{
  "$schema": "https://json-schema.org/draft/2020-12/schema",
  "$id": "https://hupe1980.github.io/nuvec/benchmarks/parity-record-v1.schema.json",
  "title": "nuvec parity record, version 1",
  "type": "object",
  "required": [
    "schema",
    "schema_version",
    "created",
    "environment",
    "settings",
    "sections",
    "models",
    "summary"
  ],
  "additionalProperties": false,
  "properties": {
    "schema": {
      "const": "nuvec-parity-record"
    },
    "schema_version": {
      "const": 1
    },
    "created": {
      "type": "string",
      "format": "date-time"
    },
    "environment": {
      "type": "object",
      "required": [
        "machine",
        "chip",
        "memory_bytes",
        "macos",
        "power",
        "swap_used_bytes_at_start",
        "nuvec",
        "rust",
        "mlx",
        "mlx_lm"
      ],
      "properties": {
        "machine": {
          "type": "string"
        },
        "chip": {
          "type": "string"
        },
        "cpu_cores": {
          "type": "object",
          "properties": {
            "performance": {
              "type": "integer"
            },
            "efficiency": {
              "type": "integer"
            }
          }
        },
        "gpu_cores": {
          "type": [
            "integer",
            "null"
          ]
        },
        "memory_bytes": {
          "type": "integer",
          "minimum": 0
        },
        "macos": {
          "type": "object",
          "required": [
            "version",
            "build"
          ],
          "properties": {
            "version": {
              "type": "string"
            },
            "build": {
              "type": "string"
            }
          }
        },
        "power": {
          "type": "object",
          "required": [
            "source",
            "low_power_mode"
          ],
          "properties": {
            "source": {
              "enum": [
                "ac",
                "battery",
                "unknown"
              ]
            },
            "low_power_mode": {
              "type": "boolean"
            }
          }
        },
        "swap_used_bytes_at_start": {
          "type": "integer",
          "minimum": 0
        },
        "nuvec": {
          "type": "object",
          "required": [
            "version",
            "commit",
            "dirty"
          ],
          "properties": {
            "version": {
              "type": "string"
            },
            "commit": {
              "type": "string"
            },
            "dirty": {
              "type": "boolean"
            }
          }
        },
        "rust": {
          "type": "string"
        },
        "mlx": {
          "type": "string"
        },
        "mlx_lm": {
          "type": "string"
        },
        "python": {
          "type": "string"
        }
      }
    },
    "settings": {
      "type": "object",
      "required": [
        "passes",
        "runs_per_pass",
        "max_tokens",
        "prompt_tokens"
      ],
      "properties": {
        "passes": {
          "type": "integer",
          "minimum": 1
        },
        "runs_per_pass": {
          "type": "integer",
          "minimum": 1
        },
        "max_tokens": {
          "type": "integer",
          "minimum": 2
        },
        "prompt_tokens": {
          "type": "array",
          "items": {
            "type": "integer",
            "minimum": 1
          }
        },
        "legacy_repeats": {
          "type": "array",
          "items": {
            "type": "integer"
          }
        },
        "concurrency": {
          "type": "object"
        },
        "load": {
          "type": "object"
        },
        "memory": {
          "type": "object",
          "required": [
            "method"
          ],
          "properties": {
            "method": {
              "const": "peak_with_mapped-vs-get_peak_memory+ru_maxrss"
            },
            "note": {
              "type": "string"
            }
          }
        },
        "gates": {
          "type": "object",
          "required": [
            "gate",
            "goal"
          ],
          "properties": {
            "gate": {
              "type": "number"
            },
            "goal": {
              "type": "number"
            }
          }
        },
        "baseline": {
          "type": "object",
          "required": [
            "ref",
            "commit"
          ],
          "properties": {
            "ref": {
              "type": "string"
            },
            "commit": {
              "type": "string"
            }
          }
        },
        "tier": {
          "enum": [
            "quick",
            "full"
          ]
        },
        "regression_baseline": {
          "type": "object",
          "required": [
            "ref",
            "commit",
            "threshold"
          ],
          "properties": {
            "ref": {
              "type": "string"
            },
            "commit": {
              "type": "string"
            },
            "threshold": {
              "type": "number"
            }
          }
        }
      }
    },
    "sections": {
      "type": "array",
      "items": {
        "enum": [
          "decode",
          "ttft",
          "memory",
          "load",
          "concurrency",
          "speculative"
        ]
      },
      "uniqueItems": true
    },
    "models": {
      "type": "array",
      "items": {
        "$ref": "#/$defs/model"
      }
    },
    "summary": {
      "type": "object",
      "required": [
        "below_gate",
        "below_goal",
        "correctness_flags"
      ],
      "properties": {
        "below_gate": {
          "type": "object",
          "additionalProperties": {
            "type": "array",
            "items": {
              "type": "object",
              "required": [
                "model",
                "ratio"
              ],
              "properties": {
                "model": {
                  "type": "string"
                },
                "ratio": {
                  "type": "number"
                }
              }
            }
          }
        },
        "below_goal": {
          "type": "object",
          "additionalProperties": {
            "type": "array",
            "items": {
              "type": "object",
              "required": [
                "model",
                "ratio"
              ],
              "properties": {
                "model": {
                  "type": "string"
                },
                "ratio": {
                  "type": "number"
                }
              }
            }
          }
        },
        "correctness_flags": {
          "type": "array",
          "items": {
            "type": "object",
            "required": [
              "model",
              "check"
            ],
            "properties": {
              "model": {
                "type": "string"
              },
              "level": {
                "type": "integer"
              },
              "check": {
                "const": "matches_solo"
              }
            }
          }
        },
        "regressions": {
          "description": "Cells below the regression threshold against settings.regression_baseline, after one re-measurement.",
          "type": "array",
          "items": {
            "type": "object",
            "required": [
              "model",
              "cell",
              "vs_baseline"
            ],
            "properties": {
              "model": {
                "type": "string"
              },
              "cell": {
                "type": "string"
              },
              "vs_baseline": {
                "type": "number"
              }
            }
          }
        }
      },
      "additionalProperties": {
        "$ref": "#/$defs/geomean"
      }
    }
  },
  "$defs": {
    "metric": {
      "type": "object",
      "required": [
        "median",
        "samples",
        "unit"
      ],
      "properties": {
        "median": {
          "type": "number"
        },
        "samples": {
          "type": "array",
          "items": {
            "type": "number"
          },
          "minItems": 1
        },
        "unit": {
          "enum": [
            "tok/s",
            "s",
            "bytes"
          ]
        }
      }
    },
    "pair": {
      "type": "object",
      "required": [
        "nuvec",
        "mlx_lm",
        "ratio"
      ],
      "properties": {
        "nuvec": {
          "oneOf": [
            {
              "$ref": "#/$defs/metric"
            },
            {
              "type": "null"
            }
          ]
        },
        "mlx_lm": {
          "oneOf": [
            {
              "$ref": "#/$defs/metric"
            },
            {
              "type": "null"
            }
          ]
        },
        "ratio": {
          "type": [
            "number",
            "null"
          ]
        },
        "reason": {
          "type": "string"
        },
        "gate_pass": {
          "type": "boolean"
        },
        "goal_pass": {
          "type": "boolean"
        },
        "baseline": {
          "description": "The baseline release's samples (settings.regression_baseline).",
          "oneOf": [
            {
              "$ref": "#/$defs/metric"
            },
            {
              "type": "null"
            }
          ]
        },
        "vs_baseline": {
          "description": "HEAD against the baseline release, oriented as ratio (above 1 favours HEAD).",
          "type": [
            "number",
            "null"
          ]
        }
      },
      "if": {
        "anyOf": [
          {
            "properties": {
              "nuvec": {
                "type": "null"
              }
            }
          },
          {
            "properties": {
              "mlx_lm": {
                "type": "null"
              }
            }
          }
        ]
      },
      "then": {
        "required": [
          "reason"
        ],
        "properties": {
          "ratio": {
            "type": "null"
          }
        }
      }
    },
    "geomean": {
      "oneOf": [
        {
          "type": "object",
          "required": [
            "value",
            "n"
          ],
          "properties": {
            "value": {
              "type": "number"
            },
            "n": {
              "type": "integer"
            }
          }
        },
        {
          "type": "object",
          "additionalProperties": {
            "$ref": "#/$defs/geomean"
          }
        }
      ]
    },
    "model": {
      "type": "object",
      "required": [
        "id",
        "kind",
        "status"
      ],
      "properties": {
        "id": {
          "type": "string"
        },
        "kind": {
          "enum": [
            "dense",
            "hybrid",
            "moe",
            "hybrid-moe"
          ]
        },
        "snapshot": {
          "type": "string"
        },
        "status": {
          "enum": [
            "ok",
            "partial",
            "skipped",
            "failed"
          ]
        },
        "reason": {
          "type": "string"
        },
        "decode": {
          "type": "array",
          "items": {
            "type": "object",
            "required": [
              "repeats",
              "prompt_tokens",
              "tok_s"
            ],
            "properties": {
              "repeats": {
                "type": "integer"
              },
              "prompt_tokens": {
                "type": "object",
                "properties": {
                  "nuvec": {
                    "type": "integer"
                  },
                  "mlx_lm": {
                    "type": "integer"
                  }
                }
              },
              "tok_s": {
                "$ref": "#/$defs/pair"
              }
            }
          }
        },
        "ttft": {
          "type": "object",
          "properties": {
            "lengths": {
              "type": "array",
              "items": {
                "type": "object",
                "required": [
                  "prompt_tokens",
                  "ttft_s",
                  "prefill_tok_s"
                ],
                "properties": {
                  "prompt_tokens": {
                    "type": "integer"
                  },
                  "ttft_s": {
                    "$ref": "#/$defs/pair"
                  },
                  "prefill_tok_s": {
                    "$ref": "#/$defs/pair"
                  }
                }
              }
            },
            "legacy": {
              "type": "array"
            }
          }
        },
        "memory": {
          "type": "object",
          "additionalProperties": {
            "type": "object",
            "required": [
              "peak_bytes",
              "os_peak_rss_bytes"
            ],
            "properties": {
              "peak_bytes": {
                "$ref": "#/$defs/pair",
                "description": "The one definition: nuvec MemoryStats::peak_with_mapped vs mlx-lm get_peak_memory()"
              },
              "os_peak_rss_bytes": {
                "$ref": "#/$defs/pair"
              },
              "nuvec_raw": {
                "type": "object",
                "required": [
                  "peak_bytes",
                  "active_bytes",
                  "mapped_bytes"
                ],
                "properties": {
                  "peak_bytes": {
                    "type": "integer"
                  },
                  "active_bytes": {
                    "type": "integer"
                  },
                  "mapped_bytes": {
                    "type": "integer"
                  }
                }
              }
            }
          }
        },
        "load": {
          "type": "object",
          "properties": {
            "warm": {
              "$ref": "#/$defs/load_phase"
            },
            "cold": {
              "$ref": "#/$defs/load_phase"
            }
          }
        },
        "concurrency": {
          "type": "array",
          "items": {
            "type": "object",
            "required": [
              "level",
              "total_tokens",
              "wall_s",
              "aggregate_tok_s"
            ],
            "properties": {
              "level": {
                "type": "integer",
                "minimum": 1
              },
              "total_tokens": {
                "type": "integer"
              },
              "wall_s": {
                "$ref": "#/$defs/pair"
              },
              "aggregate_tok_s": {
                "$ref": "#/$defs/pair"
              },
              "decode_aggregate_tok_s": {
                "$ref": "#/$defs/pair"
              },
              "nuvec_matches_solo": {
                "type": [
                  "boolean",
                  "null"
                ]
              }
            }
          }
        },
        "speculative": {
          "type": "array",
          "items": {
            "type": "object",
            "required": [
              "drafter",
              "tok_s"
            ],
            "additionalProperties": false,
            "properties": {
              "drafter": {
                "enum": [
                  "mtp",
                  "lookup"
                ]
              },
              "tok_s": {
                "$ref": "#/$defs/pair"
              }
            }
          }
        }
      },
      "if": {
        "properties": {
          "status": {
            "not": {
              "const": "ok"
            }
          }
        }
      },
      "then": {
        "required": [
          "reason"
        ]
      }
    },
    "load_phase": {
      "type": "object",
      "required": [
        "model_ready_s",
        "first_token_s"
      ],
      "properties": {
        "model_ready_s": {
          "$ref": "#/$defs/pair"
        },
        "first_token_s": {
          "$ref": "#/$defs/pair"
        }
      }
    }
  }
}
