@rulvar/evals 1.226.0 → 1.228.0

package/dist/index.js

@@ -3399,10 +3399,10 @@ const SCENARIOS = [
 	},
 	{
 		name: "validator-guidance-conflict",
-		doctrine: "the evidence-grade reason names the SAFE composition against its cited-value sibling (RV2202): the c3 trap finish repairs in ONE round by moving the run id into its own sentence, with cited-value never rejecting; the third subscription run burned both repairs between the two verdicts",
+		doctrine: "the evidence-grade reason names the SAFE composition against its cited-value sibling and names the run id it wants written (RV2202, RV2501, RV2502): the c3 trap finish repairs in ONE round by carrying THIS run's own id in the graded sentence, beside a source citation, with cited-value reading that id as identity and never rejecting; the third subscription run burned both repairs between the two verdicts, and the 1.226.0 comparison run burned both again on an id no artifact pattern could match",
 		async run() {
-			const TRAP_FINISH = "The reserve fold is live-observed under sustained load. The engine seals the journal at settle (README.md:3).";
-			const FIXED_FINISH = "The reserve fold is live-observed under sustained load in run 01ARZ3NDEKTSV4RRFFQ69G5FAV. The engine seals the journal at settle (README.md:3).";
+			const TRAP_FINISH = "The reserve fold is live-observed under sustained load. The engine seals the journal at settle (`README.md:3`).";
+			const FIXED_FINISH = "The reserve fold is live-observed under sustained load in run `fault-guidance-conflict`, where the engine seals the journal at settle (`README.md:3`).";
 			const calls = [];
 			let finishAttempts = 0;
 			const adapter = {
@@ -3453,14 +3453,14 @@ const SCENARIOS = [
 			const entries = await store.load("fault-guidance-conflict");
 			const repairRequest = calls[1];
 			const repairBytes = JSON.stringify(repairRequest?.messages ?? []);
-			const guidanceQuoted = repairBytes.includes("SEPARATE sentence") && repairBytes.includes("run id");
-			const citedValueNamed = repairBytes.includes("cited-value one");
+			const guidanceQuoted = repairBytes.includes("write this run's id fault-guidance-conflict") && repairBytes.includes("may share a sentence with a source citation");
+			const citedValueNamed = repairBytes.includes("cited-value reads a run id as identity");
 			const decisionsText = JSON.stringify(entries.filter((entry) => entry.kind === "decision").map((entry) => entry.value ?? null));
 			const citedValueRejected = decisionsText.includes("\"cited-value\"") ? decisionsText.includes("not present inside the cited window") : false;
 			return {
 				observation: {
 					matched: outcome.status === "ok" && outcome.value === FIXED_FINISH && finishAttempts === 2 && guidanceQuoted && citedValueNamed && !citedValueRejected,
-					detail: `run '${outcome.status}' after ${String(finishAttempts)} finish attempt(s); the repair exchange quoted the safe composition (SEPARATE sentence + run id: ${String(guidanceQuoted)}) and named the cited-value trade (${String(citedValueNamed)}); cited-value rejected=${String(citedValueRejected)}; final result carries the separate-sentence run id: ${String(outcome.value === FIXED_FINISH)}`
+					detail: `run '${outcome.status}' after ${String(finishAttempts)} finish attempt(s); the repair exchange named the run id and the shared-sentence composition (${String(guidanceQuoted)}) and named the identity reading (${String(citedValueNamed)}); cited-value rejected=${String(citedValueRejected)}; final result carries this run's own id beside the citation in the graded sentence: ${String(outcome.value === FIXED_FINISH)}`
 				},
 				artifacts: [
 					jsonArtifact("outcome.json", {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@rulvar/evals",
-  "version": "1.226.0",
+  "version": "1.228.0",
   "description": "Rulvar evals: eval cases, golden outputs, rubric and judge graders, matrix sweeps, canary fingerprint.",
   "type": "module",
   "license": "Apache-2.0",
@@ -22,11 +22,11 @@
     "access": "public"
   },
   "dependencies": {
-    "@rulvar/anthropic": "1.226.0",
-    "@rulvar/core": "1.226.0",
-    "@rulvar/openai": "1.226.0",
-    "@rulvar/plan": "1.226.0",
-    "@rulvar/testing": "1.226.0"
+    "@rulvar/anthropic": "1.228.0",
+    "@rulvar/core": "1.228.0",
+    "@rulvar/openai": "1.228.0",
+    "@rulvar/plan": "1.228.0",
+    "@rulvar/testing": "1.228.0"
   },
   "devDependencies": {
     "@types/node": "^22.20.1",