The War Room is a live draft co-pilot for fantasy football. Point it at any Sleeper or ESPN league or mock draft (or run a manual draft for any other platform) and it prices every player for your league's exact rules, tracks the draft in real time, simulates who'll still be available at your picks, grades every pick as it happens, and — if you want the heavy machinery — solves a full optimization model for your remaining picks and simulates the season it produces.
One box, any league. Paste your league link or ID into Your League — the app auto-detects the platform and reveals only what that source needs:
Once a league loads, set My Team — that unlocks pick markers, the on-the-clock banner, availability odds, Game Plan, and My Squad. Live drafts (Sleeper and ESPN) refresh themselves while the tab is open; a ↻ refresh link appears if you want to force it.
One row per projected player, priced for your league. Click any column header to sort (missing values always sink to the bottom). Filtering to a single position adds tier break lines. Row buttons: ★ mark as a target (tracked in Game Plan) · ✕ mark drafted (paper mode: assigns to the team on the clock; ↩ undoes) · 🔎 AI scouting report.
| Column | What it means |
|---|---|
| Value | The headline number — blended, upside-priced value (see Methodology). Sort by this early; it's what the optimizer maximizes too. |
| VOR | Plain value-over-replacement from projections alone. The gap between Value and VOR is the priced-in consensus + upside adjustment — shown side by side on purpose. |
| Upside+ | The volatility premium: how much extra a player's wide range of outcomes is worth. Sort by this in late rounds — it surfaces lottery tickets (rookies, injury-return, contested roles). |
| Ceil | 80th-percentile weekly points (floor is in the tooltip). |
| Wk Pts | Raw projected weekly points under your scoring — deliberately unblended so you can see the source signal. |
| Tier | Natural positional tiers found by gap detection — the drop between T1 and T2 matters more than any single rank. |
| Depth | Team depth chart slot (RB1, WR3…). Non-starters flagged red. |
| Sleeper Bd | The ordinal rank everyone else in a Sleeper room sees — read the room's logic against yours. |
| ADP / ECR | Market price (Fantasy Football Calculator, 2QB or 1QB to match your format) and Expert Consensus Rank (also format-matched). |
| Trend 24h | Net Sleeper adds−drops in the last 24 hours (green/red; split in tooltip). Breaking-news radar. |
| There @ Next | Monte Carlo probability the player survives to your next pick (appears once My Team is set). |
| 📊 stat view | Toggle in the filter row: swaps market columns for projected season stat lines (pass/rush/rec yards, TDs, receptions) plus Vegas lines — the books' per-game expectations for the next slate (Vg Pass/G, Vg TD% etc.), median across sportsbooks. |
Pick grades (A+–F on each pick) grade the decision, not the player: how many standard deviations of market value you gained or lost versus his ADP. Taking a great player 14 picks early is an F (you could've had him later); an at-market pick is a B in round 1 and round 14; letting a stud fall to you is an A+. Small upside bonus for smart flyers; process caps for early kickers/defenses and below-replacement players at premium picks. Hover any grade for its reasoning.
Team grades (the big letter per roster) grade the groceries: projected weekly points of each team's best legal lineup under your exact slots (flex and superflex assignments included), plus a bench-depth bonus, z-scored against the league. The two can disagree — a team of overpays can still be an A roster. Pick grades = how well you shopped; team grade = how good the food is.
Paste an Anthropic API key into the Front Office (it stays in your browser only — never uploaded, never in cloud saves) and every 🔎 becomes a live AI scouting report with web search: a headline role label ("Low-end WR2 with a path to WR1"), latest news, health, role, ceiling/floor per pundits, and a DRAFT/FAIR/FADE verdict at his ADP for your format. ⚡ Fast mode answers in seconds; 🔬 Deep mode searches more (~30s). Close the modal and keep drafting — a toast pops when the report lands.
Everything works logged out (saved to your device). Create an account (email + password) and your leagues sync across devices: each saved league keeps its ID, your team, custom ranking uploads, targets, and paper-draft progress. The Film Room tab lets you upload any custom ranking (CSV or paste) — it's validated against Sleeper's player list with did-you-mean suggestions, then becomes a sortable board column.
Base projections are converted to weekly fantasy points using your league's actual scoring settings, stat by stat — so half-PPR, superflex, TE-premium, juiced kicker scoring, all of it is priced in from the first step. Nothing assumes a "standard" league.
A player's raw points overstate his worth — what matters is his edge over the best freely-available alternative. We find replacement level by greedily filling every starting slot in the entire league (all teams, flexes and superflex included) with the best players, then taking the best leftover at each position. In a 10-team superflex, ~20 QBs start, so QB replacement is QB21 — which is why elite QBs are worth ~50% more there than in 1QB. VOR = weekly points − replacement.
Any single projection source produces some rankings no expert shares. The Value column blends 70% our league-scored projection with 30% consensus: we take each player's Expert Consensus Rank (format-matched), find where that rank lands on our own projection curve at his position, and blend the two point estimates. Experts fix the ordering; your league's scoring keeps the magnitudes. A small capped bonus rewards players experts rank well ahead of their market price.
Late-round picks aren't investments you hold — you cut busts and keep booms. Mathematically that truncates the downside, so a volatile player near replacement level is worth more than a steady one with the same average (this is the entire logic of sleepers). We estimate each player's volatility from expert disagreement (normalized against players of similar rank, plus a rookie bump — experts herd on players with no NFL sample) and price the option value exactly. The premium self-attenuates for studs: a locked-in RB1's Value ≈ his VOR, while lottery tickets float. That's the Upside+ column.
Every survival probability comes from the same model: a player's draft position is treated as a bell curve around his platform board rank (rooms follow the board in front of them; market ADP is the fallback) with the market's own measured uncertainty. Players who outlive their ADP get a hazard adjustment — someone still on the board is expected to go soon, not never. The odds board runs full Monte Carlo drafts with roster-need-aware opponents; the optimizer uses the analytic form.
The Optimal Plan solves a mixed-integer program: choose one player per remaining pick and an assignment of players to starting slots, maximizing projected starting-lineup points (flex-correct — at flex slots all positions compete on raw points) plus position-weighted bench depth and bench upside, minus three honest costs: an availability risk charge on future picks (targeting a coin-flip player later in the plan is charged; your next pick is exempt — you'll know who's there when you're on the clock), an opportunity cost for spending premium picks on plentiful players (this is what pushes K/DEF to the end without any hard rule), and a bye-stacking penalty. Roster-shape caps (max backup QBs/TEs etc.) are derived from your league's own roster rules, never hardcoded. Pinned ★ targets are forced in when plausible — with a clear warning when the odds say they won't make it.
Any roster becomes a season: each week we set the best legal lineup from non-bye players (like a real manager — on expectations, not hindsight), then simulate 2,000 seasons with every starter's weekly variance. Opponents are modeled by snake-drafting the same player pool off the platform board — a realistic room, not perfect robots. Output: season points with a p10–p90 band, projected record with a luck band (the same roster genuinely finishes anywhere in a ~5-win range on schedule variance alone — we show it instead of hiding it), and playoff odds under your league's playoff settings.
| Source | What | Refresh |
|---|---|---|
| Sleeper API | Leagues, drafts, live picks, players, depth charts, projections, platform board ranks | Live; picks poll ~12s during drafts; data file rebuilt daily |
| Fantasy Football Calculator | ADP with measured uncertainty — 2QB and 1QB formats | Daily |
| FantasyPros | Expert Consensus Ranks + expert disagreement — superflex and 1QB runs | Daily |
| Sleeper trending | 24h adds/drops | Live at page load |
| Sportsbooks (via The Odds API) | Player prop lines for the next slate: pass/rush/rec yards, receptions, TD probability (median across books) | Weekly (Tuesdays) |
Why does the Value column disagree with Sleeper's board? Because it's priced for your league — superflex QB scarcity, your PPR level, your roster slots — then consensus-blended. The Sleeper Bd column shows you the room's view side by side.
Why did my pick get a C when the player is great? Pick grades measure price, not talent. Grabbing a stud 15 picks before you needed to costs you a whole other stud. Hover the grade — it shows the math.
Why is a player's "There @ Next" 100%? Nobody in simulated rooms takes him before your pick — usually deep bench/K/DEF types. The interesting numbers are 20–80%: those are your actual sweats.
The optimizer recommends someone I hate. Star (★) the players you want and re-run — pins force them in when plausible. Or use What-If to see exactly what the roster costs you either way. The model is a co-pilot, not the captain.
Mock drafts? Paste the draft ID from the mock lobby URL. Everything works, including live picks and report cards.
Is my Anthropic key safe? It lives in your browser's local storage only, is sent per-request through the site's own proxy, and is never stored server-side or in cloud saves.
Built by Mike with Claude Code · The War Room · 2026. Not gambling advice, obviously.