@tangle-network/agent-eval 0.122.3 → 0.122.5

package/dist/{chunk-BD2OK6ZW.js → chunk-JUOZI4L3.js}

@@ -1,6 +1,6 @@
 import {
   summarizeBackendIntegrity
-} from "./chunk-N3QPYIVG.js";
+} from "./chunk-FZ26ORA6.js";
 import {
   paretoChart
 } from "./chunk-DPZAEKA6.js";
@@ -995,4 +995,4 @@ export {
   summarizeExecution,
   analyzeRuns
 };
-//# sourceMappingURL=chunk-BD2OK6ZW.js.map
+//# sourceMappingURL=chunk-JUOZI4L3.js.map

package/dist/{chunk-OZFTLMGS.js → chunk-WULRYLNC.js}

@@ -1,7 +1,7 @@
 import {
   fsCampaignStorage,
   runCampaign
-} from "./chunk-N3QPYIVG.js";
+} from "./chunk-FZ26ORA6.js";
 import {
   __export
 } from "./chunk-PZ5AY32C.js";
@@ -763,4 +763,4 @@ export {
   retrievalMetricsAtCutoff,
   benchmarks_exports
 };
-//# sourceMappingURL=chunk-OZFTLMGS.js.map
+//# sourceMappingURL=chunk-WULRYLNC.js.map

package/dist/contract/index.d.ts

@@ -1386,6 +1386,11 @@ interface Scenario {
     id: string;
     kind: string;
     tags?: string[];
+    /**
+     * Variants with the same non-empty value receive the same seed for a given
+     * replicate. Leave unset when every scenario should have an independent seed.
+     */
+    seedGroup?: string;
 }
 /** Redacted identity of a complete scenario payload retained in campaign results. */
 interface CampaignScenarioIdentity extends Pick<Scenario, 'id' | 'kind'> {

package/dist/contract/index.js

@@ -14,7 +14,7 @@ import {
 import {
   analyzeRuns,
   summarizeExecution
-} from "../chunk-BD2OK6ZW.js";
+} from "../chunk-JUOZI4L3.js";
 import {
   REFERENCE_EQUIVALENCE_INPUT_LIMITS,
   REFERENCE_EQUIVALENCE_JUDGE_VERSION,
@@ -36,7 +36,7 @@ import {
   runReferenceEquivalenceJudge,
   surfaceContentHash,
   surfaceHash
-} from "../chunk-DHJJACEX.js";
+} from "../chunk-FLCXOTWF.js";
 import {
   campaignSplitDigest,
   createRunCostLedger,
@@ -44,7 +44,7 @@ import {
   inMemoryCampaignStorage,
   resolveRunDir,
   runCampaign
-} from "../chunk-N3QPYIVG.js";
+} from "../chunk-FZ26ORA6.js";
 import {
   buildDefaultAnalystRegistry,
   createChatClient

package/dist/index.d.ts

@@ -7520,6 +7520,11 @@ interface Scenario {
     id: string;
     kind: string;
     tags?: string[];
+    /**
+     * Variants with the same non-empty value receive the same seed for a given
+     * replicate. Leave unset when every scenario should have an independent seed.
+     */
+    seedGroup?: string;
 }
 /** Redacted identity of a complete scenario payload retained in campaign results. */
 interface CampaignScenarioIdentity extends Pick<Scenario, 'id' | 'kind'> {
@@ -14081,6 +14086,35 @@ declare namespace index$2 {
   export { index$2_ROUTING_DATASET as ROUTING_DATASET, index$2_RoutingAdapter as RoutingAdapter, type index$2_RoutingDatasetItem as RoutingDatasetItem, type index$2_RoutingItem as RoutingItem, type index$2_RoutingPayload as RoutingPayload, index$2_assignSplit as assignSplit, index$2_evaluate as evaluate, index$2_extractRouteTokens as extractRouteTokens, index$2_loadDataset as loadDataset };
 }
 
+/**
+ * `CampaignStorage` — the filesystem seam `runCampaign` writes through
+ * (run/cell dirs, the resumability cache, per-cell artifacts, trace spans).
+ *
+ * The default (`fsCampaignStorage`) is the Node filesystem — identical
+ * behavior to the inline `node:fs` calls it replaces, so existing CLI
+ * consumers are unaffected. `inMemoryCampaignStorage` keeps everything in a
+ * `Map`, so the substrate runs in environments WITHOUT a filesystem
+ * (Cloudflare Workers, Deno Deploy, other edge runtimes) — the campaign
+ * still produces its `CampaignResult` (cells + aggregates) in memory;
+ * artifacts/traces simply aren't persisted to disk.
+ *
+ * Paths are opaque keys to the in-memory adapter — it does not parse them,
+ * so the same `join(...)`-built paths work unchanged across both adapters.
+ */
+interface CampaignStorage {
+    /** Ensure a directory exists (recursive). No-op for in-memory. */
+    ensureDir(dir: string): void;
+    /** Does this path exist (as a written file or an ensured dir)? */
+    exists(path: string): boolean;
+    /** Read a UTF-8 file; `undefined` when missing or unreadable. */
+    read(path: string): string | undefined;
+    /** Write a file (string or bytes). Parent dir is assumed ensured. */
+    write(path: string, content: string | Uint8Array): void;
+    /** Append only when the current UTF-8 byte length matches `expectedBytes`.
+     * Returns the new length, or undefined when another writer won. */
+    append?(path: string, content: string, expectedBytes: number): number | undefined;
+}
+
 /**
  * `openAutoPr` — thin shell-out helper for the `runImprovementLoop` preset's
  * `autoOnPromote: 'pr'` mode. Substitutes for the per-product PR-opening
@@ -14130,35 +14164,6 @@ interface OpenAutoPrResult {
  */
 declare function openAutoPr<TArtifact, TScenario extends Scenario>(options: OpenAutoPrOptions<TArtifact, TScenario>): OpenAutoPrResult;
 
-/**
- * `CampaignStorage` — the filesystem seam `runCampaign` writes through
- * (run/cell dirs, the resumability cache, per-cell artifacts, trace spans).
- *
- * The default (`fsCampaignStorage`) is the Node filesystem — identical
- * behavior to the inline `node:fs` calls it replaces, so existing CLI
- * consumers are unaffected. `inMemoryCampaignStorage` keeps everything in a
- * `Map`, so the substrate runs in environments WITHOUT a filesystem
- * (Cloudflare Workers, Deno Deploy, other edge runtimes) — the campaign
- * still produces its `CampaignResult` (cells + aggregates) in memory;
- * artifacts/traces simply aren't persisted to disk.
- *
- * Paths are opaque keys to the in-memory adapter — it does not parse them,
- * so the same `join(...)`-built paths work unchanged across both adapters.
- */
-interface CampaignStorage {
-    /** Ensure a directory exists (recursive). No-op for in-memory. */
-    ensureDir(dir: string): void;
-    /** Does this path exist (as a written file or an ensured dir)? */
-    exists(path: string): boolean;
-    /** Read a UTF-8 file; `undefined` when missing or unreadable. */
-    read(path: string): string | undefined;
-    /** Write a file (string or bytes). Parent dir is assumed ensured. */
-    write(path: string, content: string | Uint8Array): void;
-    /** Append only when the current UTF-8 byte length matches `expectedBytes`.
-     * Returns the new length, or undefined when another writer won. */
-    append?(path: string, content: string, expectedBytes: number): number | undefined;
-}
-
 /**
  * `runOptimization` — the improvement loop body. Runs N generations: the
  * `SurfaceProposer` proposes K candidate surfaces per generation, each

package/dist/index.js

@@ -9,12 +9,12 @@ import {
   checkBehavioralCanary,
   checkCanaries,
   runBehavioralCanaries
-} from "./chunk-BD2OK6ZW.js";
+} from "./chunk-JUOZI4L3.js";
 import {
   BENCHMARK_SPLIT_SEED,
   benchmarks_exports,
   deterministicSplit
-} from "./chunk-OZFTLMGS.js";
+} from "./chunk-WULRYLNC.js";
 import {
   DEFAULT_RULES,
   buildTrajectory,
@@ -53,7 +53,7 @@ import {
   pairArms,
   parseCorrectnessResponse,
   verifyCompletion
-} from "./chunk-IDC74VDL.js";
+} from "./chunk-GLUBGL77.js";
 import {
   DEFAULT_MUTATION_PRIMITIVES,
   DEFAULT_RED_TEAM_CORPUS,
@@ -91,7 +91,7 @@ import {
   scoreRedTeamOutput,
   surfaceContentHash,
   toolNamesForRun
-} from "./chunk-DHJJACEX.js";
+} from "./chunk-FLCXOTWF.js";
 import {
   BackendIntegrityError,
   assertRealAgentReceipts,
@@ -103,7 +103,7 @@ import {
   inMemoryVerdictCache,
   summarizeAgentReceiptIntegrity,
   summarizeBackendIntegrity
-} from "./chunk-N3QPYIVG.js";
+} from "./chunk-FZ26ORA6.js";
 import {
   DEFAULT_COMPLEXITY_WEIGHTS,
   FindingsStore,

package/dist/openapi.json

@@ -2,7 +2,7 @@
   "openapi": "3.1.0",
   "info": {
     "title": "@tangle-network/agent-eval — wire protocol",
-    "version": "0.122.3",
+    "version": "0.122.5",
     "description": "HTTP and stdio RPC interface to agent-eval. The TypeScript runtime is the source of truth; this spec is the contract that cross-language clients (Python, Rust, Go) generate from.\n\nWire-protocol version: 1.0.0. Bumps on breaking changes to request/response schemas.",
     "contact": {
       "name": "Tangle Network",

package/dist/rl.d.ts

@@ -3569,6 +3569,11 @@ interface Scenario {
     id: string;
     kind: string;
     tags?: string[];
+    /**
+     * Variants with the same non-empty value receive the same seed for a given
+     * replicate. Leave unset when every scenario should have an independent seed.
+     */
+    seedGroup?: string;
 }
 /** Redacted identity of a complete scenario payload retained in campaign results. */
 interface CampaignScenarioIdentity extends Pick<Scenario, 'id' | 'kind'> {

package/dist/{run-campaign-2YJ3Z6JS.js → run-campaign-56EB2UN5.js}

@@ -1,7 +1,7 @@
 import {
   planCampaignRun,
   runCampaign
-} from "./chunk-N3QPYIVG.js";
+} from "./chunk-FZ26ORA6.js";
 import "./chunk-PJQFMIOX.js";
 import "./chunk-ZMXDQ4K7.js";
 import "./chunk-VI2UW6B6.js";
@@ -11,4 +11,4 @@ export {
   planCampaignRun,
   runCampaign
 };
-//# sourceMappingURL=run-campaign-2YJ3Z6JS.js.map
+//# sourceMappingURL=run-campaign-56EB2UN5.js.map

package/docs/eval-surface-map.md

@@ -15,6 +15,9 @@ composition point already exists (see Produced-state grading below).**
 | `runOptimization` | GENERATE: measured or validated premeasured baseline → N generations of propose → measure → rank → promote. No release gate. | generations + winner |
 | `runImprovementLoop` | The release-gate shell around `runOptimization`: adds a held-out re-score + a promotion gate (+ optional auto-PR). | gate decision + winner |
 | `runEvalCampaign` | Inversion-of-control variant of `runCampaign` — the runner is handed a pre-wired trace/sink/emitter and integrity gating as a precondition. Use when you need full capture by construction. | `CampaignResult` + records |
+
+When variants of the same task run inside one `runCampaign`, give those scenarios the same `seedGroup` so each repetition uses common randomness.
+Use `runProfileMatrix` instead when profiles are separate campaign axes.
 | `runAgentMatrix` | The bare N-axis cartesian scheduler with concurrency control. The layer beneath the eval surface — reach for it only when you need raw scheduling, not eval semantics. | cell results |
 
 Mental model: **measure** (`runCampaign`/`runEval`) → **factor** (`runProfileMatrix`) →

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tangle-network/agent-eval",
-  "version": "0.122.3",
+  "version": "0.122.5",
   "description": "Evaluate and improve AI agents from runs, traces, judges, and feedback. Compare candidates, cluster failures, measure lift, and gate releases.",
   "homepage": "https://github.com/tangle-network/agent-eval#readme",
   "repository": {