Data Sources

what feeds SHELL, what plugs in next

Connected

Feeding SHELL now · 6

CFBD api connected last sync 1h ago

collegefootballdata.com REST, key set (CFBD_API_KEY). Responses cached in data/cache as md5-named JSON, 12h TTL, stale copy served when the quota or request fails. 631 cached responses on disk.

tendencies opponent dashboards advance reports recruiting tables schedules

ESPN roster api connected last sync 28m ago

Public team-roster endpoint, no key, independent of the CFBD quota. Cached as espn_-prefixed JSON in data/cache, 24h TTL, stale fallback. Maryland team id 120 verified.

live Maryland roster self scout recruiting needs

cfbfastR data file connected last sync 9d ago

sportsdataverse/cfbfastR-data on GitHub: one ~60 MB CSV per season, one row per offensive play, player-attributed. Downloaded once to data/cache (player_stats_2025.csv); per-team aggregates cached as JSON so repeat reads are instant. No key, no quota.

stats leaders player profile stats

umterps.com bios scrape connected last sync 8d ago

Official roster + per-player bio pages scraped by scripts/fetch-umterps-bios.ts into src/playerbios.generated.ts (never hand-edited). Headshots downloaded and self-hosted at /headshots. Re-run the script to refresh.

player bios headshots vitals

Research layer research connected last sync as of 2026-06-17

data/external/*.json — 29 files from research/API sweeps, each in the {asOf, source, data} envelope (src/external.ts). A missing or malformed file degrades to an empty state on the page; nothing is invented to fill a gap.

home wire opponents intel recruiting board seeds portal context

Anthropic API ai connected last sync never

Key set (ANTHROPIC_API_KEY). Model claude-sonnet-4-6 (src/aiconfig.ts; SHELL_MODEL in .env overrides). Synthesis only — every surface it touches has a deterministic fallback, so pages render with or without it.

coach's read report prose script structuring

Integration slots

Where paid data plugs in · 3

planned PFF file

Slot defined here only. No PFF data in the building; nothing on any page claims a grade.

Season + weekly grade exports land as data/external/pff-grades-<season>.json in the standard envelope {asOf, source, data}. data: one row per player — {player, team, position, facetGrades, snaps} — with facetGrades 0-100 by facet (offense, defense, passRush, coverage, runBlock, ...). Drop the file; this registry picks it up on the next probe, no code change.

data/external/pff-grades-2026.json
{ "asOf": "YYYY-MM-DD",
  "source": "PFF export",
  "data": [
    { "player": "Last, First", "team": "Maryland", "position": "ED", "snaps": 0,
      "facetGrades": { "overall": 0, "pass": 0, "passRush": 0,
                       "coverage": 0, "runBlock": 0, "runDefense": 0 } }
  ] }

powers when live

player grade strip depth chip grade tint portal evaluation opponent personnel (report §V)

planned Telemetry file

Slot defined here only. Report sections IX (Coverage & Pressure) and X (Personnel & Formation) ship today as structured, clearly-labeled blank tables (src/reportdata.ts buildChartingTables) — CFBD play-by-play has no coverage shells, blitz, pressure, or personnel, so those cells stay blank until film breakdown or this feed fills them.

Per-play charting/tracking exports land as data/external/telemetry-charting-<opponent>-<season>.json in the standard envelope {asOf, source, data}. data: one row per play — personnel, formation, motion, coverage shell, pressure/rushers. Rows drop straight into the report's charting tables and replace manual self-scout entry.

data/external/telemetry-charting-<opponent>-2026.json
{ "asOf": "YYYY-MM-DD",
  "source": "Telemetry export",
  "data": [
    { "game": "2026-09-05 vs ...", "play": 1, "down": 1, "distance": 10,
      "personnel": "11", "formation": "3x1", "motion": true,
      "coverageShell": "C3", "pressure": false, "rushers": 4 }
  ] }

powers when live

report §IX coverage & pressure report §X personnel & formation report §XI charting layer self-scout charting practice-script opponent looks

planned Zelus file

Slot defined here only. /spend compares NIL to NFL allocations and /gameday runs its own win-prob model today; this feed adds player value and projection priors on top.

Projection/valuation exports land as data/external/zelus-projections-<season>.json in the standard envelope {asOf, source, data}. data: one row per player — WAR-like value, dollar valuation, lineup/win-prob priors. Joins the roster by player name + position.

data/external/zelus-projections-2026.json
{ "asOf": "YYYY-MM-DD",
  "source": "Zelus export",
  "data": [
    { "player": "Last, First", "team": "Maryland", "position": "QB",
      "war": 0.0, "valueUsd": 0, "winProbPrior": 0.5 }
  ] }

powers when live

roster valuation vs NIL portal target value ranking win-prob priors

Drop a normalized export into data/external/ following the envelope {asOf, source, data} and it appears here. Research layer as of 2026-06-17.