@skill-harness/core 0.11.0 → 0.12.0

package/schemas/qualification-config-v1.schema.json

@@ -0,0 +1,244 @@
+{
+  "$schema": "https://json-schema.org/draft/2020-12/schema",
+  "$id": "https://skill-harness.dev/schemas/qualification-config-v1.schema.json",
+  "title": "skill-harness qualification runner configuration v1",
+  "type": "object",
+  "additionalProperties": false,
+  "required": ["schema_version", "mode", "product", "engine", "producer", "runner", "accounting", "arms"],
+  "properties": {
+    "schema_version": { "const": "qualification-config-v1" },
+    "oauth_directory_policy": { "const": "qualification-oauth-directory-policy-v2" },
+    "terminal_receipt_version": { "const": "qualification-terminal-receipt-v3" },
+    "judge_panel": { "$ref": "#/$defs/judgePanel" },
+    "board": { "$ref": "#/$defs/board" },
+    "mode": { "enum": ["production", "test"] },
+    "product": { "$ref": "#/$defs/productPin" },
+    "engine": { "$ref": "#/$defs/enginePin" },
+    "producer": { "$ref": "#/$defs/producerPin" },
+    "runner": { "$ref": "#/$defs/runner" },
+    "accounting": { "oneOf": [{ "$ref": "#/$defs/accountingLegacy" }, { "$ref": "#/$defs/accountingPanel" }] },
+    "arms": { "type": "array", "minItems": 2, "uniqueItems": true, "items": { "$ref": "#/$defs/arm" } }
+  },
+  "allOf": [
+    { "if": { "properties": { "judge_panel": { "$ref": "#/$defs/judgePanel" } }, "required": ["judge_panel"] }, "then": { "required": ["oauth_directory_policy", "terminal_receipt_version", "board"], "properties": { "accounting": { "$ref": "#/$defs/accountingPanel" }, "board": { "$ref": "#/$defs/board" }, "oauth_directory_policy": { "const": "qualification-oauth-directory-policy-v2" }, "terminal_receipt_version": { "const": "qualification-terminal-receipt-v3" } } }, "else": { "properties": { "accounting": { "$ref": "#/$defs/accountingLegacy" }, "board": false } } },
+    { "if": { "properties": { "terminal_receipt_version": { "const": "qualification-terminal-receipt-v3" } }, "required": ["terminal_receipt_version"] }, "then": { "properties": { "oauth_directory_policy": { "const": "qualification-oauth-directory-policy-v2" } }, "required": ["oauth_directory_policy"] } },
+    { "type": "object", "properties": { "arms": { "type": "array", "contains": { "type": "object", "properties": { "kind": { "const": "subject" } }, "required": ["kind"] }, "minContains": 1 } } },
+    { "type": "object", "properties": { "arms": { "type": "array", "contains": { "type": "object", "properties": { "kind": { "const": "judge" } }, "required": ["kind"] }, "minContains": 1 } } },
+    {
+      "if": { "type": "object", "properties": { "mode": { "const": "production" } }, "required": ["mode"] },
+      "then": {
+        "type": "object", "properties": {
+          "product": { "type": "object", "properties": {
+            "repository": { "const": "https://github.com/mojomanyana/principal-pi-skills" },
+            "commit": { "const": "a6596950d64a3a525f95329d5dbd3e38948be408" },
+            "tree": { "const": "960359d69deb6f216724b86e13eef67e2f6a6aa1" },
+            "package_sha256": { "const": "3677f76fb31dcaf6a28c9e8b9cf9d6358f998b78bd370f0a0f250d05e136fcc8" },
+            "package_bytes": { "const": 110168 }
+          } },
+          "engine": { "type": "object", "properties": { "repository": { "const": "https://github.com/mojomanyana/skill-harness" } } },
+          "producer": { "type": "object", "properties": {
+            "repository": { "const": "https://github.com/mojomanyana/pi-daddy" },
+            "commit": { "const": "4a9524394ca995fd74ed9bbb836dc4e73cda3b8c" },
+            "tree": { "const": "7c006bff213142634f0f911ba9bd6add363ecaae" },
+            "version": { "const": "0.21.1" },
+            "ledger_version": { "const": 3 },
+            "ledger_schema_sha256": { "const": "64e3d875e74bc32fa43fb96892605548259cd16f6ed6678646d73cc56280c511" }
+          } },
+          "arms": {
+            "type": "array", "items": {
+              "type": "object", "allOf": [
+                { "type": "object", "properties": { "provider": { "const": "openai-codex" }, "authentication": { "const": "chatgpt-oauth" } } },
+                {
+                  "if": { "type": "object", "properties": { "kind": { "const": "subject" } }, "required": ["kind"] },
+                  "then": { "type": "object", "properties": { "model": { "enum": ["gpt-5.6-luna", "gpt-5.6-terra"] } } },
+                  "else": { "type": "object", "properties": { "model": { "const": "gpt-5.6-sol" } } }
+                },
+                { "type": "object", "properties": { "allowed_environment_names": { "type": "array", "contains": { "enum": ["HOME", "PI_CODING_AGENT_DIR"] } } } }
+              ]
+            }
+          }
+        }
+      },
+      "else": {
+        "type": "object", "properties": {
+          "arms": { "type": "array", "items": { "type": "object", "properties": { "provider": { "const": "fake" }, "authentication": { "const": "test-oauth" } } } }
+        }
+      }
+    }
+  ],
+  "$defs": {
+    "sha256": { "type": "string", "pattern": "^[a-fA-F0-9]{64}$" },
+    "gitSha": { "type": "string", "pattern": "^[a-fA-F0-9]{40}$" },
+    "absolutePath": { "type": "string", "pattern": "^(?:/|[A-Za-z]:[\\\\/])" },
+    "executable": {
+      "type": "object",
+      "additionalProperties": false,
+      "required": ["path", "sha256"],
+      "properties": { "path": { "$ref": "#/$defs/absolutePath" }, "sha256": { "$ref": "#/$defs/sha256" } }
+    },
+    "repositoryPin": {
+      "type": "object",
+      "required": ["repository", "commit", "tree"],
+      "properties": {
+        "repository": { "type": "string", "minLength": 1 },
+        "commit": { "$ref": "#/$defs/gitSha" },
+        "tree": { "$ref": "#/$defs/gitSha" }
+      }
+    },
+    "productPin": {
+      "allOf": [
+        { "$ref": "#/$defs/repositoryPin" },
+        {
+          "type": "object", "additionalProperties": false,
+          "required": ["repository", "commit", "tree", "checkout_path", "package_path", "package_sha256", "package_bytes"],
+          "properties": {
+            "repository": { "type": "string", "minLength": 1 }, "commit": { "$ref": "#/$defs/gitSha" }, "tree": { "$ref": "#/$defs/gitSha" },
+            "checkout_path": { "$ref": "#/$defs/absolutePath" }, "package_path": { "$ref": "#/$defs/absolutePath" },
+            "package_sha256": { "$ref": "#/$defs/sha256" }, "package_bytes": { "type": "integer", "minimum": 1 }
+          }
+        }
+      ]
+    },
+    "enginePin": {
+      "type": "object", "additionalProperties": false,
+      "required": ["repository", "commit", "tree", "checkout_path", "package_paths", "package_sha256"],
+      "properties": {
+        "repository": { "type": "string", "minLength": 1 }, "commit": { "$ref": "#/$defs/gitSha" }, "tree": { "$ref": "#/$defs/gitSha" },
+        "checkout_path": { "$ref": "#/$defs/absolutePath" },
+        "package_paths": {
+          "type": "object", "additionalProperties": false, "required": ["core", "adapters", "cli", "meta"],
+          "properties": { "core": { "$ref": "#/$defs/absolutePath" }, "adapters": { "$ref": "#/$defs/absolutePath" }, "cli": { "$ref": "#/$defs/absolutePath" }, "meta": { "$ref": "#/$defs/absolutePath" } }
+        },
+        "package_sha256": {
+          "type": "object", "additionalProperties": false, "required": ["core", "adapters", "cli", "meta"],
+          "properties": { "core": { "$ref": "#/$defs/sha256" }, "adapters": { "$ref": "#/$defs/sha256" }, "cli": { "$ref": "#/$defs/sha256" }, "meta": { "$ref": "#/$defs/sha256" } }
+        }
+      }
+    },
+    "producerPin": {
+      "type": "object", "additionalProperties": false,
+      "required": ["repository", "commit", "tree", "checkout_path", "version", "ledger_version", "ledger_schema_sha256"],
+      "properties": {
+        "repository": { "type": "string", "minLength": 1 }, "commit": { "$ref": "#/$defs/gitSha" }, "tree": { "$ref": "#/$defs/gitSha" },
+        "checkout_path": { "$ref": "#/$defs/absolutePath" }, "version": { "type": "string", "minLength": 1 },
+        "ledger_version": { "const": 3 }, "ledger_schema_sha256": { "$ref": "#/$defs/sha256" }
+      }
+    },
+    "runner": {
+      "type": "object", "additionalProperties": false,
+      "required": ["version", "executable", "conflicting_parent_environment"],
+      "properties": {
+        "version": { "const": "qualification-runner-v1" }, "executable": { "$ref": "#/$defs/executable" },
+        "conflicting_parent_environment": { "enum": ["refuse", "remove-and-record"] }
+      }
+    },
+    "board": {
+      "type": "object", "additionalProperties": false, "required": ["schema_version", "read_only", "cells"],
+      "properties": {
+        "schema_version": { "const": "qualification-board-v1" }, "read_only": { "const": true },
+        "cells": { "type": "array", "minItems": 1, "items": { "$ref": "#/$defs/boardCell" } }
+      }
+    },
+    "boardCell": {
+      "type": "object", "additionalProperties": false, "required": ["id", "scenario_id", "measurement_identity_sha256", "scenario_version", "stimulus_sha256", "rubric_sha256", "subject_input_sha256", "subject_arm", "judge_arms", "panels", "critical", "pass_threshold"],
+      "properties": {
+        "id": { "type": "string", "pattern": "^[A-Za-z0-9][A-Za-z0-9._:-]{0,127}$" },
+        "scenario_id": { "type": "string", "pattern": "^[A-Za-z0-9][A-Za-z0-9._:-]{0,127}$" },
+        "measurement_identity_sha256": { "$ref": "#/$defs/sha256" }, "scenario_version": { "type": "string", "pattern": "^[A-Za-z0-9][A-Za-z0-9._:-]{0,127}$" },
+        "stimulus_sha256": { "$ref": "#/$defs/sha256" }, "rubric_sha256": { "$ref": "#/$defs/sha256" }, "subject_input_sha256": { "$ref": "#/$defs/sha256" },
+        "subject_arm": { "type": "string", "pattern": "^[A-Za-z0-9][A-Za-z0-9._:-]{0,127}$" },
+        "judge_arms": { "type": "array", "minItems": 3, "maxItems": 3, "items": { "type": "string", "pattern": "^[A-Za-z0-9][A-Za-z0-9._:-]{0,127}$" } },
+        "panels": { "type": "array", "minItems": 1, "items": { "type": "object", "additionalProperties": false, "required": ["id", "repetition"], "properties": { "id": { "type": "string", "pattern": "^[A-Za-z0-9][A-Za-z0-9._:-]{0,127}$" }, "repetition": { "type": "integer", "minimum": 0 } } } },
+        "critical": { "type": "boolean" }, "pass_threshold": { "type": "integer", "minimum": 1 }
+      }
+    },
+    "judgePanel": {
+      "type": "object", "additionalProperties": false,
+      "required": ["schema_version", "initial_judge_calls", "agreement", "split", "max_judge_calls", "error", "unresolved", "behavioral_failure", "integrity_failure", "wave_a_call_budget", "complete_program_call_budget"],
+      "properties": {
+        "schema_version": { "const": "qualification-judge-panel-policy-v1" },
+        "initial_judge_calls": { "const": 2 },
+        "agreement": { "const": "two-matching-clean-votes" },
+        "split": { "const": "one-conditional-tie-break" },
+        "max_judge_calls": { "const": 3 },
+        "error": { "const": "record-as-non-vote" },
+        "unresolved": { "const": "inconclusive-blocks-acceptance" },
+        "behavioral_failure": { "const": "record-and-continue-read-only-board" },
+        "integrity_failure": { "const": "halt" },
+        "wave_a_call_budget": {
+          "type": "object", "additionalProperties": false, "required": ["artifacts", "minimum", "maximum"],
+          "properties": { "artifacts": { "const": 54 }, "minimum": { "const": 108 }, "maximum": { "const": 162 } }
+        },
+        "complete_program_call_budget": {
+          "type": "object", "additionalProperties": false, "required": ["artifacts", "minimum", "maximum"],
+          "properties": { "artifacts": { "const": 642 }, "minimum": { "const": 1284 }, "maximum": { "const": 1926 } }
+        }
+      }
+    },
+    "artifact": {
+      "type": "object", "additionalProperties": false, "required": ["type", "relative_path_template"],
+      "properties": { "type": { "const": "pi-jsonl" }, "relative_path_template": { "const": "artifacts/{invocation_id}.jsonl" } }
+    },
+    "arm": {
+      "type": "object", "additionalProperties": false,
+      "required": ["id", "kind", "provider", "model", "authentication", "executable", "resources", "arguments", "allowed_environment_names", "timeout_ms", "output_limit_bytes", "artifact", "fallback", "metered_override"],
+      "properties": {
+        "id": { "type": "string", "pattern": "^[A-Za-z0-9][A-Za-z0-9._:-]{0,127}$" },
+        "kind": { "enum": ["subject", "judge"] }, "provider": { "type": "string", "minLength": 1 }, "model": { "type": "string", "minLength": 1 },
+        "authentication": { "enum": ["chatgpt-oauth", "test-oauth"] }, "executable": { "$ref": "#/$defs/executable" },
+        "resources": {
+          "type": "array", "uniqueItems": true,
+          "items": {
+            "type": "object", "additionalProperties": false, "required": ["kind", "path", "sha256"],
+            "properties": { "kind": { "enum": ["extension", "skill", "system-prompt"] }, "path": { "$ref": "#/$defs/absolutePath" }, "sha256": { "$ref": "#/$defs/sha256" } }
+          }
+        },
+        "arguments": { "type": "array", "items": { "type": "string", "pattern": "^(?!-)(?:[^{}]|\\{(?:input_path|scenario_id|invocation_id|artifact_path)\\})+$" }, "contains": { "type": "string", "pattern": "\\{input_path\\}" } },
+        "allowed_environment_names": { "type": "array", "items": { "type": "string", "pattern": "^[A-Z_][A-Z0-9_]*$" }, "uniqueItems": true },
+        "timeout_ms": { "type": "integer", "minimum": 1, "maximum": 86400000 },
+        "output_limit_bytes": { "type": "integer", "minimum": 1, "maximum": 104857600 },
+        "artifact": { "$ref": "#/$defs/artifact" }, "fallback": { "const": false }, "metered_override": { "const": false }
+      }
+    },
+    "accountingLegacy": {
+      "type": "object", "additionalProperties": false, "required": ["wave_a", "complete_program", "ceilings", "initial"],
+      "properties": { "wave_a": { "$ref": "#/$defs/waveA" }, "complete_program": { "$ref": "#/$defs/completeProgram" }, "ceilings": { "$ref": "#/$defs/ceilings" }, "initial": { "$ref": "#/$defs/initial" } }
+    },
+    "accountingPanel": {
+      "type": "object", "additionalProperties": false, "required": ["wave_a", "complete_program", "ceilings", "initial"],
+      "properties": { "wave_a": { "$ref": "#/$defs/waveAPanel" }, "complete_program": { "$ref": "#/$defs/completeProgramPanel" }, "ceilings": { "$ref": "#/$defs/ceilingsPanel" }, "initial": { "$ref": "#/$defs/initial" } }
+    },
+    "pair": {
+      "type": "object", "additionalProperties": false, "required": ["subject", "judge"],
+      "properties": { "subject": { "type": "integer" }, "judge": { "type": "integer" } }
+    },
+    "waveA": {
+      "type": "object", "additionalProperties": false, "required": ["subject", "judge"],
+      "properties": { "subject": { "const": 54 }, "judge": { "const": 54 } }
+    },
+    "completeProgram": {
+      "type": "object", "additionalProperties": false, "required": ["subject", "judge"],
+      "properties": { "subject": { "const": 642 }, "judge": { "const": 642 } }
+    },
+    "waveAPanel": {
+      "type": "object", "additionalProperties": false, "required": ["subject", "judge"],
+      "properties": { "subject": { "const": 54 }, "judge": { "const": 162 } }
+    },
+    "completeProgramPanel": {
+      "type": "object", "additionalProperties": false, "required": ["subject", "judge"],
+      "properties": { "subject": { "const": 642 }, "judge": { "const": 1926 } }
+    },
+    "ceilingsPanel": {
+      "type": "object", "additionalProperties": false, "required": ["subject", "judge"],
+      "properties": { "subject": { "const": 700 }, "judge": { "const": 2100 } }
+    },
+    "ceilings": {
+      "type": "object", "additionalProperties": false, "required": ["subject", "judge"],
+      "properties": { "subject": { "const": 700 }, "judge": { "const": 700 } }
+    },
+    "initial": {
+      "type": "object", "additionalProperties": false, "required": ["subject", "judge"],
+      "properties": { "subject": { "const": 0 }, "judge": { "const": 0 } }
+    }
+  }
+}

package/schemas/qualification-invocation-request-v1.schema.json

@@ -0,0 +1,59 @@
+{
+  "$schema": "https://json-schema.org/draft/2020-12/schema",
+  "$id": "https://skill-harness.dev/schemas/qualification-invocation-request-v1.schema.json",
+  "title": "skill-harness qualification invocation request v1",
+  "type": "object",
+  "additionalProperties": false,
+  "required": ["schema_version", "measurement_identity_sha256", "invocation_id", "continuation_authority_sha256", "continuation_authority_expires_at", "scenario", "role", "counts_as_measurement", "arms", "selected_arm", "repetition"],
+  "properties": {
+    "schema_version": { "const": "qualification-invocation-request-v1" },
+    "measurement_identity_sha256": { "$ref": "#/$defs/sha256" },
+    "invocation_id": { "$ref": "#/$defs/id" },
+    "continuation_authority_sha256": { "$ref": "#/$defs/sha256" },
+    "continuation_authority_expires_at": { "type": "string", "pattern": "^\\d{4}-\\d{2}-\\d{2}T\\d{2}:\\d{2}:\\d{2}(?:\\.\\d+)?Z$" },
+    "scenario": {
+      "type": "object", "additionalProperties": false,
+      "required": ["id", "version", "stimulus_sha256", "rubric_sha256", "input_path", "input_sha256", "working_directory"],
+      "properties": {
+        "id": { "$ref": "#/$defs/id" }, "version": { "$ref": "#/$defs/id" },
+        "stimulus_sha256": { "$ref": "#/$defs/sha256" }, "rubric_sha256": { "$ref": "#/$defs/sha256" },
+        "input_path": { "$ref": "#/$defs/absolutePath" }, "input_sha256": { "$ref": "#/$defs/sha256" },
+        "working_directory": { "$ref": "#/$defs/absolutePath" }
+      }
+    },
+    "role": { "enum": ["holdout-author", "holdout-reviewer", "subject", "judge", "calibration", "canary"] },
+    "counts_as_measurement": { "type": "boolean" },
+    "arms": {
+      "type": "object", "additionalProperties": false, "required": ["subject", "judge"],
+      "properties": { "subject": { "$ref": "#/$defs/id" }, "judge": { "$ref": "#/$defs/id" } }
+    },
+    "selected_arm": { "$ref": "#/$defs/id" },
+    "repetition": { "type": "integer", "minimum": 0 },
+    "panel": {
+      "type": "object", "additionalProperties": false,
+      "required": ["id", "member_ordinal", "subject_invocation_id", "subject_artifact_sha256"],
+      "properties": {
+        "id": { "$ref": "#/$defs/id" },
+        "member_ordinal": { "type": "integer", "minimum": 1, "maximum": 3 },
+        "subject_invocation_id": { "$ref": "#/$defs/id" },
+        "subject_artifact_sha256": { "$ref": "#/$defs/sha256" }
+      }
+    }
+  },
+  "allOf": [
+    {
+      "if": { "properties": { "panel": { "type": "object" } }, "required": ["panel"] },
+      "then": { "properties": { "role": { "const": "judge" }, "selected_arm": { "$ref": "#/$defs/id" } }, "required": ["role", "selected_arm"] }
+    },
+    {
+      "if": { "type": "object", "properties": { "role": { "enum": ["subject", "judge"] } }, "required": ["role"] },
+      "then": { "type": "object", "properties": { "counts_as_measurement": { "const": true } } },
+      "else": { "type": "object", "properties": { "counts_as_measurement": { "const": false } } }
+    }
+  ],
+  "$defs": {
+    "id": { "type": "string", "pattern": "^[A-Za-z0-9][A-Za-z0-9._:-]{0,127}$" },
+    "absolutePath": { "type": "string", "pattern": "^(?:/|[A-Za-z]:[\\\\/])" },
+    "sha256": { "type": "string", "pattern": "^[a-fA-F0-9]{64}$" }
+  }
+}

package/schemas/qualification-judge-panel-result-v1.schema.json

@@ -0,0 +1,42 @@
+{
+  "$schema": "https://json-schema.org/draft/2020-12/schema",
+  "$id": "https://skill-harness.dev/schemas/qualification-judge-panel-result-v1.schema.json",
+  "title": "skill-harness qualification judge panel result v1",
+  "type": "object",
+  "additionalProperties": false,
+  "required": ["schema_version", "panel_id", "scenario_id", "subject_arm", "repetition", "critical", "state", "members", "judge_calls", "clean_votes", "disagreement", "acceptance", "collection"],
+  "properties": {
+    "schema_version": { "const": "qualification-judge-panel-result-v1" },
+    "panel_id": { "$ref": "#/$defs/id" }, "scenario_id": { "$ref": "#/$defs/id" }, "subject_arm": { "$ref": "#/$defs/id" },
+    "repetition": { "type": "integer", "minimum": 0 }, "critical": { "type": "boolean" },
+    "state": { "enum": ["confirmed", "tie_broken", "unresolved"] }, "verdict": { "enum": ["PASS", "FAIL"] },
+    "members": { "type": "array", "minItems": 2, "maxItems": 3, "items": { "$ref": "#/$defs/member" } },
+    "judge_calls": { "type": "integer", "minimum": 2, "maximum": 3 }, "clean_votes": { "type": "integer", "minimum": 0, "maximum": 3 },
+    "disagreement": { "type": "object", "additionalProperties": false, "required": ["initial_split", "minority_rate"], "properties": { "initial_split": { "type": "boolean" }, "minority_rate": { "type": "number", "minimum": 0, "maximum": 0.5 } } },
+    "acceptance": { "enum": ["pass", "fail", "inconclusive"] }, "collection": { "const": "continue" }
+  },
+  "allOf": [
+    { "if": { "properties": { "state": { "const": "unresolved" } }, "required": ["state"] }, "then": { "properties": { "acceptance": { "const": "inconclusive" } }, "not": { "properties": { "verdict": {} }, "required": ["verdict"] } }, "else": { "required": ["verdict"], "properties": { "acceptance": { "enum": ["pass", "fail"] } } } },
+    { "if": { "properties": { "state": { "const": "tie_broken" } }, "required": ["state"] }, "then": { "properties": { "members": { "type": "array", "minItems": 3, "maxItems": 3 }, "disagreement": { "type": "object", "properties": { "initial_split": { "const": true } } } } } },
+    { "if": { "properties": { "verdict": { "const": "PASS" } }, "required": ["verdict"] }, "then": { "properties": { "acceptance": { "const": "pass" } } } },
+    { "if": { "properties": { "verdict": { "const": "FAIL" } }, "required": ["verdict"] }, "then": { "properties": { "acceptance": { "const": "fail" } } } },
+    { "if": { "properties": { "judge_calls": { "const": 2 } }, "required": ["judge_calls"] }, "then": { "properties": { "members": { "type": "array", "minItems": 2, "maxItems": 2 } } } },
+    { "if": { "properties": { "judge_calls": { "const": 3 } }, "required": ["judge_calls"] }, "then": { "properties": { "members": { "type": "array", "minItems": 3, "maxItems": 3 } } } },
+    { "if": { "properties": { "disagreement": { "type": "object", "properties": { "initial_split": { "const": true } }, "required": ["initial_split"] } }, "required": ["disagreement"] }, "then": { "properties": { "state": { "const": "tie_broken" }, "members": { "type": "array", "minItems": 3, "maxItems": 3 } } } }
+  ],
+  "$defs": {
+    "id": { "type": "string", "pattern": "^[A-Za-z0-9][A-Za-z0-9._:-]{0,127}$" },
+    "sha256": { "type": "string", "pattern": "^[a-f0-9]{64}$" },
+    "member": {
+      "type": "object", "additionalProperties": false,
+      "required": ["invocation_id", "ordinal", "judge", "verdict", "reason", "suspect", "artifact", "terminal_receipt_sha256"],
+      "properties": {
+        "invocation_id": { "$ref": "#/$defs/id" }, "ordinal": { "type": "integer", "minimum": 1, "maximum": 3 },
+        "judge": { "type": "object", "additionalProperties": false, "required": ["provider", "model"], "properties": { "provider": { "type": "string", "minLength": 1 }, "model": { "type": "string", "minLength": 1 } } },
+        "verdict": { "enum": ["PASS", "FAIL", "ERROR", "JUDGE-AMBIGUOUS"] }, "reason": { "type": "string" }, "suspect": { "type": "boolean" },
+        "artifact": { "oneOf": [{ "type": "null" }, { "type": "object", "additionalProperties": false, "required": ["sha256", "bytes"], "properties": { "sha256": { "$ref": "#/$defs/sha256" }, "bytes": { "type": "integer", "minimum": 0 } } }] },
+        "terminal_receipt_sha256": { "$ref": "#/$defs/sha256" }
+      }
+    }
+  }
+}

package/schemas/results-v2.schema.json

@@ -49,11 +49,27 @@
         "cache_read_tokens": { "type": "integer", "minimum": 0 },
         "cache_write_tokens": { "type": "integer", "minimum": 0 },
         "subject_cost_usd": { "type": "number", "minimum": 0 },
+        "cost_source": { "enum": ["provider-reported", "subscription", "unreported"] },
         "tool_calls": { "type": "integer", "minimum": 0 },
         "delegated_children": { "type": "integer", "minimum": 0 },
         "max_concurrency": { "type": "integer", "minimum": 0 }
       }
     },
+    "judgment": {
+      "type": "object",
+      "additionalProperties": false,
+      "required": ["ordinal", "judge", "verdict", "reason", "suspect"],
+      "properties": {
+        "ordinal": { "type": "integer", "minimum": 1, "maximum": 3 },
+        "judge": {
+          "type": "object", "additionalProperties": false, "required": ["provider", "model"],
+          "properties": { "provider": { "type": "string" }, "model": { "type": "string" } }
+        },
+        "verdict": { "$ref": "#/$defs/verdict" },
+        "reason": { "type": "string" },
+        "suspect": { "type": "boolean" }
+      }
+    },
     "scenario": {
       "type": "object",
       "required": ["id", "judge_verdict", "judge_reason", "suspect", "override", "note"],
@@ -105,28 +121,10 @@
             "state": { "enum": ["confirmed", "tie_broken", "unresolved"] },
             "trigger": { "enum": ["ambiguous", "contradictory", "non_unanimous", "ship_deciding"] },
             "verdict": { "enum": ["PASS", "FAIL"] },
-            "judgments": {
-              "type": "array",
-              "items": {
-                "type": "object",
-                "additionalProperties": false,
-                "required": ["ordinal", "judge", "verdict", "reason", "suspect"],
-                "properties": {
-                  "ordinal": { "type": "integer", "minimum": 1, "maximum": 3 },
-                  "judge": {
-                    "type": "object",
-                    "additionalProperties": false,
-                    "required": ["provider", "model"],
-                    "properties": { "provider": { "type": "string" }, "model": { "type": "string" } }
-                  },
-                  "verdict": { "$ref": "#/$defs/verdict" },
-                  "reason": { "type": "string" },
-                  "suspect": { "type": "boolean" }
-                }
-              }
-            }
+            "judgments": { "type": "array", "items": { "$ref": "#/$defs/judgment" } }
           }
-        }
+        },
+        "judge_history": { "type": "array", "minItems": 2, "maxItems": 3, "items": { "$ref": "#/$defs/judgment" } }
       }
     }
   }