@openpond/evals 0.4.1 → 0.4.2

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.
@@ -42,7 +42,7 @@ export const harnessRefinerBenchmarkRelease = TasksetReleaseSchema.parse({
42
42
  ]
43
43
  }
44
44
  ],
45
- "contentHash": "05dbf9058c09b75bc950ca5cdf4ac03650dba7bb0cb74a3d9e152b16c110a2f7",
45
+ "contentHash": "4cef91a9c92df39d16f741b4d901dbde6b62e72bd8a48647a4a81c0d517d9634",
46
46
  "environment": {
47
47
  "defaultTimeoutMs": 900000,
48
48
  "deterministicSeeds": false,
@@ -98,14 +98,15 @@ export const harnessRefinerBenchmarkRelease = TasksetReleaseSchema.parse({
98
98
  "weight": 1
99
99
  }
100
100
  ],
101
- "id": "harness-refiner-public-v1",
101
+ "id": "harness-refiner-08112026",
102
102
  "metadata": {
103
103
  "adaptationSplit": "validation",
104
104
  "benchmark": "harness-refiner",
105
105
  "frozenEvaluationSplit": "frozen_eval",
106
106
  "primaryMetric": "paired_foreground_provider_tokens",
107
- "protocolVersion": "1",
107
+ "protocolVersion": "2",
108
108
  "qualityPolicy": "hard_non_regression",
109
+ "refinementMode": "sequential_product_lifecycle",
109
110
  "toolDeclarationSource": "openpond-production-model-tool-definitions",
110
111
  "trainingSideEffect": false
111
112
  },
@@ -122,7 +123,7 @@ export const harnessRefinerBenchmarkRelease = TasksetReleaseSchema.parse({
122
123
  "expectedOutput"
123
124
  ]
124
125
  },
125
- "revision": 3,
126
+ "revision": 1,
126
127
  "schemaVersion": "openpond.tasksetRelease.v2",
127
128
  "tasks": [
128
129
  {
@@ -1 +1 @@
1
- {"version":3,"file":"harness-refiner.d.ts","sourceRoot":"","sources":["../../../../../src/builtin-benchmarks/harness-refiner.ts"],"names":[],"mappings":"AAGA,eAAO,MAAM,8BAA8B;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;CAuuC1C,CAAC;AAEF,eAAO,MAAM,6BAA6B,EAAE,QAAQ,CAAC,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CASzE,CAAC"}
1
+ {"version":3,"file":"harness-refiner.d.ts","sourceRoot":"","sources":["../../../../../src/builtin-benchmarks/harness-refiner.ts"],"names":[],"mappings":"AAGA,eAAO,MAAM,8BAA8B;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;CAwuC1C,CAAC;AAEF,eAAO,MAAM,6BAA6B,EAAE,QAAQ,CAAC,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CASzE,CAAC"}
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@openpond/evals",
3
- "version": "0.4.1",
3
+ "version": "0.4.2",
4
4
  "description": "Portable Tasksets, graders, evaluation runs, receipts, conformance fixtures, and Work evidence for OpenPond Harnesses",
5
5
  "license": "MIT",
6
6
  "type": "module",