@tangle-network/agent-eval 0.122.1 → 0.122.3

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.
Files changed (44) hide show
  1. package/CHANGELOG.md +6 -0
  2. package/dist/analyst/index.js +4 -5
  3. package/dist/analyst/index.js.map +1 -1
  4. package/dist/benchmarks/index.js +6 -7
  5. package/dist/campaign/index.d.ts +25 -3
  6. package/dist/campaign/index.js +5 -6
  7. package/dist/{chunk-3A246TSA.js → chunk-BD2OK6ZW.js} +2 -2
  8. package/dist/{chunk-VZ6VKOJT.js → chunk-DHJJACEX.js} +90 -38
  9. package/dist/{chunk-VZ6VKOJT.js.map → chunk-DHJJACEX.js.map} +1 -1
  10. package/dist/{chunk-S5TT5R3L.js → chunk-EAWKAVID.js} +75 -2
  11. package/dist/chunk-EAWKAVID.js.map +1 -0
  12. package/dist/{chunk-TDHX3BVA.js → chunk-IDC74VDL.js} +62 -47
  13. package/dist/chunk-IDC74VDL.js.map +1 -0
  14. package/dist/{chunk-XDIRG3TO.js → chunk-N3QPYIVG.js} +201 -103
  15. package/dist/chunk-N3QPYIVG.js.map +1 -0
  16. package/dist/{chunk-4I2E3LLO.js → chunk-NMN4WGSJ.js} +2 -2
  17. package/dist/{chunk-PICTDURQ.js → chunk-OZFTLMGS.js} +2 -2
  18. package/dist/{chunk-QWMPPZ3X.js → chunk-SFXDMFGV.js} +2 -2
  19. package/dist/{chunk-P62I523A.js → chunk-ULW7AATT.js} +4 -6
  20. package/dist/{chunk-P62I523A.js.map → chunk-ULW7AATT.js.map} +1 -1
  21. package/dist/{chunk-JHOJHHU7.js → chunk-ZMXDQ4K7.js} +8 -5
  22. package/dist/{chunk-JHOJHHU7.js.map → chunk-ZMXDQ4K7.js.map} +1 -1
  23. package/dist/cli.js +2 -2
  24. package/dist/contract/index.d.ts +18 -0
  25. package/dist/contract/index.js +6 -6
  26. package/dist/fuzz.js +1 -1
  27. package/dist/index.d.ts +7 -1
  28. package/dist/index.js +12 -12
  29. package/dist/index.js.map +1 -1
  30. package/dist/openapi.json +1 -1
  31. package/dist/{run-campaign-HNFPJET4.js → run-campaign-2YJ3Z6JS.js} +3 -3
  32. package/dist/wire/index.js +2 -2
  33. package/docs/eval-surface-map.md +1 -1
  34. package/package.json +1 -1
  35. package/dist/chunk-3YYRZDON.js +0 -45
  36. package/dist/chunk-3YYRZDON.js.map +0 -1
  37. package/dist/chunk-S5TT5R3L.js.map +0 -1
  38. package/dist/chunk-TDHX3BVA.js.map +0 -1
  39. package/dist/chunk-XDIRG3TO.js.map +0 -1
  40. /package/dist/{chunk-3A246TSA.js.map → chunk-BD2OK6ZW.js.map} +0 -0
  41. /package/dist/{chunk-4I2E3LLO.js.map → chunk-NMN4WGSJ.js.map} +0 -0
  42. /package/dist/{chunk-PICTDURQ.js.map → chunk-OZFTLMGS.js.map} +0 -0
  43. /package/dist/{chunk-QWMPPZ3X.js.map → chunk-SFXDMFGV.js.map} +0 -0
  44. /package/dist/{run-campaign-HNFPJET4.js.map → run-campaign-2YJ3Z6JS.js.map} +0 -0
package/dist/index.js CHANGED
@@ -9,12 +9,12 @@ import {
9
9
  checkBehavioralCanary,
10
10
  checkCanaries,
11
11
  runBehavioralCanaries
12
- } from "./chunk-3A246TSA.js";
12
+ } from "./chunk-BD2OK6ZW.js";
13
13
  import {
14
14
  BENCHMARK_SPLIT_SEED,
15
15
  benchmarks_exports,
16
16
  deterministicSplit
17
- } from "./chunk-PICTDURQ.js";
17
+ } from "./chunk-OZFTLMGS.js";
18
18
  import {
19
19
  DEFAULT_RULES,
20
20
  buildTrajectory,
@@ -53,7 +53,7 @@ import {
53
53
  pairArms,
54
54
  parseCorrectnessResponse,
55
55
  verifyCompletion
56
- } from "./chunk-TDHX3BVA.js";
56
+ } from "./chunk-IDC74VDL.js";
57
57
  import {
58
58
  DEFAULT_MUTATION_PRIMITIVES,
59
59
  DEFAULT_RED_TEAM_CORPUS,
@@ -91,7 +91,7 @@ import {
91
91
  scoreRedTeamOutput,
92
92
  surfaceContentHash,
93
93
  toolNamesForRun
94
- } from "./chunk-VZ6VKOJT.js";
94
+ } from "./chunk-DHJJACEX.js";
95
95
  import {
96
96
  BackendIntegrityError,
97
97
  assertRealAgentReceipts,
@@ -103,7 +103,7 @@ import {
103
103
  inMemoryVerdictCache,
104
104
  summarizeAgentReceiptIntegrity,
105
105
  summarizeBackendIntegrity
106
- } from "./chunk-XDIRG3TO.js";
106
+ } from "./chunk-N3QPYIVG.js";
107
107
  import {
108
108
  DEFAULT_COMPLEXITY_WEIGHTS,
109
109
  FindingsStore,
@@ -116,16 +116,13 @@ import {
116
116
  defaultIsMaterial,
117
117
  diffFindings,
118
118
  runSemanticConceptJudge
119
- } from "./chunk-P62I523A.js";
119
+ } from "./chunk-ULW7AATT.js";
120
120
  import {
121
121
  buildDefaultAnalystRegistry,
122
122
  computeTraceMetrics,
123
123
  createChatClient
124
- } from "./chunk-QWMPPZ3X.js";
124
+ } from "./chunk-SFXDMFGV.js";
125
125
  import "./chunk-HHWE3POT.js";
126
- import {
127
- Mutex
128
- } from "./chunk-3YYRZDON.js";
129
126
  import {
130
127
  AnalystRegistry,
131
128
  DEFAULT_RUN_SCORE_WEIGHTS,
@@ -134,6 +131,7 @@ import {
134
131
  IMPROVEMENT_KIND_SPEC,
135
132
  KNOWLEDGE_GAP_KIND_SPEC,
136
133
  KNOWLEDGE_POISONING_KIND_SPEC,
134
+ Mutex,
137
135
  POLICY_EDIT_AXES,
138
136
  POLICY_EDIT_CANDIDATE_RECORD_SCHEMA,
139
137
  POLICY_EDIT_TARGET_SURFACES,
@@ -150,6 +148,7 @@ import {
150
148
  makeFinding,
151
149
  makePolicyEdit,
152
150
  makePolicyEditCandidateRecord,
151
+ mapConcurrent,
153
152
  policyEditFromFinding,
154
153
  policyEditsFromFindings,
155
154
  renderPriorFindings,
@@ -157,7 +156,7 @@ import {
157
156
  scorePolicyEditReadiness,
158
157
  validatePolicyEdit,
159
158
  validatePolicyEditCandidateRecord
160
- } from "./chunk-S5TT5R3L.js";
159
+ } from "./chunk-EAWKAVID.js";
161
160
  import {
162
161
  allCriticalPassed,
163
162
  controlFailureClassFromVerification,
@@ -267,7 +266,7 @@ import {
267
266
  CostReservationExceededError,
268
267
  costForUsage,
269
268
  modelPriceKey
270
- } from "./chunk-JHOJHHU7.js";
269
+ } from "./chunk-ZMXDQ4K7.js";
271
270
  import {
272
271
  MODEL_PRICING,
273
272
  MetricsCollector,
@@ -11959,6 +11958,7 @@ export {
11959
11958
  makePolicyEdit,
11960
11959
  makePolicyEditCandidateRecord,
11961
11960
  mannWhitneyU,
11961
+ mapConcurrent,
11962
11962
  matchGoldens,
11963
11963
  matchSpan,
11964
11964
  maximumChargeForLlmRequest,