@tangle-network/agent-eval 0.112.0 → 0.114.0

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 (60) hide show
  1. package/CHANGELOG.md +34 -0
  2. package/dist/belief-state/index.js +1 -1
  3. package/dist/benchmarks/index.d.ts +2 -2
  4. package/dist/benchmarks/index.js +6 -6
  5. package/dist/builder-eval/index.js +1 -1
  6. package/dist/campaign/index.d.ts +55 -22
  7. package/dist/campaign/index.js +13 -7
  8. package/dist/{chunk-VLNGPUJJ.js → chunk-3LXTCTWL.js} +2 -2
  9. package/dist/{chunk-7QPK6W7R.js → chunk-ARU2PZFM.js} +2 -2
  10. package/dist/{chunk-EMKORATZ.js → chunk-DPZAEKA6.js} +2 -2
  11. package/dist/{chunk-XNCF3JU3.js → chunk-FAOEFFRT.js} +2 -2
  12. package/dist/{chunk-WZOIFFG2.js → chunk-LOBMT6SB.js} +2 -2
  13. package/dist/{chunk-RT2AFUXM.js → chunk-MOXWMGPC.js} +2 -2
  14. package/dist/{chunk-RCXMWGRY.js → chunk-N6MTC3GK.js} +69 -25
  15. package/dist/chunk-N6MTC3GK.js.map +1 -0
  16. package/dist/{chunk-YWGKQARF.js → chunk-NYFUT3B3.js} +2 -2
  17. package/dist/{chunk-VC43KQCK.js → chunk-PJQFMIOX.js} +62 -9
  18. package/dist/chunk-PJQFMIOX.js.map +1 -0
  19. package/dist/{chunk-QQVYMBCZ.js → chunk-VHPD6AXX.js} +512 -21
  20. package/dist/{chunk-QQVYMBCZ.js.map → chunk-VHPD6AXX.js.map} +1 -1
  21. package/dist/{chunk-BOLYPSAE.js → chunk-WBOGKYM4.js} +4 -4
  22. package/dist/{chunk-3LDPUYAC.js → chunk-X4UCIOTZ.js} +2 -2
  23. package/dist/contract/index.d.ts +7 -7
  24. package/dist/contract/index.js +9 -8
  25. package/dist/contract/index.js.map +1 -1
  26. package/dist/{gepa-T8T215nw.d.ts → gepa-DolL_Fko.d.ts} +2 -6
  27. package/dist/hosted/index.d.ts +1 -1
  28. package/dist/{index-Dc3VLGhp.d.ts → index-CWr5SIG-.d.ts} +1 -1
  29. package/dist/index.d.ts +8 -8
  30. package/dist/index.js +13 -11
  31. package/dist/index.js.map +1 -1
  32. package/dist/meta-eval/index.d.ts +1 -1
  33. package/dist/meta-eval/index.js +2 -2
  34. package/dist/multishot/index.d.ts +1 -1
  35. package/dist/openapi.json +1 -1
  36. package/dist/pipelines/index.js +2 -2
  37. package/dist/{pre-registration-BepVVa6P.d.ts → pre-registration-CTQbZbpX.d.ts} +1 -1
  38. package/dist/{provenance-KhY8ESVM.d.ts → provenance-BZmpWmn4.d.ts} +5 -11
  39. package/dist/reporting.d.ts +1 -1
  40. package/dist/reporting.js +4 -4
  41. package/dist/rl.d.ts +1 -1
  42. package/dist/rl.js +5 -5
  43. package/dist/{run-campaign-HJQEDCJQ.js → run-campaign-UADIM77S.js} +3 -3
  44. package/dist/{statistics-CDfpwIdX.d.ts → statistics-oUbOJe-S.d.ts} +34 -1
  45. package/dist/{types-v--ctu-b.d.ts → types-CgSlO6wT.d.ts} +26 -11
  46. package/docs/design/loop-taxonomy.md +7 -1
  47. package/docs/improvement-glossary.md +1 -1
  48. package/package.json +1 -1
  49. package/dist/chunk-RCXMWGRY.js.map +0 -1
  50. package/dist/chunk-VC43KQCK.js.map +0 -1
  51. /package/dist/{chunk-VLNGPUJJ.js.map → chunk-3LXTCTWL.js.map} +0 -0
  52. /package/dist/{chunk-7QPK6W7R.js.map → chunk-ARU2PZFM.js.map} +0 -0
  53. /package/dist/{chunk-EMKORATZ.js.map → chunk-DPZAEKA6.js.map} +0 -0
  54. /package/dist/{chunk-XNCF3JU3.js.map → chunk-FAOEFFRT.js.map} +0 -0
  55. /package/dist/{chunk-WZOIFFG2.js.map → chunk-LOBMT6SB.js.map} +0 -0
  56. /package/dist/{chunk-RT2AFUXM.js.map → chunk-MOXWMGPC.js.map} +0 -0
  57. /package/dist/{chunk-YWGKQARF.js.map → chunk-NYFUT3B3.js.map} +0 -0
  58. /package/dist/{chunk-BOLYPSAE.js.map → chunk-WBOGKYM4.js.map} +0 -0
  59. /package/dist/{chunk-3LDPUYAC.js.map → chunk-X4UCIOTZ.js.map} +0 -0
  60. /package/dist/{run-campaign-HJQEDCJQ.js.map → run-campaign-UADIM77S.js.map} +0 -0
@@ -1,9 +1,9 @@
1
1
  import {
2
2
  summarizeBackendIntegrity
3
- } from "./chunk-XNCF3JU3.js";
3
+ } from "./chunk-FAOEFFRT.js";
4
4
  import {
5
5
  paretoChart
6
- } from "./chunk-EMKORATZ.js";
6
+ } from "./chunk-DPZAEKA6.js";
7
7
  import {
8
8
  cohensD,
9
9
  pairedBootstrap,
@@ -12,7 +12,7 @@ import {
12
12
  pearsonR,
13
13
  requiredSampleSize,
14
14
  spearmanR
15
- } from "./chunk-VC43KQCK.js";
15
+ } from "./chunk-PJQFMIOX.js";
16
16
  import {
17
17
  llmSpans
18
18
  } from "./chunk-MHNQWM4I.js";
@@ -822,4 +822,4 @@ export {
822
822
  HoldoutAuditor,
823
823
  analyzeRuns
824
824
  };
825
- //# sourceMappingURL=chunk-BOLYPSAE.js.map
825
+ //# sourceMappingURL=chunk-WBOGKYM4.js.map
@@ -1,7 +1,7 @@
1
1
  import {
2
2
  pearsonR,
3
3
  spearmanR
4
- } from "./chunk-VC43KQCK.js";
4
+ } from "./chunk-PJQFMIOX.js";
5
5
 
6
6
  // src/meta-eval/rubric-predictive-validity.ts
7
7
  async function rubricPredictiveValidity(input) {
@@ -133,4 +133,4 @@ function makeRng(seed) {
133
133
  export {
134
134
  rubricPredictiveValidity
135
135
  };
136
- //# sourceMappingURL=chunk-3LDPUYAC.js.map
136
+ //# sourceMappingURL=chunk-X4UCIOTZ.js.map
@@ -1,9 +1,9 @@
1
- import { S as Scenario, M as MutableSurface, D as DispatchContext, b as JudgeConfig, f as SurfaceProposer, g as Gate, L as LabeledScenarioStore, C as CampaignResult, h as GateDecision } from '../types-v--ctu-b.js';
2
- export { i as CampaignAggregates, j as CampaignArtifactWriter, k as CampaignCellResult, l as CampaignCostMeter, d as CampaignTraceWriter, m as CodeSurface, c as Dispatch, n as GateContext, G as GateResult, o as GenerationCandidate, e as GenerationRecord, a as JudgeDimension, J as JudgeScore, p as Mutator, O as OptimizationProposer, q as OptimizerConfig, r as SessionScript } from '../types-v--ctu-b.js';
3
- import { L as LoopProvenanceRecord, P as PowerPreflight, R as RunEvalOptions } from '../provenance-KhY8ESVM.js';
4
- export { A as AxisEvidence, a as AxisVerdict, B as BuildEvidenceVectorOptions, D as DefaultProductionGateOptions, E as EvidenceVector, b as EvolutionaryProposerOptions, H as HeldOutGateOptions, O as ObjectiveSource, c as ParetoSignificanceGateOptions, d as PromotionObjective, e as PromotionPolicy, f as buildEvidenceVector, g as composeGate, h as defaultProductionGate, i as evolutionaryProposer, j as heldOutGate, p as paretoPolicy, k as paretoSignificanceGate, r as runEval } from '../provenance-KhY8ESVM.js';
5
- import { R as RunOptimizationOptions, a as RunImprovementLoopResult } from '../gepa-T8T215nw.js';
6
- export { G as GepaProposerOptions, b as RunCampaignOptions, c as RunImprovementLoopOptions, g as gepaProposer, r as runCampaign, d as runImprovementLoop } from '../gepa-T8T215nw.js';
1
+ import { S as Scenario, M as MutableSurface, D as DispatchContext, b as JudgeConfig, f as SurfaceProposer, g as Gate, L as LabeledScenarioStore, C as CampaignResult, h as GateDecision } from '../types-CgSlO6wT.js';
2
+ export { i as CampaignAggregates, j as CampaignArtifactWriter, k as CampaignCellResult, l as CampaignCostMeter, d as CampaignTraceWriter, m as CodeSurface, c as Dispatch, n as GateContext, G as GateResult, o as GenerationCandidate, e as GenerationRecord, a as JudgeDimension, J as JudgeScore, p as Mutator, O as OptimizationProposer, q as OptimizerConfig, r as SessionScript } from '../types-CgSlO6wT.js';
3
+ import { L as LoopProvenanceRecord, P as PowerPreflight, R as RunEvalOptions } from '../provenance-BZmpWmn4.js';
4
+ export { A as AxisEvidence, a as AxisVerdict, B as BuildEvidenceVectorOptions, D as DefaultProductionGateOptions, E as EvidenceVector, b as EvolutionaryProposerOptions, H as HeldOutGateOptions, O as ObjectiveSource, c as ParetoSignificanceGateOptions, d as PromotionObjective, e as PromotionPolicy, f as buildEvidenceVector, g as composeGate, h as defaultProductionGate, i as evolutionaryProposer, j as heldOutGate, p as paretoPolicy, k as paretoSignificanceGate, r as runEval } from '../provenance-BZmpWmn4.js';
5
+ import { R as RunOptimizationOptions, a as RunImprovementLoopResult } from '../gepa-DolL_Fko.js';
6
+ export { G as GepaProposerOptions, b as RunCampaignOptions, c as RunImprovementLoopOptions, g as gepaProposer, r as runCampaign, d as runImprovementLoop } from '../gepa-DolL_Fko.js';
7
7
  import { C as CampaignStorage } from '../storage-Dw_f7WMt.js';
8
8
  export { f as fsCampaignStorage, i as inMemoryCampaignStorage } from '../storage-Dw_f7WMt.js';
9
9
  export { D as DeploymentOutcome, F as FileSystemOutcomeStore, a as FileSystemOutcomeStoreOptions, I as InMemoryOutcomeStore, b as OutcomeStore } from '../outcome-store-rnXLEqSn.js';
@@ -16,7 +16,7 @@ export { A as AnalystFinding } from '../kind-factory-20hcaYpf.js';
16
16
  import { A as AnalyzeRunsOptions } from '../analyze-runs-Dmz6LA9e.js';
17
17
  export { a as analyzeRuns } from '../analyze-runs-Dmz6LA9e.js';
18
18
  export { C as CodeAgentSessionDiagnostic, a as CodeAgentSessionIntakeOptions, b as CodeAgentSessionIntakeResult, c as CodeAgentSessionMetrics, d as CodeAgentSessionSource, P as ParsedCodeAgentJsonl, f as fromClaudeCodeSession, e as fromCodexSession, g as fromKimiCodeSession, h as fromOpenCodeSession, i as fromPiSession, j as fromPigraphSession, p as parseCodeAgentJsonl } from '../code-agent-session-yitf9I-F.js';
19
- import '../statistics-CDfpwIdX.js';
19
+ import '../statistics-oUbOJe-S.js';
20
20
  import '../judge-calibration-7C-IDmKr.js';
21
21
  import '../types-C7DGg5ex.js';
22
22
  import '@tangle-network/tcloud';
@@ -12,7 +12,7 @@ import {
12
12
  } from "../chunk-ZZUXHH3R.js";
13
13
  import {
14
14
  analyzeRuns
15
- } from "../chunk-BOLYPSAE.js";
15
+ } from "../chunk-WBOGKYM4.js";
16
16
  import {
17
17
  buildEvidenceVector,
18
18
  composeGate,
@@ -26,14 +26,15 @@ import {
26
26
  powerPreflight,
27
27
  runEval,
28
28
  runImprovementLoop,
29
- surfaceContentHash
30
- } from "../chunk-RCXMWGRY.js";
29
+ surfaceContentHash,
30
+ surfaceHash
31
+ } from "../chunk-N6MTC3GK.js";
31
32
  import "../chunk-VI2UW6B6.js";
32
33
  import {
33
34
  fsCampaignStorage,
34
35
  inMemoryCampaignStorage,
35
36
  runCampaign
36
- } from "../chunk-XNCF3JU3.js";
37
+ } from "../chunk-FAOEFFRT.js";
37
38
  import {
38
39
  buildDefaultAnalystRegistry
39
40
  } from "../chunk-LVTGFSHF.js";
@@ -42,9 +43,9 @@ import {
42
43
  FileSystemOutcomeStore,
43
44
  InMemoryOutcomeStore
44
45
  } from "../chunk-3RF76KTD.js";
45
- import "../chunk-7QPK6W7R.js";
46
- import "../chunk-EMKORATZ.js";
47
- import "../chunk-VC43KQCK.js";
46
+ import "../chunk-ARU2PZFM.js";
47
+ import "../chunk-DPZAEKA6.js";
48
+ import "../chunk-PJQFMIOX.js";
48
49
  import "../chunk-MHNQWM4I.js";
49
50
  import "../chunk-GGE4NNQT.js";
50
51
  import {
@@ -286,7 +287,7 @@ async function shipEvalRunToHosted(tenant, opts, summary, raw, runDir) {
286
287
  const compositeMean = cells.length === 0 ? 0 : cells.reduce((s, c) => s + c.compositeMean, 0) / cells.length;
287
288
  return {
288
289
  index,
289
- surfaceHash: typeof surface === "string" ? hashString(surface) : hashString(JSON.stringify(surface ?? "")),
290
+ surfaceHash: surfaceHash(surface),
290
291
  surface,
291
292
  cells,
292
293
  compositeMean,