How the Model Works

K Watch projects a full line — strikeouts, innings, walks, hits, earned runs and home runs allowed — for every probable starter on the slate. The skeleton is the same as the batter model: a Marcel-style weighted average of the last three seasons (5/4/3 weights, computed per batter faced), the deliberately simple baseline Tom Tango proposed that serious projection systems get measured against.

Each per-BF rate is then regressed toward league average in proportion to how quickly that stat becomes reliable — strikeout rate stabilizes fast, home run rate very slowly — following the stabilization-point research popularized by Russell Carleton. A recent call-up gets pulled hard toward the league line; a three-year workhorse barely moves.

Context comes in as league-relative multipliers: how much the opposing lineup strikes out, walks, and hits (team K%, BB% and wOBA) against the starter's throwing hand — kept neutral when the split sample is under 150 PA — plus the ballpark's HR and contact factors. Expected workload blends the last five starts with season batters-faced per start, capped for openers and call-ups. Earned runs assume roughly 92% of runs allowed are earned, and innings convert from outs at ~4.25 batters per inning of work.

The headline number is the closed-form product — K rate × expected batters faced — and the K/IP ranges come from 1,000 seeded simulations of the outing (25th/50th/75th percentiles). Every row's expandable trace shows this exact arithmetic, and it reconciles to the displayed number by construction. Like every board on the site, lines freeze at first pitch, are graded against the official MLB Stats API box score the next morning, and the misses stay on the record — K MAE, IP MAE and the ±1K hit rate are on the Accuracy page.

The Trade Ledger grades real MLB trades in expected surplus value — wins and dollars, never letter grades. Each grade is frozen the day the trade is announced, then re-graded on a public schedule as the players' careers actually play out. Every number ships with its full math trace: the steps, the constants, and the sources.

The valuation chain, for MLB players:

1. Project the player's remaining production from current-season rates (a deliberately simple Marcel-style projection), convert it to runs — basic Bill James Runs Created for hitters, ERA-based runs prevented for pitchers — measured against FanGraphs' published replacement level.
2. Convert runs to wins at the canonical ~10 runs per marginal win, and wins to dollars at the open-market price of a win (FanGraphs' 2026 estimate: $11.23M per WAR).
3. Subtract the salary the receiving team actually takes on (cited to a public payroll source on every trade), then apply a win-now leverage multiplier (0.9×–1.2× by contention tier, an internal heuristic, disclosed on the trace) to rest-of-season dollars only.
4. Value future control years the same way with an 8%/year decay, without the leverage multiplier.

Prospects are valued from Craig Edwards' FanGraphs research putting dollar values on Future Value grades — including the documented haircut for pitching prospects, who bust more often. That research is denominated in 2018 dollars; we say so on every prospect trace rather than silently inflating it.

Uncertainty bands come from the outcome spread behind each method — prospect bust/star rates, projection error on veterans — and each side's band is the sum of its assets' bands. Re-grades count realized production at its flat market price (no leverage on sunk wins) plus a fresh valuation of what remains.

Minor leaguers in the Trade Analyzer (the fantasy-style 0–100 tool, distinct from the Ledger's dollar grades) are priced with classic minor-league equivalencies: current-season AAA or AA stats are discounted to MLB terms — AAA ×0.80, AA ×0.70 on rate production, the Bill James MLE idea carried forward by Clay Davenport and Dan Szymborski — then scored through the exact same 162-game-pace math as a partial MLB season. Every translated value is marked ~ and labeled an estimate. Below AA we don't price at all: tiny samples against teenage competition would be false precision. The moment a player has real current-season MLB stats, those take over and the translation disappears.

See it applied: the Trade Ledger — every trade, every grade, and the date it gets re-graded next.

Career pages cover roughly 29,000 players across MLB and every affiliated minor-league level (Triple-A through Rookie ball). Two kinds of data live side by side, and we hold them to different standards: automated stats, rebuilt nightly from primary sources, and curated facts (prospect ranks, contracts), entered by hand with a citation on every row.

Automated, from primary sources:

• Game logs, rosters, schedules and transactions come from the MLB Stats API — the same feed MLB.com and MiLB.com render. Box scores are ingested per game each night for all six levels, then summed into per-season, per-level lines. Stats displayed on career pages are served only from those pre-computed lines, so a sync problem can slow updates but never corrupt what you see mid-rebuild.
• Player identity — who someone is across id systems — is anchored by the Chadwick Bureau register, an open-source baseball id crosswalk, refreshed monthly. It links MLBAM ids to FanGraphs and Baseball-Reference ids where a confident match exists (matches are near-complete for MLB players; many minor leaguers have no cross-ids yet, and we leave those blank rather than guess).
• A verification harness compares our season lines field-by-field against MLB's official totals for a fixed panel of players — including a mid-season three-level promotion arc — and must pass exactly before we trust a backfill.

Curated, with citations:

• Prospect ranks and contract terms are not licensed feeds — each row is entered manually and must carry a public citation URL (the outlet that published the rank, the reporter who broke the deal). Rankings are always presented as the cited outlet's opinion, never as our own scouting.
• Corrections never overwrite history: snapshots are append-only and contracts are supersede-only, so an updated figure sits alongside the record of what we previously showed — the same discipline the Trade Ledger uses for re-grades. Fields no outlet has published (like scouting grades) stay blank; we don't fabricate to fill a column.

Team control is an estimate:

MLB does not publish service time in its public API, so the contract-control years shown on career pages are a documented heuristic estimated from debut date: a player debuting on or before April 15 accrues a full first service year (club control through debut year + 5); any later debut adds a partial year (control through debut year + 6). That deliberately ignores option yo-yos, Super Two arbitration, and extensions — when a specific player's real number matters (Trade Ledger grades, for example), it's set manually from a cited report and the manual value is never overwritten by the heuristic. Estimates are labeled as such wherever they appear.

Stats © MLB Advanced Media, L.P. — this site is not affiliated with or endorsed by MLB. Identity data: Chadwick Bureau (open-source). Prospect ranks and contract figures: credited to their cited outlets on each row.