Methodology
How a Fantasy Football Draft Algorithm Actually Works.
Most draft tools ask you to trust a number. This page explains where the number comes from.
GREEN18 runs a deterministic pipeline: open football data becomes player value, your league settings condition that value, recorded picks move it, and survival odds turn it into take now or wait.
No step in that chain is generated prose. Every step is computed.
Available for iPhone. Built for the 2026 fantasy football draft.
Determinism First.
The model is deterministic. The same inputs always produce the same output.
Replay a draft and you replay the recommendations exactly.
That is what makes the model checkable rather than merely persuasive.
If you can’t reproduce it, you can’t audit it.
The Pipeline
1. Open Football Data.
Player facts come from open football data — nflverse, credited under its CC-BY licence — plus the picks you record yourself. A hard allowlist rejects unapproved sources at build time.
2. Your League Conditions Value.
Scoring rules and quarterback format are applied to the player pool. Superflex does not put a badge on quarterbacks; it genuinely re-ranks them against everyone else.
3. Recorded Picks Move Live Value.
Each pick you record changes the value of everyone still on the board, driven by how fast that position is coming off the board and what the teams picking before your next turn actually need.
4. Survival to Your Next Pick.
The model estimates each remaining player’s chance of still being there when you pick again, given the picks that sit between now and then.
5. Take Now or Wait.
Value and survival combine into the only question that matters on the clock: is this player worth the pick now, or can you come back for them?
The Draft Has a State.
At any moment a draft holds a unique combination of information. These are the five parts of it the model reads — the components of a Draft-State Valuation — and why the board can change after every selection.
League State
Scoring rules, league size, roster requirements, flex structure, Superflex or 2QB requirements, starting positions, bench depth.
Market State
Players selected, players available, positional runs, unexpected falls, unexpected reaches, and the depth remaining at each position.
Team State
Players already drafted, positions filled, positions still needed, roster balance, and the flexibility you keep for later rounds.
User State
Target players, preferred profiles, risk tolerance, positional preferences, strategic tendencies. These live in your personal rank.
Draft Position State
The current pick, the distance until your next pick, and how many opportunities sit between the two.
The Combined State
The board is recomputed against all five together, not against any one of them. That is what makes it stateful rather than a re-sorted list.
Why Does the Board Keep Moving?
Because fantasy value is relative.
A player’s expected production matters. But so do:
- the alternatives available
- the positional scarcity behind him — a rate of decline, not a label a position carries
- your roster
- your scoring
- your next pick
- the managers drafting between your picks
- the players already gone
GREEN18 recomputes the remaining board as those variables change.
The player didn’t change.
Your situation did.
And that changes his value to you.
How GREEN18 Thinks About a Pick.
When your turn approaches, the model is effectively asking several questions at once.
- How good is the player? Start with fundamental player value.
- How does this league score him? Scoring changes the value of production.
- How badly does your roster need him? The same player creates different value for different team constructions.
- What happens if you wait? Replacement value matters — a player is worth the margin by which he beats whoever else would fill that slot.
- How scarce is the position? Remaining depth matters.
- What has the league been doing? The actual draft matters.
- When do you pick again? Pick horizon matters, because distance to your next selection is what converts a valuation into a decision.
- What do you prefer? Your strategy matters, and it stays in your own personalized rankings.
Then the board reorganizes around the combined answer.
None of those eight answers is written prose. Each is a named, computed component.
Three Properties Worth Naming.
Replayable
Every recommendation can be reproduced from the board that produced it.
Separated
Your subjective preferences live in your personal rank only. They never move the objective market values.
Sourced
Player facts come from open football data and your own recorded observations. Nothing else gets in.
Objective and Personal Are Not the Same Column.
You will always like some players more than the numbers do. That is fine.
GREEN18 keeps those opinions in your own personal rank.
They never leak into the objective values the model computes.
So you can see both at once: what the board says, and what you think.
A tool that quietly folds your bias into its “objective” number is telling you what you already believe.
What the Model Does Not Do.
It does not forecast the NFL season.
It does not predict injuries.
It does not guarantee an outcome.
It does not generate explanations as text — the components are named and shown, not written.
It does not make the pick. See automated draft tools for exactly where that line is drawn.
Why This Beats a Static Ranking.
A preseason ranking answers one question, once, before any information exists.
This pipeline answers it again after every pick.
That is the difference between a ranking and a live board — and why ADP is the starting line, not the answer.
Recomputation after a pick is effectively instant, so the board is never behind your draft.
Rankings Are Inputs. Draft Decisions Are Outputs.
GREEN18 treats drafting as a continuously changing allocation problem rather than a static player-ranking problem.
Each selection alters the remaining player pool, positional replacement options, scarcity, roster construction, and future opportunity.
The model combines those changing conditions with your league settings and your draft preferences to reassess the situational value of the players still available.
The objective is not to produce one universal ranking.
The objective is to produce a useful ranking for this manager, in this league, at this pick, given this draft.
The best available player is a moving target.
See the Model Run.
The methodology is only useful in a draft.
Download GREEN18 for iPhone and watch the board recompute pick by pick.
Available for iPhone. Player data from open football sources; nflverse credited under CC-BY.
Questions
How does a fantasy football draft algorithm work?
A draft model turns player data into a value for each available player, adjusts that value for the specific league’s scoring and roster rules, updates it as picks are recorded, and then estimates whether each player will still be available at the user’s next pick. The output is a take-now-or-wait decision rather than a fixed ranking.
What data does the GREEN18 model use?
Player facts come from open football data — nflverse, credited under its CC-BY licence — combined with the draft picks the user records during their own draft. A hard allowlist rejects any source that has not been approved, and it is enforced when the app is built.
Is the GREEN18 model deterministic?
Yes. The same league settings and the same sequence of recorded picks always produce the same recommendation. Recommendations are replayable, which means any change in the board can be traced to a change in the inputs rather than to randomness in the model.
Do a user’s own player preferences change the objective values?
No. Subjective preferences are confined to the user’s personal rank. They are displayed alongside the computed values but never alter them, so the objective view of the market stays independent of what the user hopes is true.
What is a stateful draft model?
A stateful draft model evaluates players against the current condition of the draft rather than against a fixed list. GREEN18 reads five kinds of state: league settings, the market of selected and available players, the roster drafted so far, the user’s own preferences, and the distance until the user’s next pick. The board is recomputed against that combined state, which is why it can change after every selection.
Why does a player’s value change when someone else is drafted?
Fantasy value is relative to the board around it. When a player is removed, the alternatives at that position change, the depth remaining changes, and the odds that a comparable player survives until the next pick change. The drafted player’s expected production did not change; the situation the decision is made in did.
How do league settings change player value?
Scoring rules and quarterback format are applied before the board is built, so they re-rank players rather than relabel them. A Superflex league, for example, changes how quarterbacks compare against every other position, not just how they are grouped.