 October 2026; the part-by-part tests an earlier one.
Pikola is a Chrome extension that advises live in Yahoo fantasy basketball drafts, built on this model. Try it free in a mock draft · How Pikola is tested · More Draft Science