bullet-heaven.ahraz.me

calibrate_pilot

Scores a candidate pilot profile against recorded human sessions by playing it through the very encounters the person played, once per session, then comparing both how it moved and how its runs ended. Returns fitError (lower is better, 0 is identical), a per-metric 'deltas' list showing which knob to move, 'outcomeMatch', and 'warnings'. Search one knob at a time: change a single value, call again, and keep the change only if fitError falls. Search switchMargin, focusBias and wallWeight; hold reactionSeconds and aimErrorDegrees at the values analyze_replays measured, because the recordings already answer those and moving them is reported in 'warnings'. A single call is not a calibration: expect several rounds, and after a knob improves the fit, search the remaining knobs again from that new best profile rather than from the original seed. Two candidates scoring an identical fitError means the knob you moved changed nothing in this encounter, so stop searching it and move to another. A profile is only usable as a difficulty stand-in if it fails where the person failed. 'outcomeMatch' reports that directly: it compares how far the candidate got and whether it died against the recordings, and sets matches=false when the candidate finishes an encounter the person kept losing. Keep searching while matches is false; a low fitError with matches=false describes a different player. Save the winner with save_pilot_profile, then pass its name to evaluate_encounter. The generated runs become pilot replays you can watch.

Page: /editor

Input schema

{
  "type": "object",
  "properties": {
    "profile": {
      "type": "object",
      "properties": {
        "reactionSeconds": {
          "type": "number",
          "minimum": 0.02,
          "maximum": 1
        },
        "aimErrorDegrees": {
          "type": "number",
          "minimum": 0,
          "maximum": 90
        },
        "switchMargin": {
          "type": "number",
          "minimum": 0.5,
          "maximum": 1
        },
        "focusBias": {
          "type": "number",
          "minimum": 0.25,
          "maximum": 4
        },
        "wallWeight": {
          "type": "number",
          "minimum": 0,
          "maximum": 4
        }
      },
      "required": [
        "reactionSeconds",
        "aimErrorDegrees",
        "switchMargin",
        "focusBias",
        "wallWeight"
      ],
      "additionalProperties": false,
      "description": "Candidate knobs to score. aimErrorDegrees is in degrees."
    },
    "replayIds": {
      "minItems": 1,
      "maxItems": 10,
      "type": "array",
      "items": {
        "type": "string",
        "minLength": 1
      },
      "description": "Sessions to fit against; defaults to every human session."
    },
    "seed": {
      "type": "integer",
      "minimum": 0,
      "description": "Reproducibility seed for the candidate's trials. Hold it fixed while searching a knob, or a changed fitError may just be a different seed."
    }
  },
  "required": [
    "profile"
  ],
  "additionalProperties": false
}