@tangle-network/agent-eval 0.123.0 → 0.123.2

package/dist/{chunk-GS3FJGUF.js → chunk-RQP5UTK5.js}

@@ -15,7 +15,7 @@ import {
   runImprovementLoop,
   surfaceContentHash,
   surfaceHash
-} from "./chunk-HZJF4IUO.js";
+} from "./chunk-HQY7LBV2.js";
 import {
   SearchLedgerConflictError,
   SearchLedgerError,
@@ -51,14 +51,14 @@ import {
   mapConcurrent,
   policyEditsFromFindings,
   validatePolicyEditCandidateRecord
-} from "./chunk-FC5NDO3E.js";
+} from "./chunk-WXQTVEKM.js";
 import {
   callLlm,
   callLlmJson,
   costReceiptFromLlm,
   costReceiptFromLlmError,
   maximumChargeForLlmRequest
-} from "./chunk-NJC7U437.js";
+} from "./chunk-J7S4YM27.js";
 import {
   eProcess,
   mcnemar,
@@ -80,10 +80,10 @@ import {
 import {
   modelHasSnapshot,
   validateRunRecord
-} from "./chunk-S3UZOQ5Y.js";
+} from "./chunk-LOW3U7JZ.js";
 import {
   buildAgentProfileCell
-} from "./chunk-XJYR7XFV.js";
+} from "./chunk-GC4ATIKK.js";
 import {
   canonicalize
 } from "./chunk-VSMTAMNK.js";
@@ -4960,9 +4960,24 @@ function compositeProposer(opts) {
   const members = opts.proposers;
   if (members.length === 0)
     throw new Error("compositeProposer: at least one member proposer required");
+  const memberKinds = /* @__PURE__ */ new Set();
+  for (const member of members) {
+    if (!member.kind || member.kind.trim() !== member.kind) {
+      throw new Error("compositeProposer: member kinds must be trimmed and non-empty");
+    }
+    if (member.kind.includes(":")) {
+      throw new Error(`compositeProposer: member kind '${member.kind}' must not contain ':'`);
+    }
+    if (memberKinds.has(member.kind)) {
+      throw new Error(`compositeProposer: duplicate member kind '${member.kind}'`);
+    }
+    memberKinds.add(member.kind);
+  }
   const weights = opts.weights ?? members.map(() => 1);
-  if (weights.length !== members.length || weights.some((w) => !(w > 0))) {
-    throw new Error("compositeProposer: weights must match proposers length and be positive");
+  if (weights.length !== members.length || weights.some((weight) => !Number.isFinite(weight) || weight <= 0)) {
+    throw new Error(
+      "compositeProposer: weights must match proposers length and be finite and positive"
+    );
   }
   return {
     kind: `composite(${members.map((m) => m.kind).join("+")})`,
@@ -4985,7 +5000,11 @@ function compositeProposer(opts) {
         const share = shares[i] ?? 0;
         if (!member || share === 0 || ctx.signal.aborted) continue;
         try {
-          const proposals = await member.propose({ ...ctx, populationSize: share });
+          const proposals = await member.propose({
+            ...ctx,
+            history: historyForMember(ctx.history, member.kind),
+            populationSize: share
+          });
           for (const proposal of proposals) {
             const isCandidate = typeof proposal === "object" && proposal !== null && "surface" in proposal;
             const surface = isCandidate ? proposal.surface : proposal;
@@ -5019,7 +5038,11 @@ function compositeProposer(opts) {
       return pool.slice(0, ctx.populationSize);
     },
     decide(args) {
-      const votes = members.filter((m) => typeof m.decide === "function").map((m) => m.decide(args));
+      const votes = members.filter((m) => typeof m.decide === "function").map(
+        (m) => m.decide({
+          history: historyForMember(args.history, m.kind)
+        })
+      );
       if (votes.length === 0) return { stop: false };
       const allStop = votes.every((v) => v.stop);
       return allStop ? {
@@ -5029,6 +5052,15 @@ function compositeProposer(opts) {
     }
   };
 }
+function historyForMember(history, memberKind) {
+  const prefix = `${memberKind}:`;
+  return history.map((generation) => ({
+    ...generation,
+    candidates: generation.candidates.map(
+      (candidate) => candidate.label?.startsWith(prefix) ? { ...candidate, label: candidate.label.slice(prefix.length) } : candidate
+    )
+  }));
+}
 
 // src/campaign/proposers/halo.ts
 import { execFile } from "child_process";
@@ -5327,6 +5359,7 @@ var JSON_POLICY_EDIT_TARGET_SURFACES = [
   "agent-profile"
 ];
 var NonEmptyStringSchema = z.string().trim().min(1);
+var JsonObjectKeySchema = z.string().min(1).refine((key) => key.trim() === key, "JSON object keys must not have surrounding whitespace");
 var JsonValueSchema = z.lazy(
   () => z.union([
     z.string(),
@@ -5334,7 +5367,7 @@ var JsonValueSchema = z.lazy(
     z.boolean(),
     z.null(),
     z.array(JsonValueSchema),
-    z.record(NonEmptyStringSchema, JsonValueSchema)
+    z.record(JsonObjectKeySchema, JsonValueSchema)
   ])
 );
 var AuthoredJsonChangeSchema = z.discriminatedUnion("mode", [
@@ -5539,6 +5572,13 @@ var POLICY_EDIT_AUTHOR_SYSTEM = [
   "Treat expectedGain and confidence as forecasts, never as measured evidence. Learn from baselineOutcome, incumbentOutcome, and observedDeltaFromParent.",
   "Do not invent a finding, path, field, score, or task fact. Do not include schemaVersion, editId, metadata, prose, or undeclared keys."
 ].join("\n");
+function policyEditAuthorSystem(responseSchema) {
+  return [
+    POLICY_EDIT_AUTHOR_SYSTEM,
+    "The exact required response JSON Schema follows. Obey it even when the provider does not enforce response_format:",
+    JSON.stringify(responseSchema)
+  ].join("\n");
+}
 function llmPolicyEditProposer(opts) {
   const allowedJsonPaths = validateAllowedJsonPaths(opts.allowedJsonPaths);
   const allowedPathSet = new Set(allowedJsonPaths);
@@ -5593,6 +5633,12 @@ function llmPolicyEditProposer(opts) {
         { currentSurface, allowedJsonPaths, objectives, targetSurface: opts.targetSurface },
         scenarioIds
       );
+      const modelSurface = redactCurrentSurfaceForModel(
+        currentSurface,
+        allowedJsonPaths,
+        opts.redactCurrentSurfaceForModel
+      );
+      assertSurfaceIsTaskAgnostic(modelSurface, scenarioIds);
       const measuredSources = measuredSourceMeasurements(ctx);
       const findings = citableFindings(ctx.findings, measuredSources, maxFindings);
       const findingByKey = new Map(
@@ -5605,7 +5651,7 @@ function llmPolicyEditProposer(opts) {
         objectives,
         candidateCount: limit,
         generation: ctx.generation,
-        currentSurface,
+        currentSurface: modelSurface,
         findings: findings.map(
           (finding, index) => renderFinding(finding, `finding-${index + 1}`, scenarioIds, measuredSources)
         ),
@@ -5633,15 +5679,16 @@ function llmPolicyEditProposer(opts) {
         allowedJsonPaths,
         objectives
       );
+      const system = policyEditAuthorSystem(responseSchema);
       assertPolicyEditAuthorContextBudget(
-        { system: POLICY_EDIT_AUTHOR_SYSTEM, authorContext, responseSchema },
+        { system, authorContext, responseSchema },
         maxAuthorContextChars
       );
       const userContent = JSON.stringify(authorContext);
       const request = {
         model: opts.model,
         messages: [
-          { role: "system", content: POLICY_EDIT_AUTHOR_SYSTEM },
+          { role: "system", content: system },
           { role: "user", content: userContent }
         ],
         jsonSchema: {
@@ -6209,6 +6256,65 @@ function parseJsonSurface(surface) {
   }
   return parsed;
 }
+function redactCurrentSurfaceForModel(surface, allowedJsonPaths, redact) {
+  if (!redact) return surface;
+  const redacted = redact(structuredClone(surface));
+  const parsed = JsonValueSchema.safeParse(redacted);
+  if (!parsed.success) {
+    const detail = formatJsonValidationError(parsed.error);
+    throw new Error(
+      `llmPolicyEditProposer: redactCurrentSurfaceForModel returned invalid JSON (${detail})`
+    );
+  }
+  if (!parsed.data || typeof parsed.data !== "object" || Array.isArray(parsed.data)) {
+    throw new Error("llmPolicyEditProposer: redactCurrentSurfaceForModel must return a JSON object");
+  }
+  for (const path of allowedJsonPaths) {
+    if (!jsonValuesEqual(readJsonPath(surface, path), readJsonPath(parsed.data, path))) {
+      throw new Error(
+        `llmPolicyEditProposer: redactCurrentSurfaceForModel must not change or hide editable JSON path '${path}'`
+      );
+    }
+  }
+  return parsed.data;
+}
+function formatJsonValidationError(error) {
+  const messages = [...new Set(collectZodMessages(error.issues))];
+  const informative = messages.filter(
+    (message) => message !== "Invalid input" && message !== "Invalid key in record"
+  );
+  const custom = informative.filter((message) => !message.startsWith("Invalid input: expected"));
+  return (custom.length > 0 ? custom : informative).join("; ") || "invalid JSON value";
+}
+function collectZodMessages(value) {
+  if (Array.isArray(value)) return value.flatMap(collectZodMessages);
+  if (!value || typeof value !== "object") return [];
+  const issue = value;
+  return [
+    ...typeof issue.message === "string" ? [issue.message] : [],
+    ...collectZodMessages(issue.errors),
+    ...collectZodMessages(issue.issues)
+  ];
+}
+function readJsonPath(root, path) {
+  let cursor = root;
+  for (const part of path.split(".").map((segment) => segment.trim()).filter(Boolean)) {
+    if (!cursor || typeof cursor !== "object" || Array.isArray(cursor)) return void 0;
+    cursor = cursor[part];
+  }
+  return cursor;
+}
+function jsonValuesEqual(left, right) {
+  if (left === right) return true;
+  if (left === void 0 || right === void 0 || left === null || right === null) return false;
+  if (Array.isArray(left) || Array.isArray(right)) {
+    return Array.isArray(left) && Array.isArray(right) && left.length === right.length && left.every((value, index) => jsonValuesEqual(value, right[index]));
+  }
+  if (typeof left !== "object" || typeof right !== "object") return false;
+  const leftKeys = Object.keys(left);
+  const rightKeys = Object.keys(right);
+  return leftKeys.length === rightKeys.length && leftKeys.every((key) => Object.hasOwn(right, key) && jsonValuesEqual(left[key], right[key]));
+}
 function citableFindings(inputs, measuredSources, limit) {
   if (inputs.length === 0) {
     throw new Error("llmPolicyEditProposer: at least one analyst finding is required");
@@ -8279,4 +8385,4 @@ export {
   verifyCodeSurface,
   resolveWorktreePath
 };
-//# sourceMappingURL=chunk-GS3FJGUF.js.map
+//# sourceMappingURL=chunk-RQP5UTK5.js.map