JerryGM Projections API — v1

A rule is worth points × the player's expected number of such plays per game — a rate, not a probability, because one game can hold two. The rate comes from that player's own long plays (recent games weighted more, last season behind them), shrunk toward the positional rate, so a receiver with a quiet stretch is not priced at zero. breakdown.longPlays shows each rule with its perGame rate and the ppg it added, and breakdown.expectedLongPlayPPG is their sum.

• Any threshold from 20 yards up. min is a yard count, not a bucket: 37 works. Below 20 is a 400 — play-by-play is only kept from there up, so a lower threshold would quietly undercount rather than be wrong loudly.
• max closes a band. {"type":"passCmp","min":40,"max":49,"points":4} plus {"type":"passCmp","min":50,"max":59,"points":5} pays a 55-yard completion 5, once. Without max a rule is open-ended and overlapping rules stack (see below). 999 means open, so a provider's 40-999 is the same rule as a hand-written 40+ and shares its cached board.
• One 42-yard completion feeds two rules: passCmp for the quarterback, rec for the receiver. A 42-yard touchdown feeds the Td types as well, so a league paying for both the play and the score pays twice, as it should.
• Overlapping thresholds stack. Paying 1 at 40+ and 2 at 50+ scores 3 on a 55-yarder. If your league means them exclusively, add max — the response's scoringWarnings will point this out whenever rules overlap.
• Play-by-play covers 1999 on. A player with no long plays in the sample is priced at the positional rate, not zero.

Example — 6-pt pass TD, full PPR, +0.5 TE premium:

CopyGET /projections?season=2026&week=3&scoring={"passYd":0.04,"passTd":6,"passInt":-2,"rushYd":0.1,"rushTd":6,"rec":1,"tePremiumRec":0.5,"recYd":0.1,"recTd":6,"fumbleLost":-2,"twoPt":2,"specialTd":6}

(URL-encode the JSON.) GET /sources returns the preset-equivalent weight objects to start from.

#### Threshold bonuses

Add a bonuses array to the weights object for rules like *+5 for 300 passing yards* or *+3 for 100 rushing/receiving yards*. Any threshold works — min is yours to set:

Copy{"passYd":0.04,"passTd":4,"rushYd":0.1,"recYd":0.1,"rec":1,
 "bonuses":[{"stat":"passYds","min":300,"points":5},
            {"stat":"rushRecYds","min":100,"points":3},
            {"stat":"passTds","min":3,"points":2}]}

stat is any statLine key, plus rushRecYds (rushing + receiving yards combined); GET /sources lists them under bonusStats. Up to 12 per object.

Add max to close a band, for leagues that pay tiers rather than a single threshold. 100-149 pays only in that range, so consecutive tiers each pay once instead of stacking:

Copy"bonuses":[{"stat":"recYds","min":100,"max":149,"points":2},
           {"stat":"recYds","min":150,"max":199,"points":3},
           {"stat":"recYds","min":200,"points":5}]

Add positions to scope a rule, for leagues that tier the same stat by position — a tight end at 75 receiving yards where a receiver needs 100:

Copy"bonuses":[{"stat":"recYds","min":75,"points":2,"positions":["TE"]},
           {"stat":"recYds","min":100,"points":2,"positions":["WR"]}]

Listing every position is the same as omitting positions, and a rule without it applies to everyone.

Overlapping rules stack, and unknown keys are refused. Two rules that both match a game each pay — 100+ worth 2 and 150+ worth 3 pay 5 on a 160-yard game. That is deliberate: plenty of leagues write the upper tier as an extra on top. If yours means exclusive tiers, give the lower rule a max. When rules overlap the response carries a scoringWarnings array saying so; nothing is silently reinterpreted. Any key we do not recognize is a 400 rather than a quiet drop, because a discarded max turns exclusive bands into stacking ones and overpays the big games without telling you.

How it's scored. A bonus is worth points × P(stat ≥ min) per game, and that probability is *not* derivable from the mean stat line — two players projected at 260 yards can have very different 300-yard rates — so it is modeled from each player's game-to-game spread (his own per-game variance, shrunk toward his position's when the sample is thin). Count stats treat "3 or more" as "more than 2.5". Every rule's probability and expected points appear in breakdown.bonuses, and their sum in breakdown.expectedBonusPPG. With bonuses, projectedPPG = your weights applied to statLine plus expectedBonusPPG.

Per-*play* rules (a 40+ yard completion) are a different thing — see Long plays below.

Boards are cached per rule set, so a second caller asking for the same thresholds gets the cached board.

Matching players by name

Name matching is case-, punctuation-, suffix-, and word-order-insensitive: Allen, Josh, josh allen, and st brown amon-ra all resolve. If your roster source has only proprietary ids + first/last names (e.g. balldontlie.io), join by names= on first sync, then cache our ids.gsis per player — it's stable — and use ids=/idType=gsis afterwards. Requested ids/names that don't resolve come back in unmatched (never an error).

Response

Copy{
  "season": 2026, "week": 3, "scoring": "half", "source": "jgm_baseline",
  "count": 612,
  "unmatched": [ { "type": "name", "value": "Fake Person" } ],   // only if any
  "players": [
    {
      "gsisId": "00-0034857", "name": "Josh Allen", "position": "QB", "team": "BUF",
      "projectedPPG": 20.06,
      "seasonTotal": 341.0,                       // projectedPPG × 17
      "statLine": {                                // per-game, points-consistent
        "passAtt": 33.1, "passCmp": 21.9, "passYds": 249.8, "passTds": 1.9,
        "passInts": 0.7, "rushAtt": 6.8, "rushYds": 38.2, "rushTds": 0.4,
        "targets": 0, "receptions": 0, "recYds": 0, "recTds": 0,
        "fumblesLost": 0.1, "twoPt": 0.1, "stTds": 0
      },
      "gamesSampled": { "current": 8, "prior": 16 },
      "ageMultiplier": 1,
      "breakdown": {
        "trailingPPG": 21.4, "positionMean": 15.2, "shrinkWeight": 0.75,
        "cofactors": [                             // jgm_cofactor source only
          { "key": "tdRegression", "delta": -0.8,
            "detail": "Scoring 1.3 more TDs than usage supports — regression expected" }
        ]
      },
      "ids": { "gsis": "00-0034857", "espn": "3918298", "sleeper": "4984",
               "mfl": "13593", "yahoo": "30977", "cbs": "2071510", "sportradar": "…" }
    }
  ],
  "attribution": "JerryGM projections. Derived from openly licensed nflverse data (nflverse.com)."
}

Field reference (what the numbers mean)

Copy{
  "season": 2026, "count": 650,
  "players": [
    { "gsisId": "00-0034857", "name": "Josh Allen", "position": "QB", "team": "BUF",
      "ids": { "gsis": "00-0034857", "espn": "3918298", "sleeper": "4984",
               "mfl": "13593", "yahoo": "30977", "cbs": "2071510", "sportradar": "…" } }
  ],
  "attribution": "JerryGM. Player ids derived from openly licensed nflverse data (nflverse.com)."
}