@tangle-network/agent-eval 1.0.13 → 1.0.15
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CHANGELOG.md +10 -0
- package/dist/analyst/index.js +1 -1
- package/dist/cli.js +2 -2
- package/dist/contract/index.js +1 -1
- package/dist/{default-registry-eCkZFfuF.js → default-registry-BB9_jzyH.js} +2 -2
- package/dist/{default-registry-eCkZFfuF.js.map → default-registry-BB9_jzyH.js.map} +1 -1
- package/dist/diagnosis/index.d.ts.map +1 -1
- package/dist/diagnosis/index.js +19 -13
- package/dist/diagnosis/index.js.map +1 -1
- package/dist/{exporters-ByLcILYU.js → exporters-BKZLt1Px.js} +2 -2
- package/dist/{exporters-ByLcILYU.js.map → exporters-BKZLt1Px.js.map} +1 -1
- package/dist/{hf-dataset-DPVQXW5n.js → hf-dataset-Bv114tzB.js} +11 -11
- package/dist/{hf-dataset-DPVQXW5n.js.map → hf-dataset-Bv114tzB.js.map} +1 -1
- package/dist/index-f09MX_H_.d.ts.map +1 -1
- package/dist/index.d.ts.map +1 -1
- package/dist/index.js +6 -9
- package/dist/index.js.map +1 -1
- package/dist/{integrity-Vx5xgn3s.js → integrity-BHM4voOx.js} +3 -3
- package/dist/{integrity-Vx5xgn3s.js.map → integrity-BHM4voOx.js.map} +1 -1
- package/dist/{mint-BH_3Ak2x.js → mint-yF51iBgY.js} +2 -2
- package/dist/{mint-BH_3Ak2x.js.map → mint-yF51iBgY.js.map} +1 -1
- package/dist/multishot/golden/index.d.ts.map +1 -1
- package/dist/multishot/golden/index.js +3 -5
- package/dist/multishot/golden/index.js.map +1 -1
- package/dist/openapi.json +1 -1
- package/dist/{render-BVPfP7dc.js → render-CsHLoMUl.js} +70 -47
- package/dist/render-CsHLoMUl.js.map +1 -0
- package/dist/rl.js +3 -3
- package/dist/rollout/index.js +5 -5
- package/dist/{rollout-CbOGxuHl.js → rollout-Cu7kHGIe.js} +5 -5
- package/dist/{rollout-CbOGxuHl.js.map → rollout-Cu7kHGIe.js.map} +1 -1
- package/dist/{rollout-nodes-DKZ3vAYY.js → rollout-nodes-DhSPJ2X7.js} +4 -4
- package/dist/rollout-nodes-DhSPJ2X7.js.map +1 -0
- package/dist/{schema-YUIXT8iW.js → schema-ChIZfnqY.js} +37 -5
- package/dist/schema-ChIZfnqY.js.map +1 -0
- package/dist/supervisor-run/index.d.ts +1 -0
- package/dist/supervisor-run/index.d.ts.map +1 -1
- package/dist/supervisor-run/index.js +3 -3
- package/dist/trace-repair/index.d.ts.map +1 -1
- package/dist/trace-repair/index.js +2 -5
- package/dist/trace-repair/index.js.map +1 -1
- package/package.json +1 -1
- package/dist/render-BVPfP7dc.js.map +0 -1
- package/dist/rollout-nodes-DKZ3vAYY.js.map +0 -1
- package/dist/schema-YUIXT8iW.js.map +0 -1
package/dist/index.js
CHANGED
|
@@ -36,7 +36,7 @@ import { a as LlmCallError, i as stripFencedJson, n as isRecord, o as LlmRespons
|
|
|
36
36
|
import { r as selfImprove, t as defineAgentEval } from "./define-agent-eval-CjF2nTA3.js";
|
|
37
37
|
import { n as equivalenceVerdict, t as certificationEvidenceDigest } from "./verdict-C8TcdwfD.js";
|
|
38
38
|
import { c as createBoundedTraceAnalysisStore, p as DEFAULT_TRACE_ANALYST_BUDGETS, t as createTraceAnalyst } from "./kind-factory-Cfz8DUCc.js";
|
|
39
|
-
import { C as assertUniqueFindingIds, D as validateAnalystReviewDecisions, E as snapshotAnalystRun, S as analystRunDigest, T as readAnalystReview, c as FAILURE_MODE_KIND_SPEC, i as DEFAULT_TRACE_ANALYST_KINDS, n as AnalystRegistry, t as buildDefaultAnalystRegistry, w as completedAnalystReviewQuality, x as analystFindingDigest } from "./default-registry-
|
|
39
|
+
import { C as assertUniqueFindingIds, D as validateAnalystReviewDecisions, E as snapshotAnalystRun, S as analystRunDigest, T as readAnalystReview, c as FAILURE_MODE_KIND_SPEC, i as DEFAULT_TRACE_ANALYST_KINDS, n as AnalystRegistry, t as buildDefaultAnalystRegistry, w as completedAnalystReviewQuality, x as analystFindingDigest } from "./default-registry-BB9_jzyH.js";
|
|
40
40
|
import { OUTPUT_VALUE } from "./trace-attributes.js";
|
|
41
41
|
import { n as extractUsageFromSse, t as extractUsage } from "./extract-usage-ISFqhfDz.js";
|
|
42
42
|
import { n as InMemoryRawProviderSink, r as NoopRawProviderSink, t as FileSystemRawProviderSink } from "./raw-provider-sink-RbaEv2vB.js";
|
|
@@ -51,8 +51,8 @@ import { n as runBoundedProcess, t as DEFAULT_BOUNDED_PROCESS_TIMEOUT_MS } from
|
|
|
51
51
|
import { t as runEvalCampaign } from "./eval-campaign-CZSCjxVb.js";
|
|
52
52
|
import { _ as assertCrossFamily, a as maximumChargeForLlmRequest, d as assertServedModel, g as CrossFamilyError, h as describeServedModelCheck, i as costReceiptFromLlmError, l as ModelSubstitutionError, m as checkServedModel, p as assertServedModels, r as costReceiptFromLlm, s as isTransientLlmError, u as assertCrossFamilyServed, v as judgeFamily } from "./llm-client-BdSiZ4qz.js";
|
|
53
53
|
import { n as createLlmCorrectnessChecker, r as verifyCompletion, t as completionVerdict } from "./completion-verifier-ClmJV_Yc.js";
|
|
54
|
-
import "./rollout-
|
|
55
|
-
import { t as mintRolloutRows } from "./mint-
|
|
54
|
+
import "./rollout-Cu7kHGIe.js";
|
|
55
|
+
import { t as mintRolloutRows } from "./mint-yF51iBgY.js";
|
|
56
56
|
import { n as pairedEvalueSequence, t as evaluateInterimReleaseConfidence } from "./sequential-Bjs5roiK.js";
|
|
57
57
|
import { n as paidJsonChat } from "./chat-json-call-C473CwTt.js";
|
|
58
58
|
import { t as createDspyRlmTraceEngine } from "./dspy-rlm-engine-_MEe2jIO.js";
|
|
@@ -5952,7 +5952,7 @@ const productBenchmarkSplits = [
|
|
|
5952
5952
|
"safety",
|
|
5953
5953
|
"sentinel"
|
|
5954
5954
|
];
|
|
5955
|
-
function isObject
|
|
5955
|
+
function isObject(value) {
|
|
5956
5956
|
return Boolean(value) && typeof value === "object" && !Array.isArray(value);
|
|
5957
5957
|
}
|
|
5958
5958
|
function fail(path, message) {
|
|
@@ -5963,7 +5963,7 @@ function wrapValidationError(path, err) {
|
|
|
5963
5963
|
throw new ValidationError(`${path}: ${err instanceof Error ? err.message : String(err)}`, { cause: err });
|
|
5964
5964
|
}
|
|
5965
5965
|
function expectObject(value, path) {
|
|
5966
|
-
if (!isObject
|
|
5966
|
+
if (!isObject(value)) fail(path, "must be an object");
|
|
5967
5967
|
return value;
|
|
5968
5968
|
}
|
|
5969
5969
|
function expectString(value, path) {
|
|
@@ -7936,7 +7936,7 @@ function jsonShape(expected) {
|
|
|
7936
7936
|
id,
|
|
7937
7937
|
check(obs) {
|
|
7938
7938
|
const json = obs.json;
|
|
7939
|
-
if (!
|
|
7939
|
+
if (!isRecord(json)) return {
|
|
7940
7940
|
id,
|
|
7941
7941
|
pass: false,
|
|
7942
7942
|
detail: "observation.json missing or not an object"
|
|
@@ -7995,9 +7995,6 @@ function evaluateOracles(obs, oracles) {
|
|
|
7995
7995
|
}
|
|
7996
7996
|
};
|
|
7997
7997
|
}
|
|
7998
|
-
function isObject(v) {
|
|
7999
|
-
return typeof v === "object" && v !== null && !Array.isArray(v);
|
|
8000
|
-
}
|
|
8001
7998
|
// src/produced-state.ts
|
|
8002
7999
|
function artifactKind(mimeType) {
|
|
8003
8000
|
if (!mimeType) return "file";
|