@tangle-network/agent-eval 0.49.0 → 0.50.0

package/dist/index.js

@@ -1,3 +1,10 @@
+import {
+  HoldoutAuditor,
+  canaryLeakView,
+  checkBehavioralCanary,
+  checkCanaries,
+  runBehavioralCanaries
+} from "./chunk-SHTXZ4O2.js";
 import {
   DEFAULT_MUTATION_PRIMITIVES,
   DEFAULT_RED_TEAM_CORPUS,
@@ -82,10 +89,10 @@ import {
   evaluateReleaseConfidence,
   judgeReplayGate,
   renderReleaseReport
-} from "./chunk-KQ26DYTQ.js";
+} from "./chunk-UBQGWD3O.js";
 import {
   runEvalCampaign
-} from "./chunk-PD3MH6WU.js";
+} from "./chunk-5KSDYBYH.js";
 import {
   AGENT_PROFILE_KINDS,
   AgentProfileCellValidationError,
@@ -106,14 +113,16 @@ import {
   verifyAgentProfileCell
 } from "./chunk-BWZEGTES.js";
 import {
-  RESEARCH_REPORT_HARD_PAIR_FLOOR,
   evaluateInterimReleaseConfidence,
+  pairedEvalueSequence
+} from "./chunk-MAZ26DC7.js";
+import {
+  RESEARCH_REPORT_HARD_PAIR_FLOOR,
   gainHistogram,
-  pairedEvalueSequence,
   paretoChart,
   researchReport,
   summaryTable
-} from "./chunk-MNL6LXGQ.js";
+} from "./chunk-EGIPWXHL.js";
 import {
   benjaminiHochberg,
   bonferroni,
@@ -6938,107 +6947,6 @@ function defaultExtract(metric) {
   };
 }
 
-// src/contamination-guard.ts
-function checkCanaries(output, scenarios) {
-  const leaks = [];
-  for (const s of scenarios) {
-    if (!s.canary) continue;
-    if (output.includes(s.canary)) {
-      leaks.push({ scenarioId: s.id, canary: s.canary, evidence: excerpt2(output, s.canary) });
-    }
-  }
-  return leaks;
-}
-function checkBehavioralCanary(output, scenario) {
-  const pattern = scenario.forbiddenPattern ?? scenario.canary;
-  if (!pattern) return null;
-  const hit = matchForbidden(output, pattern);
-  if (!hit) return null;
-  return {
-    scenarioId: scenario.id,
-    canary: pattern,
-    evidence: excerpt2(output, hit)
-  };
-}
-function runBehavioralCanaries(cases) {
-  const leaks = [];
-  for (const c of cases) {
-    const leak = checkBehavioralCanary(c.output, c.scenario);
-    if (leak) leaks.push({ ...leak, runId: c.runId ?? leak.runId });
-  }
-  return leaks;
-}
-function matchForbidden(output, pattern) {
-  const re = tryParseRegex(pattern);
-  if (re) {
-    const m = output.match(re);
-    return m && m[0].length > 0 ? m[0] : null;
-  }
-  return output.includes(pattern) ? pattern : null;
-}
-function tryParseRegex(pattern) {
-  if (pattern.length < 2 || pattern[0] !== "/") return null;
-  const last = pattern.lastIndexOf("/");
-  if (last <= 0) return null;
-  const body = pattern.slice(1, last);
-  const flags = pattern.slice(last + 1);
-  if (!/^[gimsuy]*$/.test(flags)) return null;
-  try {
-    return new RegExp(body, flags);
-  } catch {
-    return null;
-  }
-}
-async function canaryLeakView(store, scenarios) {
-  const targets = scenarios.filter((s) => !!s.canary);
-  if (targets.length === 0) return [];
-  const spans = await llmSpans(store);
-  const leaks = [];
-  for (const span of spans) {
-    const output = span.output ?? "";
-    for (const s of targets) {
-      if (s.canary && output.includes(s.canary)) {
-        leaks.push({
-          scenarioId: s.id,
-          canary: s.canary,
-          runId: span.runId,
-          evidence: excerpt2(output, s.canary)
-        });
-      }
-    }
-  }
-  return leaks;
-}
-var HoldoutAuditor = class {
-  scenarios;
-  accessLog = [];
-  constructor(scenarios) {
-    this.scenarios = scenarios;
-  }
-  /** Retrieve a holdout scenario for a declared purpose. Non-'evaluation' throws. */
-  get(scenarioId, purpose) {
-    if (purpose !== "evaluation" && purpose !== "debugging") {
-      throw new Error(
-        `HoldoutAuditor.get: purpose must be 'evaluation' or 'debugging', got ${purpose}`
-      );
-    }
-    const s = this.scenarios.find((x) => x.id === scenarioId);
-    if (!s) throw new Error(`holdout scenario "${scenarioId}" not found`);
-    this.accessLog.push({ scenarioId, purpose, at: Date.now() });
-    return s;
-  }
-  getAccessLog() {
-    return this.accessLog;
-  }
-};
-function excerpt2(source, needle) {
-  const at = source.indexOf(needle);
-  if (at < 0) return "";
-  const start = Math.max(0, at - 30);
-  const end = Math.min(source.length, at + needle.length + 30);
-  return (start > 0 ? "\u2026" : "") + source.slice(start, end) + (end < source.length ? "\u2026" : "");
-}
-
 // src/observability.ts
 async function toLangfuseEnvelope(store, runId) {
   const run = await store.getRun(runId);