An auditable NFL player comparison

The question and the recorded tool call

Question: compare Josh Allen and Patrick Mahomes in the 2024 regular season, using passing yards, passing touchdowns, and interceptions. Resolve the players by team and state the season type before drawing a conclusion.

The public tool-reference snapshot includes both regular-season and postseason rows. The example below keeps only passing rows for the requested year with season_type=REG. It does not sum seasons, mix rushing into passing, or treat a missing row as zero.

{
  "tool": "nfl_compare",
  "snapshotGeneratedAt": "2026-09-25T03:34:52Z",
  "provenance": "Saved public tool-reference example; not a new live query or independent accuracy benchmark.",
  "arguments": {
    "entity1": "Josh Allen",
    "entity2": "Patrick Mahomes",
    "season": 2024,
    "entity1_team": "BUF",
    "entity2_team": "KC"
  },
  "selection": "Only passing rows matching the requested season and season_type=REG.",
  "results": [
    {
      "player": "Josh Allen",
      "season": 2024,
      "season_type": "REG",
      "yards": 3731,
      "touchdowns": 28,
      "interceptions": 6
    },
    {
      "player": "Patrick Mahomes",
      "season": 2024,
      "season_type": "REG",
      "yards": 3928,
      "touchdowns": 26,
      "interceptions": 11
    }
  ]
}

• Full saved tool response and schema

Reproduce the selection

The public JSON artifact contains the unfiltered saved response as well as the selected results, arguments, and source timestamp. Run the JavaScript below in a modern browser console or Node.js with module input enabled to reproduce the selection without access to our source repository. No model call, account, API key, or live database is required.

To run a new comparison rather than replay the saved example, connect your authenticated client, pass the recorded arguments to nfl_compare, and inspect its returned season-type rows. Live results can differ after dataset corrections or tool changes.

const response = await fetch("https://nflmcp.com/docs/research-methodology/evidence.json");
if (!response.ok) throw new Error("Saved evidence is unavailable");
const saved = await response.json();
const results = [saved.arguments.entity1, saved.arguments.entity2].map((player) => {
  const rows = saved.savedResponse[player].passing.filter(
    (row) => row.season === saved.arguments.season && row.season_type === "REG"
  );
  if (rows.length !== 1) throw new Error("Expected one regular-season row");
  const { season, season_type, yards, touchdowns, interceptions } = rows[0];
  return { player, season, season_type, yards, touchdowns, interceptions };
});
console.log(JSON.stringify(results, null, 2));

• Download the saved evidence JSON
• How to query NFL statistics
• Connect your assistant

What this example does not prove

This demonstrates an auditable selection from a saved product example. We have not independently reconciled those saved figures to official gamebooks here. Reproducing the selection is not the same as verifying the underlying statistics.

We have not run a controlled connected-versus-unconnected model experiment for this report. There is no measured hallucination-reduction percentage, forecast accuracy, betting edge, or fantasy win-rate claim. A fair accuracy study would need independent reference answers, fixed questions and client versions, repeated runs, and published failures as well as successes.

NFL MCP publishes and maintains this example about its own product. The saved source timestamp is shown in the evidence block; the page update date describes editorial changes, not a new data refresh. NFL MCP is independent of the NFL, Anthropic, and OpenAI.

• Dataset coverage and limitations
• Compare research approaches