@tangle-network/agent-eval 0.123.2 → 0.123.3

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 (65) hide show
  1. package/CHANGELOG.md +24 -0
  2. package/README.md +48 -4
  3. package/dist/analyst/index.d.ts +18 -3
  4. package/dist/analyst/index.js +5 -5
  5. package/dist/benchmarks/index.d.ts +10 -3
  6. package/dist/benchmarks/index.js +7 -7
  7. package/dist/campaign/index.d.ts +241 -221
  8. package/dist/campaign/index.js +20 -16
  9. package/dist/{chunk-RQP5UTK5.js → chunk-22VO7T2I.js} +689 -356
  10. package/dist/chunk-22VO7T2I.js.map +1 -0
  11. package/dist/{chunk-EBDOTTZJ.js → chunk-4KO6D2BO.js} +3 -3
  12. package/dist/{chunk-3FCG7FBV.js → chunk-D5JZ7UDZ.js} +2 -2
  13. package/dist/{chunk-BGVTIE2C.js → chunk-GRCDRKII.js} +13 -2
  14. package/dist/chunk-GRCDRKII.js.map +1 -0
  15. package/dist/{chunk-FHFTYX2Q.js → chunk-JKDNAOF5.js} +2 -2
  16. package/dist/{chunk-WXQTVEKM.js → chunk-MHPEGJHC.js} +3 -3
  17. package/dist/{chunk-J7S4YM27.js → chunk-MK3Q6GCL.js} +28 -14
  18. package/dist/chunk-MK3Q6GCL.js.map +1 -0
  19. package/dist/{chunk-DNVPOYUS.js → chunk-N7ZS6PEF.js} +2 -2
  20. package/dist/{chunk-4SOQ4ND2.js → chunk-QFQZ3U3X.js} +1 -1
  21. package/dist/chunk-QFQZ3U3X.js.map +1 -0
  22. package/dist/{chunk-WMJR67FX.js → chunk-QNVBALEX.js} +4 -4
  23. package/dist/{chunk-E3HAD4A3.js → chunk-SJT4OBVL.js} +2 -2
  24. package/dist/{chunk-HQY7LBV2.js → chunk-SUN7QLPB.js} +6 -6
  25. package/dist/{chunk-HQY7LBV2.js.map → chunk-SUN7QLPB.js.map} +1 -1
  26. package/dist/{chunk-5YMKIFYP.js → chunk-WP3GEXKI.js} +3 -3
  27. package/dist/cli.js +3 -3
  28. package/dist/contract/index.d.ts +18 -3
  29. package/dist/contract/index.js +7 -7
  30. package/dist/fuzz.d.ts +8 -1
  31. package/dist/fuzz.js +4 -4
  32. package/dist/index.d.ts +27 -9
  33. package/dist/index.js +14 -12
  34. package/dist/index.js.map +1 -1
  35. package/dist/multishot/index.d.ts +2 -2
  36. package/dist/openapi.json +1 -1
  37. package/dist/rl.d.ts +15 -2
  38. package/dist/rl.js +4 -2
  39. package/dist/rl.js.map +1 -1
  40. package/dist/{run-campaign-ZKR5MQMQ.js → run-campaign-I3JXKVAK.js} +3 -3
  41. package/dist/traces.d.ts +2 -2
  42. package/dist/traces.js +1 -1
  43. package/dist/wire/index.d.ts +16 -1
  44. package/dist/wire/index.js +3 -3
  45. package/docs/campaign-proposers.md +142 -131
  46. package/docs/concepts.md +30 -34
  47. package/docs/design/loop-taxonomy.md +2 -3
  48. package/docs/design.md +0 -1
  49. package/docs/multi-shot-optimization.md +41 -49
  50. package/package.json +1 -1
  51. package/dist/chunk-4SOQ4ND2.js.map +0 -1
  52. package/dist/chunk-BGVTIE2C.js.map +0 -1
  53. package/dist/chunk-J7S4YM27.js.map +0 -1
  54. package/dist/chunk-RQP5UTK5.js.map +0 -1
  55. package/docs/improvement-glossary.md +0 -204
  56. package/docs/self-improvement-map.md +0 -115
  57. /package/dist/{chunk-EBDOTTZJ.js.map → chunk-4KO6D2BO.js.map} +0 -0
  58. /package/dist/{chunk-3FCG7FBV.js.map → chunk-D5JZ7UDZ.js.map} +0 -0
  59. /package/dist/{chunk-FHFTYX2Q.js.map → chunk-JKDNAOF5.js.map} +0 -0
  60. /package/dist/{chunk-WXQTVEKM.js.map → chunk-MHPEGJHC.js.map} +0 -0
  61. /package/dist/{chunk-DNVPOYUS.js.map → chunk-N7ZS6PEF.js.map} +0 -0
  62. /package/dist/{chunk-WMJR67FX.js.map → chunk-QNVBALEX.js.map} +0 -0
  63. /package/dist/{chunk-E3HAD4A3.js.map → chunk-SJT4OBVL.js.map} +0 -0
  64. /package/dist/{chunk-5YMKIFYP.js.map → chunk-WP3GEXKI.js.map} +0 -0
  65. /package/dist/{run-campaign-ZKR5MQMQ.js.map → run-campaign-I3JXKVAK.js.map} +0 -0
package/dist/campaign/index.js CHANGED
@@ -12,22 +12,24 @@ import {
12
12
  aceProposer,
13
13
  acquireSingleRunLock,
14
14
  analyzeCrossSurfaceInteractions,
15
+ analyzeOtlpTraceFile,
15
16
  applySkillPatch,
16
17
  assertPolicyEditAuthorContextBudget,
17
18
  buildAnalystSurfaceDispatch,
18
19
  callbackGovernor,
19
20
  campaignLineageStore,
20
21
  classifyUngroundedLiterals,
21
- compareProposers,
22
+ compareOptimizationMethods,
22
23
  compositeProposer,
24
+ costFromLedgerSummary,
23
25
  discoverEvalFixtures,
24
26
  extractFapoAttributionSignals,
25
27
  failureModeRecallJudge,
26
- fapoEscalationEntry,
28
+ fapoEscalationMethod,
27
29
  fapoProposer,
28
30
  fsLineageStore,
29
- gepaParetoEntry,
30
- gepaReflectionEntry,
31
+ gepaParetoMethod,
32
+ gepaReflectionMethod,
31
33
  gitWorktreeAdapter,
32
34
  haloProposer,
33
35
  heuristicGovernor,
@@ -62,13 +64,13 @@ import {
62
64
  selectPolicyEditAuthorRows,
63
65
  sequentialDecide,
64
66
  sequentialPairedGate,
65
- skillOptEntry,
67
+ skillOptMethod,
66
68
  skillOptProposer,
67
69
  traceAnalystProposer,
68
70
  userStoryScoreboard,
69
71
  validateSearchLedgerEvent,
70
72
  verifyCodeSurface
71
- } from "../chunk-RQP5UTK5.js";
73
+ } from "../chunk-22VO7T2I.js";
72
74
  import {
73
75
  assertCodeSurfaceIdentity,
74
76
  buildEvidenceVector,
@@ -109,7 +111,7 @@ import {
109
111
  surfaceContentHash,
110
112
  surfaceHash,
111
113
  verifyLoopProvenanceRecord
112
- } from "../chunk-HQY7LBV2.js";
114
+ } from "../chunk-SUN7QLPB.js";
113
115
  import {
114
116
  SearchLedgerConflictError,
115
117
  SearchLedgerError,
@@ -126,15 +128,15 @@ import {
126
128
  resolveRunDir,
127
129
  runCampaign,
128
130
  tangleTracesRoot
129
- } from "../chunk-3FCG7FBV.js";
131
+ } from "../chunk-D5JZ7UDZ.js";
130
132
  import {
131
133
  POLICY_EDIT_CANDIDATE_RECORD_SCHEMA,
132
134
  validatePolicyEditCandidateRecord
133
- } from "../chunk-WXQTVEKM.js";
135
+ } from "../chunk-MHPEGJHC.js";
134
136
  import "../chunk-ARU2PZFM.js";
135
- import "../chunk-J7S4YM27.js";
137
+ import "../chunk-MK3Q6GCL.js";
136
138
  import "../chunk-PJQFMIOX.js";
137
- import "../chunk-BGVTIE2C.js";
139
+ import "../chunk-GRCDRKII.js";
138
140
  import "../chunk-VI2UW6B6.js";
139
141
  import "../chunk-NUKSVU3W.js";
140
142
  import "../chunk-GGE4NNQT.js";
@@ -165,6 +167,7 @@ export {
165
167
  aceProposer,
166
168
  acquireSingleRunLock,
167
169
  analyzeCrossSurfaceInteractions,
170
+ analyzeOtlpTraceFile,
168
171
  applySkillPatch,
169
172
  assertCampaignDesign,
170
173
  assertCampaignSplitIdentity,
@@ -184,9 +187,10 @@ export {
184
187
  canonicalDigest,
185
188
  classifyUngroundedLiterals,
186
189
  codeSurfaceIdentityMaterial,
187
- compareProposers,
190
+ compareOptimizationMethods,
188
191
  composeGate,
189
192
  compositeProposer,
193
+ costFromLedgerSummary,
190
194
  countSentenceEdits,
191
195
  createReferenceEquivalenceJudge,
192
196
  createRunCostLedger,
@@ -199,13 +203,13 @@ export {
199
203
  extractFapoAttributionSignals,
200
204
  extractH2Sections,
201
205
  failureModeRecallJudge,
202
- fapoEscalationEntry,
206
+ fapoEscalationMethod,
203
207
  fapoProposer,
204
208
  fsCampaignStorage,
205
209
  fsLineageStore,
206
- gepaParetoEntry,
210
+ gepaParetoMethod,
207
211
  gepaProposer,
208
- gepaReflectionEntry,
212
+ gepaReflectionMethod,
209
213
  gitWorktreeAdapter,
210
214
  haloProposer,
211
215
  heldOutGate,
@@ -262,7 +266,7 @@ export {
262
266
  selectPolicyEditAuthorRows,
263
267
  sequentialDecide,
264
268
  sequentialPairedGate,
265
- skillOptEntry,
269
+ skillOptMethod,
266
270
  skillOptProposer,
267
271
  surfaceContentHash,
268
272
  surfaceHash,