| # | Regime | Tenure | Draft | Trades | Free Agency | Waiver / Minor | Int'l | Ext | Ret | Net value vs. expectation | Dev | Success | FOR |
|---|
| Org | Style | Shops for | Draft value | Trade value (net) | Pitchers improve here (wins/yr) | Defense improves here (runs/yr) | Hitters improve here (wins/yr) | Keeps its talent | IL stints per year | Overslot picks | Buys after a down year | Most similar |
|---|
The unit of account is the decision: a dated transaction attributed to the executive regime in office when it was made. Each decision is scored in WAR captured (production received during the club-control window acquired) and surplus dollars (WAR × market $/WAR, minus salary paid). Departing players are scored symmetrically, so every deal nets to one number.
Draft picks are measured against the historical expected value of their slot; the #1 overall pick carries a far higher bar than pick #38. Prospects moved in trades are valued at the deal date using published prospect-tier surplus curves, so the ledger can score a trade both as it looked at the time and as it turned out; the gap between the two is the luck-versus-skill diagnostic. Free agency is graded against what a typical club gets for the same money in the same contract years, on a deal of the same size: across every priced free-agent deal since 2006, a season of a contract returned about 1.06 wins per win of cost at market $/win in its first year, 0.80 in the second, 0.74 in the third, 0.69 in the fourth and 0.50 from the fifth on (0.87 pooled), and small deals return far more per dollar than big ones (deals under half a win of cost a season about 2x the curve, $15-30M deals about 0.86x), so a deal halfway through is judged against the years it has played and against deals its own size. Posting fees count as year-one money. Extensions are graded against the path they replaced (see below). Every individual move also carries a letter that ranks it among real moves of the same kind and size, so a $1M signing is judged against $1M signings. Before 2009 the transaction feed files a waiver claim or a Rule 5 pick only on the claiming club; the losing club's side is now added from the claim itself, so the channel is two-sided in every era. International expectations count only the signings made on a regime's watch. A channel with too little evidence to grade counts at its expectation, zero. The waiver/minor-league channel is graded against a baseline of zero: pure found money, the cleanest read on pro scouting. Net-of-cost is the universal display rule wherever a contract exists: FA signings and extensions never show raw WAR accumulation, because signing a star who then plays like a star is a pass, not a triumph.
FOR (Front Office Rating) measures decision quality: value added versus expectation across the six channels. Team Success measures on-field results during the tenure: winning percentage, division titles, playoff appearances, playoff series wins, pennants, and World Series. They are deliberately separate: a front office can out-decide the league and still miss Octobers (payroll, luck, timing), or ride inherited stars to banners. The gap between the two grades is often the most interesting number on the page. Grades display as letters (A+ through F) mapped from the underlying 20–80 scouting scale, which stays available on hover.
Every grade in the ledger is denominated in a wins-above-replacement measure built for this project from the live feed, so no number here depends on a licensed source. Offense is standard linear-weights run value converted to wins with park factors computed from each park's home/away run environments (three-year rolling, halved for the road split, back to 1996). Defense uses the SABR Defensive Index where published and Statcast fielding runs plus catcher framing elsewhere; baserunning comes from Statcast. Pre-2014 seasons are bat-only, disclosed wherever they appear. The cache covers 28,460 players, 1998-2026, and is completeness-audited: a self-audit sweep of every player with a qualifying season found 133 career-one-club regulars the transaction-driven build had missed and restored them, moving the accounting validation from r 0.853 to 0.864 against the public reference. The audit now runs after every data refresh.
Every substantial decision is shown through a stack of named lenses rather than one verdict: windows-only (the control years actually exchanged), option-inclusive (adding retention rights and their exercise), and the counterfactual (this regime against its own realistic alternative), with the raw reading beside it. Perspective rule: every lens on a regime's card is signed from that regime's seat. The other side of the same deal grades on the other regime's card, from theirs; the Betts trade reads as Friedman's win on Friedman's card and as Bloom's defensible sale on Boston's, and both are correct because each is measured against its own alternative. Alongside the lenses, every decision and every branch of a transaction tree carries a status: CLOSED means the window is finished and the number is final; OPEN means value is still accruing and the verdict can move. The site never averages the lenses into one number; the disagreement between them is the analysis.
A regime is graded only on its own decisions; inherited players are excluded. Extensions, non-tenders, and passive losses (a free agent walking, a player lost on waivers) are decisions too. Recent regimes carry a small-sample badge rather than being hidden, and the leaderboard defaults to realized value only, with projected-remaining value shown separately. Every rating is clickable down to the individual decisions that produced it, and every methodology version is archived so grades are reproducible. The two Betts decisions on the Dodgers card are scored with real data as a proof of concept; the trade and the extension are separate, chainable decisions, and that one example shows why: on windows alone the deal reads as an overpay, while the full chain reads as one of the great decisions of the era. Same inputs, different lenses, both on the page.
The contract book (v5.7). Free-agent grades and contract pricing rest on a book of guaranteed major-league deals: ESPN's free-agent tables for the 2006-21 winters and, for 2022-25, every guaranteed major-league signing (560 deals) from MLB Trade Rumors' team-by-team offseason reviews, cross-checked against NBC, MLB.com, ESPN, Fox, CBS and AP. Totals include buyouts; accepted qualifying offers count as one-year re-signings; split and minor-league deals are listed on the cards but not priced. Extensions come from MLB Trade Rumors' extension tracker and deal-by-deal reports, 286 deals from 2015 on.
One contract book, one ownership rule, one money unit (v5.8, revised in v5.9). Every guaranteed contract the ledger knows (2,779 free-agent deals from the 1997-98 winter on, 826 extensions from 1998 on) sits in one book. Ownership: a free-agent deal or a re-signing elsewhere owns its seasons, so no earlier decision's window runs into them. An extension owns only the seasons beyond the player's original club control (six years of service, dated from his MLB debut: a debut after mid-April runs control one season longer). The controlled seasons stay with the decision that acquired him, the draft pick, the international signing or the trade, and the extension is graded on what it changed there (its salary against the arbitration path it replaced) and on the free-agent seasons it bought. Before v5.9 an extension cut the acquiring decision short: when Detroit extended Miguel Cabrera three months after trading for him, the trade lost his two controlled seasons and the heist read as a loss. A traded player is an asset: the club that acquires him is credited with his window wherever he plays it (six seasons, or to the end of his club control if later), so a quick flip is graded in the flip, until a new contract, a free-agent filing or a release. Money: salary that moves in a trade or a waiver claim is charged at what that money typically buys: a free-agent contract's season at the typical return for that contract year and deal size (so taking on or shedding a typical contract is neutral in expectation, and dumping the late years of a deal is not a windfall), other salary at the pooled 0.87. Arbitration salaries are charged (v5.9): an arbitration season with no contract costs the estimated award, (0.6 + the better of the prior season and the two-season average) x the class step (1.0, 1.5, 3.1) x the $/win drift, the same model the Desk uses. A season lost to injury still charges the salary. Where a public report gives the cash sent by season, the acquiring club is charged only what it paid (Nolan Arenado to Arizona, January 2026). Positional credit in Ledger WAR is weighted by the games a player actually played at each position each season.
Assets compound across decisions: a fourth-round pick becomes a rental ace becomes a comp pick becomes a new draftee. The Trees view follows these chains generation by generation, marking each branch open or closed, with captured value in solid cards and outflows in dashed ones. Trees are also the honest way to display sell-offs: the Orioles' 2025 deadline reads as a dozen open branches, not a graded verdict.
Future-year value discounts at 10%/yr from the first season of a window (6–14% sensitivity shown): a 2029 prospect win was never worth a 2024 pennant-race win to a contender, and both sides of a deal are discounted alike. An October-December deal starts its window with the next season. From v5.2 to v5.8 trade value was also multiplied by a win-curve leverage weight, L(w) = 0.5 + 1.3·e−(w−89)²/200, at the club where the wins were produced. The round-2 review showed what that does to a zero-sum ledger: the selling club is charged the departing player's wins at the buyer's leverage, so a rebuilding club that trades a star to a contender is billed at up to 1.8x, and value flows from bad clubs to good ones by construction (seats on clubs under 72 wins netted −375 wins with the weight and +33 without). Leverage is real for the club that holds it, but it is not a property of the decision on the other side of the table, so since v5.9 the grades are unweighted. Leverage stays where it belongs, in the forward tools (the Deal Desk prices near-term wins for the club you choose) and in the Scenario Room's context. The Burnes trade now reads −6.4 for Baltimore in the trade book: Ortiz's rebound in Milwaukee outweighs Burnes's season in Baltimore whether or not October is weighted. Every parameter (discount rate, windows, money unit) is versioned and public.
A trade transfers a third asset beyond the control windows: the exclusive right to extend the player. Pricing it at zero produces indefensible verdicts: the windows-only ledger says Boston "won" the Betts trade. Every trade therefore shows three ledgers: windows-only; option-inclusive (realized extension surplus flows to the acquirer once the right is exercised, the headline number); and the seller's counterfactual, charged p(retain) × forfeited extension surplus, where p is a documented, versioned judgment drawn from negotiation reporting. The same machinery scores passive retention failures (letting a star walk is charged p-weighted forfeited surplus with the comp pick credited), so "traded him with team control left" and "let him leave for nothing" become directly comparable. Under this accounting deals stop being zero-sum: in the Betts case both front offices beat their realistic alternatives, which is the honest and more interesting answer.
Raw leakage (wins produced elsewhere by departed free agents) scales with how many stars a club employs, so an earlier pass left it display-only: it graded payroll, not decisions. The channel now conditions on who left. From every free-agency departure league-wide 1998-2023 (2,827 of them with real value behind them), we build the empirical curve of what departures of each caliber go on to do over the next three seasons elsewhere (a 2-win departure goes on to about 1.5; a 10-win departure about 5.8). A filing followed by a season back with the same club is a re-signing, not a departure, and is excluded: counting those had credited clubs that re-sign their own stars with the whole expectation (one card carried +133 wins of it before this was caught). Since v5.9 the expectation also depends on age (departures 36 and older go on to about half of what the caliber curve alone predicts; 29 and under about 1.27x), so letting a 37-year-old walk is no longer scored as skill, and a deadline rental who walks is left out (his walk is part of the trade, whose window already ends at his filing). A regime is graded on the gap between its departures' actual next three years and that expectation, plus a +1.0 win compensation-pick credit for each player who declined its qualifying offer and signed elsewhere; the signing club carries the matching 1.0-win charge on its free-agent ledger. The QO table is complete for the 2012-13 through 2025-26 offseasons (98 signings with a new club, from MLB.com's QO decision history; signings after that June's draft, which forfeited no pick, are left out). Before 2012 the Elias Type A/B compensation system applied; it is not modeled. The curve is centered so the mean gap per departure is zero; per regime the channel averages about +2 wins over a tenure (median +1), and it joins the FOR composite. Letting a six-win player walk is not a mistake; repeatedly letting the ones walk who still had the most left is. Departures after 2023 are right-censored (their next three seasons are not observable yet) and excluded, disclosed.
An extension is graded against the path it replaced (v5.9). In the seasons the club already controlled, the player's production belongs to whoever acquired him; the extension changes only the bill, so it is credited the estimated arbitration salary it replaced (the minimum in pre-arbitration seasons) and charged its own. In the free-agent seasons it buys, it is a free-agent deal: production against the cost at market $/win times the typical free-agent return for that year of free agency and deal size. Each service class (pre-debut, early career, arbitration, veteran) is then centered on the league's average result for that class per win of cost, so a club is graded on whom it locked up and at what price, not on which kind of extension it favors. (The v5.7-5.8 fixed class discounts, 0.35 to 0.90 of market $/win, left early-career extensions returning 1.39x their expectation and veteran ones 0.84x, so an extension letter mostly measured class mix.) Deferral-heavy deals are costed at CBT/present value (Betts $306.7M; Ohtani $460.8M in the FA channel). Extensions and free agency are graded for archive regimes too, from the same book.
Development is a third top-level grade beside FOR and Success, because turning a 16-year-old into a top-100 prospect is neither an acquisition nor an on-field result. Prospects are marked to market on a rank-to-surplus-wins curve using the MLB Pipeline top-100 archive (2011-2026, 1,548 rank-rows), with the holding org derived from minor-league rosters, not the rankings. Entry credit is net of acquisition cost (own draft pick vs. the pick-value curve; own international signing; acquired as prospect) with an inherited screen: a prospect developed to the list under a predecessor earns the new regime nothing at entry. Rank moves accrue to the org holding the player, graduation to MLB is a neutral exit, and falling off without debuting is charged. Credit stays home when the prospect is later traded, which is the point. Grades are league-relative because development is positive-sum (about +2.5 wins of prospect value per club-season league-wide).
Grades say how well a front office buys; the fingerprint says WHAT it buys. For every acquisition a sitting regime has made since 2016 (trades in and major league free agent signings), the ledger looks up the player's public Statcast profile in the season the front office actually evaluated (the one before the deal): contact quality, chase and whiff rates, sprint speed, and walk rate for hitters; fastball velocity, bat-missing, ground-ball rate, chase induced, and walk rate for pitchers. Each skill is z-scored within its season, so a 2016 velocity buy and a 2026 velocity buy are comparable, and the regime's average deviation across all its buys is its fingerprint. The persistent leanings read out in plain language on the Org DNA tab (a desk that pays for velocity, one that buys strike-zone discipline, one that lives with swing-and-miss for power), feed the assistant GM's counterparty counsel, and appear on every Deal Desk counterparty card. Regimes with fewer than eight joined acquisitions on a side read as thin sample rather than being forced into a label, and a lean toward slower, older players is reported honestly as what it is: a taste for veterans. Pointed at a club's internal evaluations instead of public Statcast aggregates, this same construction answers the sharper question: what does each rival pay for, and what do they undervalue that you can sell them.
The lens stack now has an empirical anchor: 224 deadline rental deals classified exactly from the league feed (acquired June-August, declared free agency within five months), with the buyer's receipt measured as rest-of-season Ledger WAR and the price as what the surrendered players actually became for the seller. Findings: a rental win costs about 0.8 realized wins across the market (the nine true star rentals ran nearer 1.2, too few to call); the median deadline rental costs approximately nothing that ever materializes; and more than half of rental buys deliver under 0.3 wins. With a market price in hand, a trade grades on two questions: paid vs. the market at the time, and realized vs. paid. Applied to the canonical case: Boston's return for one expected year of Betts was roughly market rate; the ledger's criticism was never the price, it is the decision to be a seller of that asset at all. Full study: the methodology addendum in the repo; the recent-era ratio is disclosed as right-censored while surrendered prospects mature.
The complement of the rental study: 1,372 deadline deals 2005-2023 (1,370 with a measurable control bucket) where acquired players came with control beyond the season, with control measured from the feed (seasons until the player's first free-agency declaration, capped at six). The price curve by control: one extra year of control trades at about 1.0 wins paid per win received (the deadline premium on controlled players is real and larger than the 0.8 star-rental rate); two extra years at ~0.86; and the three-plus bucket, which is mostly the seller's seat buying prospect packages, at ~0.37 paid per win eventually received, with a 71% bust rate. In other words: sellers buy future wins at about a third of the price buyers pay for present ones, and they pay for that discount in variance. The biggest controlled buys in the archive (Turner, Arrieta, Verlander, Gallen, the Soto package) were all acquired at single-digit prices against double-digit returns. Full tables in data/rental_market.json.
Every graded trade on an active or archive card carries up to three market comps: the nearest deals by eventual value received, same season phase (deadline / offseason), drawn from the 1998-2023 league-wide archive of 5,804 buyer-seat deals. A verdict reads differently when you can see what deals of that size usually cost; the comps put each trade's price against the market that actually existed. One boundary worth knowing: the trade archive's market rates are fitted on deals through 2023, so deadline pricing for 2024-26 extrapolates the fitted curves rather than re-estimating them; the curves move slowly, but a sharp reader should know where the fit ends.
Open deals show, beside their realized grade, a projected rest-of-window and a projected final verdict. The projection is self-built (no licensed projection system is consulted): Marcel-style weighted recent form (5/4/3), regressed 20% to the population mean, walked forward on delta-method aging curves computed from our own 28,000-player cache (hitters peak at 27-28 and decline ~0.3 wins/yr through the 30s; pitchers decline flatter: the textbook curve, recovered independently), with a 10%/yr roster-uncertainty haircut and a zero floor. Prospects without an MLB track record carry their Pipeline-rank expectation from the same curve the deal marks use; unranked traded prospects carry the empirical mean outcome of 1,327 unranked traded pieces (0.41 wins). Backtested: projecting each season from only prior data gives r = 0.62 (2023) and r = 0.67 (2025) against actual next-season WAR, with MAE ~0.8 wins, competitive with public projection systems. The overlay is disclosed and quarantined: grades never move on a projection. The one expectation inside a grade is the one the deal was made on: a prospect not yet in the majors carries his rank-curve value at the time of the deal, fading as his real seasons replace it (the same rule the draft uses for unrealized picks; v5.9).
Projections as rate times playing time (v5.8). A player's base is his WAR per plate appearance (or per inning) over 2024-26, weighted 5/4/3 and regressed toward his role's league rate by sample size (1,200 weighted PA for hitters, 450 innings for starters, 180 for relievers), times his expected 2027 playing time: the same weights over every season since his debut, with a missed season counting as zero, capped at a full season. Missed time is counted once; before, a 30-inning season counted as a full season of 0.3-WAR pitching and was then cut again for durability. A player with no big-league time in either 2025 or 2026 gets no projection and is priced at a one-year minimum.
For signings and extensions the Desk now does the work before you guess: pick a player and it proposes terms. Two components are shown in full. The market-comp offer takes the eight nearest real contracts by blended recent form, age, and position type from a book of 2,049 free-agent deals (2006 to the current winter; extensions are never used as free-agent comps; each comp shown with what it actually returned) and offers their median years at their median AAV inflated to next winter's dollars. The model-fair offer runs years until the player's projected season value thins below ~0.75 wins and pays open-market $/win for the projection, with the class discount on extensions. The headline Ledger offer is a calibrated blend, and the calibration is backtested the only honest way: reconstructing the suggestion for all 1,607 book contracts signed 2010-2025 using only information available that winter (comps strictly earlier, that era's $/win; signings with no MLB track record, such as NPB/KBO postings, are excluded as unknowable in-model and flagged in the tool). The market-comp offer alone is nearly unbiased on ordinary deals but reads star contracts badly light: median-of-comps compresses, and the $100M+ class came in roughly $79M light on average. The current rules close most of that gap with market structure rather than curve-fitting. A pedigree gate rates stardom on a recency-weighted read of several seasons, discounted with age, so a star coming off a down platform year still prices as a star instead of as his worst season. Pitcher wins are paid at a premium over flat $/win, which is how ace auctions actually end. Elite seasons carry a convex premium per win, applied to regression-shrunk projections, because the market pays for scarcity but also prices elite regression risk. Mid-tier regulars earn model years even while their dollars stay comp-driven, and true stars take spread years: extra seasons at the same total, the AAV-flattening structure of real megadeals. One more filter guards the comp median itself: when the subject is an everyday player, comps whose own contract repudiated their platform value (non-tenders, injury collapses) are dropped from the pricing median, because they carry information about risk, not price; depth signings keep them, since down there those outcomes are the market. Backtested on the same walk-forward protocol, the suggested offer misses by 0.57 years, $3.0M AAV, and $11.4M total guarantee on average across 1,607 signings. The biggest real deals run above any offer (on the 22 deals of $200M+ the rules suggestion is about $77M light, the comp median alone far more): sorting by the size of the real contract selects the auctions that broke high, so the fair test is by size of offer, where the live blended offers of $50M or more land at a median 97% of the real deal. One further disclosure: the in-page suggester applies the durability factor (recent playing time) to projections, while the historical backtest cannot, because season-by-season playing-time data was not reconstructed for past winters; the direction is conservative and the divergence is small, but the quoted backtest numbers describe the model without that factor. A separate data honesty note: this re-fit began by refetching primary positions for every player then in the value cache from the live feed, which uncovered roughly 440 backtest contracts whose player type was wrong or missing and quietly polluting comp pools, the sort of bug only found by reading individual misses.
Committed payroll is computed from the ledger's own contract book: guaranteed FA and extension dollars, AAV-spread across seasons 2027-2031 per club. Desk verdicts on signings and extensions show the signing club's committed payroll before and after the proposed deal and its distance to the $244M competitive-balance-tax reference line (the 2026 threshold; 2027+ awaits the next CBA). The disclosure matters: arb, pre-arb, options, and mid-contract trades sit outside the book, so true payrolls run higher; the layer measures the guaranteed core, which is the part a GM is actually committing. The Horizon tab is the same machinery pointed at planning: every club's roster walked forward on the aging curves through 2031, bounded by control (feed-derived service defaults, extended by guaranteed contract years), with per-season totals, top contributors, who falls off when, committed dollars beside projected wins, and ranked prospects on a farm line at rank expectation. Trade verdicts at the Desk show what the proposed deal does to your club's horizon, season by season. Any Desk verdict exports as a one-page deal memo (print or save as PDF) carrying the full verdict, suggested terms, and comps.
Three tightenings. Durability: Desk projections now carry a playing-time factor (0.7 + 0.3 x weighted recent availability: plate appearances against a full season for hitters, innings against a starter's or reliever's workload for pitchers). Validated the only honest way, by projecting 2026 from 2024-25 with and without it across 1,081 players: error improves (MAE 0.86 to 0.85) and bias centers (+0.06 to -0.01), at a small cost in rank correlation (r 0.58 to 0.56), so it is a calibration fix for totals and dollar figures, which is exactly where it is applied. Arb estimates: the payroll layer adds estimated salaries for controlled players outside the contract book: league minimum pre-arb, and arbitration awards from a model calibrated against 48 real 2026 arbitration salaries (award = (0.6 + 1.0 x platform WAR) x a class step of 1.0, 1.5, and 3.1 for the first, second, and later awards, scaled by the $/WAR drift; leave-one-out miss about $1.6M per player, median $1.3M, no bias, where the old 25/40/60-of-market rule missed by $3.3M and ran $3.0M hot). Service time is real MLB service rebuilt from each player's transaction history (options, recalls, injured list, Super Two, the top 22% of two-year players who reach arbitration a year early, at about 2 years 124 days), not a count of seasons with an appearance; six-plus-service players are free agents, not arb cases. The contract book itself also now carries 83 extensions (CBT values where deferrals diverge), which is how Bobby Witt Jr.'s $289M stopped being invisible to Kansas City's payroll. Contract comps still draw from free-agent deals only, because extensions price off-market by construction. The draft room: a fourth Desk mode prices pick packages on the same maturity-adjusted slot curve the draft channel grades with: trade up, trade down, or pick-for-player, each package translated to its nearest single-pick equivalent, with the variance warning stated plainly (a pick is a lottery ticket priced at its mean). MLB only permits trading Competitive Balance round picks, so the mode is framed as the valuation chart a draft room argues from.
The second self-audit. While building the Scenario Room, a decision scenario about the most famous extension in the book came back reading zero. The cause was structural: the player-value universe had been built from transaction-touched players (trades, waivers, signings, draft picks), so a career-one-club star who never hits free agency touches almost no transactions and was invisible to the ledger. A full completeness audit (every player with 150+ PA or 40+ IP in any season 1998-2026 against the cache) found 133 significant missing players, including Jose Altuve, Jose Ramirez, Salvador Perez, Ronald Acuna Jr., Ozzie Albies, Chipper Jones, and Todd Helton. All were rebuilt at Ledger WAR and the scoring chain rerun. Effects: the accounting validation improved from r = 0.853 to r = 0.864 (fuller rosters sum closer to records, which is the metric working), channel grades barely moved (these players were rarely traded, which is why they were missing), Salvador Perez rejoined the 2015 championship core and Altuve the 2017 core, and the Deal Desk pool grew (it now holds 3,130 players). Disclosed with numbers, like the 2009 feed gap before it: a system that grades others audits itself in public.
Process only (v5.7). Famous archive deals are easy to grade with hindsight, so the archive now plays blind: names, club, executive and date stay hidden until you decide, and the card shows only the numbers that were knowable. Your grade is process alone: the price against that day's market, adjusted for the seat (a 74-win club handing a 31-year-old a long deal, or a club out of the race buying a rental, is a process miss even at a fair price). What the deal became is shown afterward as hindsight and never moves the grade. A Winter 2027 pack adds decisions nobody knows the answer to yet: this winter's top free agents offered to the clubs whose depth charts need them most, young extension candidates, and controllable stars on rebuilding clubs, each priced by the Ledger's market model with asking prices set above, at, and below it. Those settle after the 2027 season.
The Scenario Room replays real decision points from the archive with only what was knowable at the time: the record and standing on the morning of the deal (standings API, dated), win-curve leverage, computed roster-need groups, era $/win, and the pieces with their platform seasons and at-the-time prospect ranks (the rank archive starts in 2011; a handful of pre-2011 pieces carry a curated consensus tier, flagged on the card). You accept, reject, or counter; the counterparty answers by deterministic market rules derived from the control-bucket price curves, no dice, and the reveal explains exactly why it took or walked. Grading is the project's thesis operationalized: a process score against a deliberately crude ex-ante expectation (last full season's rate over typical control, prospects at rank expectation) and an outcome score from what the archive says actually happened. Accepting the Chapman rental grades F on outcome and your counter to trim the price grades A on process, in the same room. One 2026 deadline deal plays with its return still open, graded at the trade book's value to date, and one fictional scenario is generated from the current data pack each load. Curated setup lines are color; every number is computed.
The Deal Desk tab is the forward-looking half of the system: propose a hypothetical trade, free-agent signing, or extension and get a verdict assembled from the ledger's own parts. Trades are evaluated with Ledger Projection over each piece's remaining control (players, top-100 prospects by rank, and draft picks off the slot curve), leverage-weighted at your club's current standing, priced against the historical rate for deals of that shape (the 5,800-deal archive's control-bucket curves, with bust rates), matched to the three nearest real comps and their outcomes, and accompanied by counterparty intelligence from the regime cards (their trade grade, tendencies, and your head-to-head record). Signings and extensions grade with the same accounting the FA and extension channels use. Everything is a sandbox on public data: proposals are never recorded, and real card grades never move on a projection. Pointed at internal projections and pref lists instead of public ones, this is a deadline war-room tool; that swap is architectural, not a rebuild.
The offseason simulator runs your club's actual coming winter, staged the way the calendar flows: qualifying offers and arbitration tenders first, then the exclusive re-sign window, then the open market, the depth chart, and a graded close. Every price in it comes from the engines above. Free agents negotiate through a deterministic camp: each player opens at the calibrated Ledger offer times a stable personal factor, counters when your offer is close, and walks after repeated lowballs, so the same winter replays identically. Trades are priced in surplus value (projected production net of salary over the control window, arb years from the calibrated award model), which is why a rebuilding club will not move its franchise cornerstone for a star on a backloaded deal: the names are bigger but the surplus runs the wrong way. Acquired money lands on your payroll against an ownership capacity ceiling, and book money leaves with a traded player. The depth chart fills by projection under true position fits (a shortstop can slide to second or third; a second baseman is not an answer at first base), and an assistant GM reads the same numbers out loud: tender and QO counsel, re-sign or let-walk advice, negotiation reads against the camp's ask, trade counsel with named pieces from your own system that bridge a short offer, and real contract comps tagged with how each aged. The closing grade is projected surplus across your moves, with your projected record set against the five real winters most like yours from 450 club-offseasons since 2009.
The assistant GM's guardrails (v5.8). It never recommends a qualifying offer the CBA forbids (players traded during the season or QO'd before), it names the actual hole a player fills, it only says to go past our number for a player projected at a win or more, it filters free-agent targets by the payroll ownership will approve, and it reads a club's posture from last season and the 2027 roster together (a $260M club coming off 74 wins is retooling, not rebuilding). Deal Desk trades are graded on surplus net of each player's salary, with leverage applied only to near-term big-league value.
How the assistant GM picks fits (v5.7). A recommendation must be a natural fit for the hole (no starting pitcher for a bullpen hole, no catcher at second base), and it is ranked by what the signing adds to the whole depth chart after every player re-slots, not by his own projection. A fielder who would only play designated hitter loses the glove and positional credit his projection carries (about one win for a corner outfielder, more for a shortstop or catcher), so a club with a strong outfield is pointed to an infielder who takes a weak position and moves an incumbent to DH, or to a true DH. Posture filters the list: a rebuilding club is only shown players 29 or younger, a fringe club 32 or younger.
Every World Series winner 1998-2026 carries a computed roster ledger: core contributors (≥ 1.0 Ledger WAR in the title season) traced to the club's last acquisition event (feed 2009+, Retrosheet backfill for 1985-2008 provenance; scoring floors at 1998), with inherited meaning acquired before the sitting regime's true seat date. Across all 29 titles the median champion was 56% built by its sitting regime, in year 5 of the tenure (seasons counted the leaderboard's way: a fall hire's first season is the next one), and, counting only the title regime's own moves, free agency has been a majority channel exactly once (the 2009 Yankees). The strongest through-line is generational: the median title core was 46% self-built in 1998-2009 and 72% in 2010-2026. The handed-down championship has not gone away, though: seven of the seventeen titles since 2010 were won on majority-inherited cores (the 2021 Braves, 2022 Astros, and 2023 Rangers, back to back to back; the Rangers' core reads 85% inherited only on final authority, since 75% of it arrived while Chris Young was already GM under Jon Daniels, the continuity lens on that card), and the 1998 Yankees remain the extreme at 93%.
The ledger is designed to backfill decade by decade; transactions exist back to the 1870s (Retrosheet), so earlier eras can join the same scale as the method extends; today the graded archive starts in 1998 (Schuerholz, Gillick, and their peers from that season on). Channel definitions adapt by era: before the 1965 amateur draft, the "draft" channel reads as amateur and international signings. Era-specific $/WAR baselines keep cross-era comparisons honest.
Before launch, three reviewers were asked to break the site: one on facts, one on the page, one on the method. The fixes are listed in the repository's methodology addendum with the evidence for each. The ones that change grades: extensions now own only the free-agent seasons they buy, so a trade keeps the controlled seasons it acquired (the Cabrera, Rizzo and Lindor trades had been losing them to the extension that followed); leverage is out of the grades (it billed sellers at the buyer's leverage); arbitration salaries are charged; free-agent money is priced by contract year and deal size everywhere; open trades carry the at-the-time expectation of prospects not yet in the majors; club control is dated from the debut and trade windows run to its end; the draft is centered by class; archive regimes are graded on free agency and extensions; interim GMs who traded are listed; retention is age-conditioned; and every letter is on one curve: a rate is net / (seasons + 2), read in units of its own channel's spread across all graded regimes, so A+ is about the top 5% of a channel and F about the bottom 5%.
How much of a grade is skill? We checked whether a regime's decisions agree with themselves: split each regime's decisions into alternating halves and correlate the halves across regimes. Free agency holds up best (split-half r 0.28 across 90 regimes) and extensions somewhat (0.11, 54 regimes); trades (-0.11, 131 regimes), the draft (-0.11, 128) and waivers (-0.01, 118) show no measurable persistence. Read the letters accordingly: they are an accounting of results (what the decisions returned against what a typical club gets from the same move), and in the trade, draft and waiver channels those results are dominated by outcomes no front office controls. A great trade letter says the trades worked, not that the next one will. That is also why the composite shrinks short tenures toward average.
Who watches the Watchmen? A site that grades front offices owes the same scrutiny to its own grades. Two checks, both reproducible from the public feed. Accounting: a WAR-class metric must sum to team record. Across 326 team-seasons (2015-2025, the full-coverage era), roster-summed Ledger WAR vs actual wins above a 48-win replacement baseline: correlation 0.865, R² 0.749, slope 0.824, mean absolute error 5.3 wins. The slope below 1 reflects known gaps (defense coverage thins before 2016; pre-2014 seasons are bat-plus-position only), disclosed rather than patched. Prediction: does a front office's decision record carry information about its FUTURE results beyond the record to date? Pooled across 576 regime-years from 112 regimes (active and archive, 1998-2024, futures bounded inside each regime's own window), with decision value counted only as it had been realized by each season (a trade's later seasons are not allowed to predict the wins they themselves produced), net decision value per season (trades, the draft, waivers, free agency and extensions, each counted in the season its value arrived) has a partial correlation of 0.11 with the next two seasons' W% after controlling for the record to date; the cluster-bootstrap 95% interval is [-0.02, 0.28] (2,000 resamples of whole regimes), and under grouped 5-fold cross-validation (folds split by regime) it adds nothing out of sample (R² 0.186 without it, 0.154 with it). An honest reading: the grades measure what decisions returned against what was expected; they are not, on their own, a forecast of next year's standings. An earlier version of this check reported a partial correlation of 0.27; our own audit found that it let seasons produced after each date leak into the predictor, and this is the corrected result.
This site is the public, free-data build of a decision-support system. Pointed at a club's internal data, the same machinery gives you: deadline pricing (the live rental-market curve says what a win costs this July, and whether an offer is above or below the market that actually clears); extension pricing (the class-market baselines say what the comparable pre-arb or arb extension really costs, with CBT-aware deferral valuation); counterparty preparation (every GM's trade record, channel tendencies, buy-low habits, and development effects, so you know who overvalues what before the call); self-audit (which of your channels is actually producing, process-vs-outcome separation so you stop firing scouts for variance); and regime benchmarking for ownership. Swap Ledger WAR for your internal projections and the prospect-rank archive for your own pref lists and every grade sharpens; nothing in the architecture assumes public data, it was simply built to prove the concept without any.
Built by Sean Zerillo as a working demonstration: a complete decision-level accounting system for front-office performance, from raw transaction feed to graded, clickable, versioned product, with every number computed from free public data. The stack is deliberately simple (Python ingestion and scoring, SQLite, a single-file front end, zero frameworks, zero scraping of licensed sources) so the methodology, not the engineering, is the point. Everything is reproducible from the public repository, including this page. Contact: sean.zerillo@gmail.com.
Data status: the leaderboard, all 27 active cards badged COMPUTED, all 113 archive-regime cards, and all 29 championship ledgers are generated end-to-end by Ledger WAR and the sweep engine from the live MLB Stats API feed (plus the curated Dodgers, Orioles, and Brewers cards, whose decision pages are hand-verified). Archive regimes (1998-2026) grade trade/draft/waiver channels over their actual windows; FA is shown raw and DEV/EXT are not scored for past windows, disclosed on each card. The MLB Stats API transaction feed is empty before 2009, so 1985-2008 transactions come from the Retrosheet archive (mapped to MLBAM ids via the Chadwick Bureau register); the project found and fixed this itself when a 2005 champion's roster came back untraceable. The information used here was obtained free of charge from and is copyrighted by Retrosheet (retrosheet.org). Sample values persist only where labeled: parts of the curated decision pages, the curated Trees showcase, and the Champions tab's contribution estimates.