Bring-your-own-data: your organization licenses the feed; Prism carries it. Credentials are configured server-side in Cloudflare (Settings → Environment variables, e.g. CONNECTOR_TRACKMAN_URL + CONNECTOR_TRACKMAN_KEY) and never touch the browser. Requests route through /api/connector with bearer, custom-header, or query-key auth.
Raw statistics from different levels are not comparable: a .350 wOBA earned against Double-A pitching does not represent .350 wOBA of major-league production. When a cross-level matchup is generated, Prism translates minor-league performance into an MLB-equivalent baseline before the matchup model runs, following the Minor League Equivalency approach established in sabermetric research (originating with Bill James' MLE work and refined by subsequent public studies).
| Level | Hitter factor (wOBA ×) | Pitcher ERA (÷) | Pitcher K/9 (×) |
|---|---|---|---|
| MLB | 1.00 | 1.00 | 1.00 |
| Triple-A | 0.84 | 0.84 | 0.84 |
| Double-A | 0.75 | 0.75 | 0.75 |
| High-A | 0.68 | 0.68 | 0.68 |
| Single-A | 0.62 | 0.62 | 0.62 |
| Rookie / complex | 0.55 | 0.55 | 0.55 |
| Mexican League (LMB) | 0.72 | 0.72 | 0.72 |
| NPB (imported data) | 0.88 | 0.88 | 0.88 |
| KBO (imported data) | 0.78 | 0.78 | 0.78 |
| NPB (imported) | 0.88 | 0.88 | 0.88 |
| KBO (imported) | 0.78 | 0.78 | 0.78 |
Formulas. Hitter: wOBA_MLB-eq = wOBA_level × F. Pitcher: ERA_MLB-eq = ERA_level ÷ F (a minor-league ERA inflates against MLB hitters) and K/9_MLB-eq = K/9_level × F. Reports and matrices carry a gold ⚖ MLE badge whenever any participant's stats were translated.
Known limitations. Factors are league-wide averages: they do not yet adjust for run environment within a level (the Pacific Coast League inflates offense relative to the International League; the Mexican League is an extreme hitting environment at altitude, so its 0.72 factor carries wider uncertainty), park effects, player age relative to level, or small samples. A 21-year-old posting .350 in AA projects better than a 27-year-old doing the same; the current model treats them identically. NPB and KBO factors reflect the public research consensus on posted-player translations (NPB roughly a strong Triple-A-plus environment; KBO somewhat below), applied only to data imported through the Data tab — Prism does not license or redistribute NPB/KBO data itself. These are deliberate simplifications, and the factor table is configurable per client.
Where pitch-level data is unavailable, player wOBA is estimated from season slash lines: wOBA ≈ (OBP × 0.9 + SLG × 0.3) × 0.72, clamped to [.200, .450]. This tracks true wOBA closely for typical profiles and is labeled an estimate wherever shown. When the Statcast engine has event data for a player, the modal displays real wOBA computed from Baseball Savant's per-event woba_value / woba_denom, which supersedes the estimate.
Projected matchup wOBA starts at the league baseline (.318) and applies: batter quality — 70% of the batter's (MLE-translated) distance from league average; platoon handedness — offsets of +.012 (L bat vs RHP), −.010 (R vs R), +.015 (R vs LHP), −.018 (L vs L); pitcher run prevention — .018 wOBA per run of (MLE-translated) ERA distance from 4.20; and strikeout suppression — .004 per K/9 below/above 8.5. Output clamps to [.200, .480]. Situational matrices apply game-state (±3–4%) and inning-third (±2–3%) modifiers. This is a transparent linear model chosen for explainability in a coaching context, not a proprietary black box.
The live splits card pulls per-pitch event data from Baseball Savant for the selected player and season, computing wOBA overall, by opposing handedness, and by inning third from official woba_value/woba_denom fields, plus batted-ball exit velocity and hard-hit rate (≥95 mph). Results are cached at the edge for six hours. Coverage: MLB (and Statcast-equipped AAA parks) only.
Prism operates a two-tier profile system. Auto profiles infer language from birthplace and apply standard per-language coaching vocabulary — clearly badged UNVERIFIED. ★ Verified profiles are hand-built communication intelligence, and each contains seven researched fields:
| Field | What it captures | Source basis |
|---|---|---|
| Native identity | Native-script name, position, number | Official league records |
| Language & register | Formality and directness calibration (formal · peer-level analytical · motivational-direct · encouraging-simple) | Interview demeanor, media conduct, career stage |
| Development lineage | Amateur/pro system and coaching philosophy (e.g., Orix Hosokawa 形 methodology; Kuriyama-era Nippon-Ham) | League history, coaching-tree knowledge |
| Processing style | How instruction lands: spatial/trajectory · tactile/sensation · rhythm/timing · analytical | Native-language interviews and pressers |
| Mechanical vocabulary | The player's own recurring terms (感覚·指先 vs 軌道·形; 割れ; arranque; cadencia) | Verbatim interview language, beat coverage |
| Urgency style | What a <5-second in-game cue should look like for this specific player | Register + processing style synthesis |
| Sensitivity flags | IL/rehab framing rules, media-exposure risk, org-newcomer status | Injury reports, transaction history |
Why register matters more than language: three Spanish-speaking players may correctly receive three different deliveries — cold tactical precision for an analytics-fluent veteran, simple encouragement for a 23-year-old post-surgery, and an explicit prohibition on mechanical overload for a rehabbing power arm. Identical Spanish for all three is the generic-translation failure Prism exists to prevent.
Verification standard: every vocabulary and register claim traces to a native-language source. Profiles are living documents — injuries, role changes, and career maturation trigger re-verification. Profile research and maintenance for additional players is available as a service engagement.
Leagues without public APIs are supported through client-side data import: paste CSV rosters (hitters and pitchers) tagged with a league, and they become fully functional teams in PerformanceFlow — roster tables, lineup builder, and cross-league matchups against MLB opposition with the league's MLE factor applied automatically (NPB ×0.88, KBO ×0.78, other ×0.75). NPB and KBO factors reflect published translation research placing NPB between Triple-A and MLB quality and KBO near Double-A/Triple-A; both are configurable per client. Prism does not source, license, or redistribute this data — clients import datasets they hold rights to (their own scouting records, licensed feeds, or internal databases). Imported data is stored locally in the browser.
Prism ships with a generic secure connector for club-held tracking data. The club configures three deployment variables — a base URL for their data service (a TrackMan export API, Hawk-Eye pipeline, or internal gateway), an authorization token, and a display label — and player modals automatically surface a live card from that feed alongside MLB/Statcast data. The contract is deliberately simple: the club service returns flat JSON for /players/{id}/summary and Prism renders it. This is the BYO-data architecture end to end: the club's licenses, the club's infrastructure, Prism's intelligence layer — no tracking data ever passes through or is stored by Lumira Intelligence.
MLB Stats API (rosters, season statistics, all affiliated levels via sportId), Baseball Savant (Statcast event data), MLB Film Room (video deep links — opens on mlb.com), Anthropic Claude (localization and matchup narrative generation; Sonnet default, Opus selectable). No client data is shared across tenants.