@remnic/cli 1.0.5 → 1.0.6

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 (6) hide show
  1. package/README.md +15 -3
  2. package/dist/assets/download-datasets.sh +468 -3
  3. package/dist/index.js +3958 -568
  4. package/package.json +47 -4
  5. package/dist/chunk-GAZ3DFWX.js +0 -12027
  6. package/dist/dist-7DCVQLUB.js +0 -292
package/dist/dist-7DCVQLUB.js DELETED
@@ -1,292 +0,0 @@
1
- import {
2
- ASSISTANT_AGENT_CONFIG_KEY,
3
- ASSISTANT_JUDGE_CONFIG_KEY,
4
- ASSISTANT_MEETING_PREP_SCENARIOS,
5
- ASSISTANT_MEETING_PREP_SMOKE_SCENARIOS,
6
- ASSISTANT_MORNING_BRIEF_SCENARIOS,
7
- ASSISTANT_MORNING_BRIEF_SMOKE_SCENARIOS,
8
- ASSISTANT_NEXT_BEST_ACTION_SCENARIOS,
9
- ASSISTANT_NEXT_BEST_ACTION_SMOKE_SCENARIOS,
10
- ASSISTANT_RUBRIC_DIMENSIONS,
11
- ASSISTANT_RUBRIC_ID_KEY,
12
- ASSISTANT_SEEDS_CONFIG_KEY,
13
- ASSISTANT_SPOT_CHECK_DIR_KEY,
14
- ASSISTANT_SYNTHESIS_SCENARIOS,
15
- ASSISTANT_SYNTHESIS_SMOKE_SCENARIOS,
16
- BENCHMARK_INTEGRITY_META_SCHEMA,
17
- BENCHMARK_RESULT_SCHEMA,
18
- BENCHMARK_SPLIT_TYPES,
19
- CANARY_FIXED_RECALL,
20
- CANARY_SCORE_FLOOR,
21
- DEFAULT_ASSISTANT_RUBRIC_ID,
22
- EMPTY_CONTAMINATION_MANIFEST,
23
- INTEGRITY_CIPHER_ALGORITHM,
24
- INTEGRITY_HASH_ALGORITHM,
25
- INTEGRITY_META_FIELDS,
26
- REQUIRED_FRONTMATTER_FIELDS,
27
- SCHEMA_TIER_FIXTURE,
28
- SCHEMA_TIER_SMOKE_FIXTURE,
29
- SEALED_PROMPT_REGISTRY,
30
- addContaminationEntry,
31
- aggregateTaskScores,
32
- assertCanaryUnderFloor,
33
- assertIntegrityMetaPresent,
34
- assertPublishableIntegrity,
35
- assertSha256Hex,
36
- assistantMeetingPrepDefinition,
37
- assistantMorningBriefDefinition,
38
- assistantNextBestActionDefinition,
39
- assistantSynthesisDefinition,
40
- backlinkF1,
41
- bootstrapMeanConfidenceInterval,
42
- buildBenchmarkPublishFeed,
43
- buildBenchmarkRunSeeds,
44
- buildJudgePayload,
45
- buildSchemaTierFixture,
46
- buildSchemaTierSmokeFixture,
47
- calendarFixture,
48
- canonicalJsonStringify,
49
- chatFixture,
50
- checkDatasetContamination,
51
- checkRegression,
52
- clampScore,
53
- cohensD,
54
- compareResults,
55
- computeSealHash,
56
- containsAnswer,
57
- createAnthropicProvider,
58
- createCanaryAdapter,
59
- createDeterministicSpotCheckLogger,
60
- createLightweightAdapter,
61
- createLiteLlmProvider,
62
- createOllamaProvider,
63
- createOpenAiCompatibleProvider,
64
- createProvider,
65
- createRemnicAdapter,
66
- createSeededRng,
67
- createSpotCheckFileLogger,
68
- defaultBenchmarkBaselineDir,
69
- defaultBenchmarkPublishPath,
70
- deleteBenchmarkResults,
71
- discoverAllProviders,
72
- emailFixture,
73
- entityRecall,
74
- exactMatch,
75
- f1Score,
76
- generateReport,
77
- getBenchmark,
78
- getBenchmarkLowerIsBetter,
79
- hashBytes,
80
- hashCanonicalJson,
81
- hashString,
82
- integrityMetaIsComplete,
83
- interpretEffectSize,
84
- isContaminationEntry,
85
- isContaminationManifest,
86
- isSealedQrelsArtifact,
87
- isSha256Hex,
88
- linkMatches,
89
- listBenchmarkBaselines,
90
- listBenchmarkResults,
91
- listBenchmarks,
92
- llmJudgeScore,
93
- loadBaseline,
94
- loadBenchmarkBaseline,
95
- loadBenchmarkResult,
96
- loadCustomBenchmarkFile,
97
- loadSealKeyFromEnv,
98
- loadSealedQrels,
99
- loadSealedRubric,
100
- matchEntity,
101
- mergeContaminationManifests,
102
- openSeal,
103
- orchestrateBenchmarkRuns,
104
- pairedDeltaConfidenceInterval,
105
- parseCustomBenchmark,
106
- parseRubricResponse,
107
- parseSealedQrels,
108
- precisionAtK,
109
- projectFolderFixture,
110
- recallAtK,
111
- renderBenchmarkResultExport,
112
- renderMemorySummaryForJudge,
113
- renderMemoryViewForAgent,
114
- resolveAssistantAgent,
115
- resolveAssistantRubricId,
116
- resolveAssistantSeeds,
117
- resolveAssistantSpotCheckDir,
118
- resolveBenchmarkResultReference,
119
- resolveBenchmarkRunCount,
120
- resolveStructuredJudge,
121
- rotateDistractors,
122
- rougeL,
123
- runAssistantBenchmark,
124
- runAssistantMeetingPrepBenchmark,
125
- runAssistantMorningBriefBenchmark,
126
- runAssistantNextBestActionBenchmark,
127
- runAssistantSynthesisBenchmark,
128
- runBenchSuite,
129
- runBenchmark,
130
- runCustomBenchmarkFile,
131
- runExplain,
132
- runSealedJudge,
133
- safeHexEqual,
134
- saveBaseline,
135
- saveBenchmarkBaseline,
136
- schemaCompleteness,
137
- sealPayload,
138
- selectFixtureVariant,
139
- serializeSealedQrels,
140
- shuffleTasks,
141
- timed,
142
- verifyRubricDigest,
143
- writeBenchmarkPublishFeed,
144
- writeBenchmarkResult,
145
- zeroScores
146
- } from "./chunk-GAZ3DFWX.js";
147
- export {
148
- ASSISTANT_AGENT_CONFIG_KEY,
149
- ASSISTANT_JUDGE_CONFIG_KEY,
150
- ASSISTANT_MEETING_PREP_SCENARIOS,
151
- ASSISTANT_MEETING_PREP_SMOKE_SCENARIOS,
152
- ASSISTANT_MORNING_BRIEF_SCENARIOS,
153
- ASSISTANT_MORNING_BRIEF_SMOKE_SCENARIOS,
154
- ASSISTANT_NEXT_BEST_ACTION_SCENARIOS,
155
- ASSISTANT_NEXT_BEST_ACTION_SMOKE_SCENARIOS,
156
- ASSISTANT_RUBRIC_DIMENSIONS,
157
- ASSISTANT_RUBRIC_ID_KEY,
158
- ASSISTANT_SEEDS_CONFIG_KEY,
159
- ASSISTANT_SPOT_CHECK_DIR_KEY,
160
- ASSISTANT_SYNTHESIS_SCENARIOS,
161
- ASSISTANT_SYNTHESIS_SMOKE_SCENARIOS,
162
- BENCHMARK_INTEGRITY_META_SCHEMA,
163
- BENCHMARK_RESULT_SCHEMA,
164
- BENCHMARK_SPLIT_TYPES,
165
- CANARY_FIXED_RECALL,
166
- CANARY_SCORE_FLOOR,
167
- DEFAULT_ASSISTANT_RUBRIC_ID,
168
- EMPTY_CONTAMINATION_MANIFEST,
169
- INTEGRITY_CIPHER_ALGORITHM,
170
- INTEGRITY_HASH_ALGORITHM,
171
- INTEGRITY_META_FIELDS,
172
- REQUIRED_FRONTMATTER_FIELDS,
173
- SCHEMA_TIER_FIXTURE,
174
- SCHEMA_TIER_SMOKE_FIXTURE,
175
- SEALED_PROMPT_REGISTRY,
176
- addContaminationEntry,
177
- aggregateTaskScores,
178
- assertCanaryUnderFloor,
179
- assertIntegrityMetaPresent,
180
- assertPublishableIntegrity,
181
- assertSha256Hex,
182
- assistantMeetingPrepDefinition,
183
- assistantMorningBriefDefinition,
184
- assistantNextBestActionDefinition,
185
- assistantSynthesisDefinition,
186
- backlinkF1,
187
- bootstrapMeanConfidenceInterval,
188
- buildBenchmarkPublishFeed,
189
- buildBenchmarkRunSeeds,
190
- buildJudgePayload,
191
- buildSchemaTierFixture,
192
- buildSchemaTierSmokeFixture,
193
- calendarFixture,
194
- canonicalJsonStringify,
195
- chatFixture,
196
- checkDatasetContamination,
197
- checkRegression,
198
- clampScore,
199
- cohensD,
200
- compareResults,
201
- computeSealHash,
202
- containsAnswer,
203
- createAnthropicProvider,
204
- createCanaryAdapter,
205
- createDeterministicSpotCheckLogger,
206
- createLightweightAdapter,
207
- createLiteLlmProvider,
208
- createOllamaProvider,
209
- createOpenAiCompatibleProvider,
210
- createProvider,
211
- createRemnicAdapter,
212
- createSeededRng,
213
- createSpotCheckFileLogger,
214
- defaultBenchmarkBaselineDir,
215
- defaultBenchmarkPublishPath,
216
- deleteBenchmarkResults,
217
- discoverAllProviders,
218
- emailFixture,
219
- entityRecall,
220
- exactMatch,
221
- f1Score,
222
- generateReport,
223
- getBenchmark,
224
- getBenchmarkLowerIsBetter,
225
- hashBytes,
226
- hashCanonicalJson,
227
- hashString,
228
- integrityMetaIsComplete,
229
- interpretEffectSize,
230
- isContaminationEntry,
231
- isContaminationManifest,
232
- isSealedQrelsArtifact,
233
- isSha256Hex,
234
- linkMatches,
235
- listBenchmarkBaselines,
236
- listBenchmarkResults,
237
- listBenchmarks,
238
- llmJudgeScore,
239
- loadBaseline,
240
- loadBenchmarkBaseline,
241
- loadBenchmarkResult,
242
- loadCustomBenchmarkFile,
243
- loadSealKeyFromEnv,
244
- loadSealedQrels,
245
- loadSealedRubric,
246
- matchEntity,
247
- mergeContaminationManifests,
248
- openSeal,
249
- orchestrateBenchmarkRuns,
250
- pairedDeltaConfidenceInterval,
251
- parseCustomBenchmark,
252
- parseRubricResponse,
253
- parseSealedQrels,
254
- precisionAtK,
255
- projectFolderFixture,
256
- recallAtK,
257
- renderBenchmarkResultExport,
258
- renderMemorySummaryForJudge,
259
- renderMemoryViewForAgent,
260
- resolveAssistantAgent,
261
- resolveAssistantRubricId,
262
- resolveAssistantSeeds,
263
- resolveAssistantSpotCheckDir,
264
- resolveBenchmarkResultReference,
265
- resolveBenchmarkRunCount,
266
- resolveStructuredJudge,
267
- rotateDistractors,
268
- rougeL,
269
- runAssistantBenchmark,
270
- runAssistantMeetingPrepBenchmark,
271
- runAssistantMorningBriefBenchmark,
272
- runAssistantNextBestActionBenchmark,
273
- runAssistantSynthesisBenchmark,
274
- runBenchSuite,
275
- runBenchmark,
276
- runCustomBenchmarkFile,
277
- runExplain,
278
- runSealedJudge,
279
- safeHexEqual,
280
- saveBaseline,
281
- saveBenchmarkBaseline,
282
- schemaCompleteness,
283
- sealPayload,
284
- selectFixtureVariant,
285
- serializeSealedQrels,
286
- shuffleTasks,
287
- timed,
288
- verifyRubricDigest,
289
- writeBenchmarkPublishFeed,
290
- writeBenchmarkResult,
291
- zeroScores
292
- };