Previous value: -"The condition the finished response must satisfy before this run counts as accepted. Pass a LIST when the work has more than one honest outcome, for example a deliverable OR a clear statement that it could not be produced. A list passes when ANY entry is present, which is how a loop that correctly refuses rather than inventing is not scored as a failure."New value: +"The condition the finished response must satisfy before this run counts as accepted. Pass a LIST when the work has more than one honest outcome, for example a deliverable OR a clear statement that it could not be produced. In text and exact mode a list passes when ANY entry matches, which is how a loop that correctly refuses rather than inventing is not scored as a failure. In fields mode a list names fields that must ALL be non-empty."

Previous value: -"How the criterion is checked. text (default): the response must contain the criterion phrase (or any entry of a criterion list). Use this when the work has an honest decline alongside its deliverable. exact: the response must BE the criterion (or any entry of a list) after stripping quotes, backticks, emphasis and whitespace; a response that merely mentions the phrase fails. Use exact for sealed checks (canary tokens, holdout answers) where a refusal that names the answer must not count as accepted. fields: the criterion names structured output fields that must be present and non-empty. evidence: the criterion names artifact kinds (screenshot, trace, test, …) the result must carry, with a retrievable reference for each."New value: +"How the criterion is checked. text (default): the response must contain the criterion phrase (or any entry of a criterion list). Use this when the work has an honest decline alongside its deliverable. exact: the response must BE the criterion (or any entry of a list) after stripping quotes, backticks, emphasis and whitespace; a response that merely mentions the phrase fails. Use exact for sealed checks (canary tokens, holdout answers) where a refusal that names the answer must not count as accepted. fields: every named field must be present and non-empty; a list means all of them, not any, and an empty string, list or object counts as missing. When either of two fields may carry the result (a deliverable, or a stated reason there is none), name neither here and state the OR with anyOf in outputSchema. evidence: the criterion names artifact kinds (screenshot, trace, test, …) the result must carry, with a retrievable reference for each."

Previous value: -"JSON Schema the run's structured output must satisfy. A response that violates it fails the run instead of being recorded as a success."New value: +"JSON Schema the run's structured output must satisfy. A response that violates it fails the run instead of being recorded as a success. A string field listed in required must also be non-empty, so leave a field that may be empty out of required, and state 'this OR that' with anyOf."

Previous value: -"Optional one-time, interval, or cron schedule."New value: +"Optional one-time, interval, or cron schedule. Omit to run once now."

Added value: +{
+  "expr": {
+    "description": "For cron: a five-field cron expression, e.g. \"0 9 * * 1-5\".",
+    "type": "string"
+  },
+  "intervalMs": {
+    "description": "For interval: milliseconds between runs.",
+    "exclusiveMinimum": 0,
+    "maximum": 9007199254740991,
+    "type": "integer"
+  },
+  "runAtMs": {
+    "description": "For once: unix ms of the run.",
+    "type": "number"
+  },
+  "type": {
+    "description": "once (default) runs at runAtMs; interval repeats every intervalMs; cron follows expr in tz.",
+    "enum": [
+      "once",
+      "interval",
+      "cron"
+    ],
+    "type": "string"
+  },
+  "tz": {
+    "description": "For cron: IANA time zone. Default UTC.",
+    "type": "string"
+  }
+}

Removed value: -{
-  "type": "string"
-}