eval-quality 0.1.0 → 0.2.0

package/schemas/sealed-run-record.schema.json

@@ -38,6 +38,14 @@
       "minLength": 1,
       "description": "An opaque caller label with no product semantics, per AD-24. The prior art's five-member enum does not survive, and its own extension history is the reason: an enum a local amendment had to widen once for `self-review` will be widened again."
     },
+    "mode": {
+      "type": "string",
+      "enum": [
+        "production",
+        "contract-scoring"
+      ],
+      "description": "AD-21's run mode, supplied by the caller on the record and never derived, recomputed, or defaulted afterwards. AD-21 requires mode to be \"fixed before ingest\", and owed item 4 records what its absence costs: the same sealed run could be relabelled after ingest and scored under the same scoring version. Required rather than optional, which makes this a BREAKING `schemaVersion` bump under AD-11, whose rule is that \"adding an optional field is a `schemaVersion` bump recorded in the field's own description; removing or retyping is breaking\". A version-1 record carries no mode, and no default may repair one into a version-2 record, because a defaulted mode is the relabelling this field exists to stop. A record presenting no mode fails to parse, which AD-28 makes a `schema-parse-failure` fault rather than an AD-21 verdict or an AD-5 code, the same routing `evaluatorRecommendation` already records for an unrecognised value. This field is where mode is read from; the evidence artifact restates it and is never the source."
+    },
     "trialIndex": {
       "type": "integer",
       "minimum": 1,
@@ -124,6 +132,12 @@
             "pattern": "^[a-z0-9]+(?:-[a-z0-9]+)*$",
             "description": "A kebab-case slug. Excludes \"/\" and \"~\" so an identifier can be embedded in an interaction-rooted pointer without escaping."
           },
+          "sequence": {
+            "type": "integer",
+            "exclusiveMinimum": 0,
+            "maximum": 9007199254740991,
+            "description": "The total order ADR-006 forbids reading off array position (owed item 2). A positive integer, unique across every observation in the same record: uniqueness is what a sorted-by-`sequence` read needs to be strictly increasing, so no separate monotonicity check is required. Neither starting at 1 nor contiguous is required across a record: only positivity and per-record uniqueness are enforced, so a record whose sequences are e.g. [5, 12, 40] is equally valid. Required rather than optional, which makes this a BREAKING `schemaVersion` bump under AD-11: adding an optional field is additive, and this field is not optional. A record with an absent or duplicated `sequence` fails to parse."
+          },
           "operationId": {
             "type": "string",
             "pattern": "^[a-z0-9]+(?:-[a-z0-9]+)*$",
@@ -290,6 +304,7 @@
         },
         "required": [
           "observationId",
+          "sequence",
           "operationId",
           "provenance",
           "callInputs",
@@ -301,7 +316,8 @@
           "exitCode"
         ],
         "additionalProperties": false
-      }
+      },
+      "description": "Per-observation `sequence` uniqueness is enforced here. The published JSON Schema dialect has no keyword for uniqueness of a nested field across array items, so this constraint is Zod-only: a non-Zod consumer must reimplement it, exactly as with any other cross-item invariant this dialect cannot state."
     },
     "judgeResults": {
       "type": "array",
@@ -438,6 +454,7 @@
     "revisionCount",
     "runId",
     "conditionArm",
+    "mode",
     "trialIndex",
     "contractDigest",
     "sealedBriefDigest",
@@ -454,7 +471,7 @@
     "invalidReason"
   ],
   "additionalProperties": false,
-  "description": "One sealed evaluator trial, as the caller presents it. Succeeds the prior-art `h0-run-result` schema per AD-24, keeping its run identifier, condition arm, findings, action-log reference, resource use, invalidation reason, evaluator recommendation as a closed enum, and per-finding confidence on a declared scale. Divergences: `condition` is demoted to the opaque `conditionArm`, `verdict` becomes `evaluatorRecommendation` without `NOT_APPLICABLE`, money is a decimal string, and `taskId`, `note`, and per-finding `actionIds` do not survive: the contract is pinned by `contractDigest`, an unstructured orchestrator annotation is the free-prose channel the Conventions close everywhere else, and two citation vocabularies on one finding is the ambiguity ADR-009 removed. Two constructions are deliberately absent and each is owed to an open item: the run MODE, which AD-21 requires to be fixed before ingest and to enter AD-11's identity inputs rather than appearing first in the evidence artifact (Owed item 4), and observation ORDERING, which ADR-006 forbids reading off array position (Owed item 2). Both arrive as additive `schemaVersion` bumps under AD-11.",
+  "description": "One sealed evaluator trial, as the caller presents it. Succeeds the prior-art `h0-run-result` schema per AD-24, keeping its run identifier, condition arm, findings, action-log reference, resource use, invalidation reason, evaluator recommendation as a closed enum, and per-finding confidence on a declared scale. Divergences: `condition` is demoted to the opaque `conditionArm`, `verdict` becomes `evaluatorRecommendation` without `NOT_APPLICABLE`, money is a decimal string, and `taskId`, `note`, and per-finding `actionIds` do not survive: the contract is pinned by `contractDigest`, an unstructured orchestrator annotation is the free-prose channel the Conventions close everywhere else, and two citation vocabularies on one finding is the ambiguity ADR-009 removed. The run MODE landed here as a required field under a BREAKING `schemaVersion` bump, which is where AD-21's \"fixed before ingest\" puts it; owed item 4 is now closed: mode enters AD-11's identity inputs as `ScoringVersionInputs`'s sixth field, and `core/score/ladder.ts` carries `ProductionAssessment`/`ContractAssessment` as the two assessment input types with their own total ladders. Observation ORDERING landed here too, under its own BREAKING `schemaVersion` bump: `sequence` is required and unique per record, closing owed item 2's ADR-006 gap, since array position was never a legal ordering.",
   "$defs": {
     "EvaluatorRecommendation": {
       "type": "string",