Most fantasy AI answers from a generic player database and guesses at the rest. The GWTTKB Coach connects to your real league, scouts the people in it, and refuses to state a number it wasn't given. This page is the honest inventory — what it does, how, and where it stops.
Connect Sleeper, Yahoo or ESPN and the Coach reads your live roster, scoring settings, roster slots and league size — then adjusts positional value to that format. Superflex pushes quarterbacks up. Full PPR lifts pass-catching backs. TE premium lifts target-earning tight ends. A shallower league moves replacement level.
This matters because most advice is written for 12-team PPR and quietly assumes it. If your league isn't that, the ranking you're reading is for a different game.
This is the part nothing else does. In a Sleeper league, the Coach pulls each of your leaguemates' other dynasty leagues — their rosters there, and roughly their last twelve trades across a full season — plus every draft they've entered this year, real, mock and best ball, with their actual picks by round and slot.
So when someone offers you a trade, the read isn't a value chart. It's what that manager has actually done: which positions he reaches for, which he waits on, whether he buys age or sells it, and what he's paid for this kind of player before.
Only Sleeper exposes a manager's other leagues. In Yahoo and ESPN leagues there is no cross-league scouting, and the Coach is instructed to say that plainly rather than invent a tendency. It still reads everything inside the connected league itself.
Every number the Coach states — a projection, an ADP, a target share, an EPA figure, a manager's history — has to come from data placed in front of it that turn. The constraints are hard-coded, not suggestions:
An AI that confidently invents your leaguemate's roster is worse than no AI. Most of the engineering here went into the refusals.
When you name a player, the Coach is handed his card. Not a summary — the numbers.
| Layer | What's in it |
|---|---|
| Production | 2025 counting stats: carries, targets, receptions, rushing and receiving yards, touchdowns, PPR points |
| Role | Snap share, target share, air-yards share, WOPR, red-zone and goal-line carry share, third-down usage |
| Efficiency | EPA per play, yards per target, separation, cushion, YAC over expected, broken tackles, drop rate |
| Coverage | Man versus zone splits — EPA per target and yards per target against each |
| Projection | Floor, median and ceiling with boom and bust rates |
| Market | Consensus ADP plus the gap against ESPN, Yahoo, Sleeper and CBS separately |
| Draft capital | Round, pick, team, college and combine — 40, weight, vertical, broad, cone, shuttle |
It leads with what matters for the position: carries and goal-line share for a running back, targets and air-yards share for a receiver. It won't open a running back read with target share.
Every projection comes with a calibrated floor (10th percentile), median (50th) and ceiling (90th), drawn from a backtested residual distribution, plus boom and bust rates. The Coach is explicitly instructed never to give a locked number — "he projects around 13.6, middle of his range, and the profile can beat or miss it" rather than "he will score 13.6."
A projection stated as a fact is a projection you can't plan around. The range is the useful part.
During a Sleeper draft the Coach tracks the board pick by pick and answers the only question that matters on the clock: can I wait on this guy? Availability comes from an ADP-anchored Monte Carlo that samples each player's own standard deviation — some players go within three picks of their ADP every time, others swing fifteen — and returns whether he's live now, should still be there at your next pick, or is usually gone.
It also knows the board it's drafting from: GWTTKB's own 2026 rankings, with the value gap against consensus and against each platform separately. On the clock, answers are capped short. The draft timer is running.
| Build | What it's for |
|---|---|
| Redraft | 2026 season-long. Draft prep, start/sit, waivers, live draft decisions. Refuses dynasty questions rather than answering them badly. |
| Dynasty | Multi-year. Adds six years of daily market values (533 players, 2020–present, 21,650 monthly data points), so it reads a player's real trajectory — when he peaked, by how much, what followed. |
| College | Yahoo college fantasy. Recruiting pedigree, portal history and usage across Power Four programs. |