Edit your model by asking
Type what you want in plain English — “Move Shakir to 40% and take it from DJ Moore” — and the shares rebalance while every downstream number updates. It reads your intent and turns it into share moves; it never invents a projection. Each number stays yours: your share × your efficiency.
How this works.A planner converts your words into structured share operations; a separate deterministic executor applies them and recomputes the numbers. The planner is the swappable half: with an API key configured it's a conversational model, otherwise a rule-based parser — and neither can touch anything that produces a number (the model is constrained to emit only share moves, never a projection). Scenario branches (“what if Josh Allen misses 4 games”) run off your existing projection; nothing here fits a curve you didn't author.
49%
29%
14%
over-allocated by 0.5%99.5%
| Player | %Ctch | Catches | Yards | TD |
|---|---|---|---|---|
| Chris Olave | 49.0% | 96 | 1122 | 6.0 |
| Jordyn Tyson | 29.0% | 57 | 681 | 3.0 |
| Devaughn Vele | 14.0% | 28 | 318 | 2.0 |
| Bryce Lance | 6.0% | 12 | 180 | 0.0 |
| Barion Brown | 1.5% | 3 | 31 | 0.0 |
This is a non-persistent what-if on Shawn's 2023 projection. Every number is recomputed from the share you set × Shawn's own per-player efficiency — the tool moves shares, it never invents a projection. Editing shares doesn't balance a drifted pool; the pool sum is preserved and shown.