@pentatonic-ai/ai-agent-sdk 0.9.5 → 0.10.0

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 (131) hide show
  1. package/README.md +3 -3
  2. package/bin/cli.js +1 -1
  3. package/bin/commands/config.js +1 -1
  4. package/dist/index.cjs +39 -72
  5. package/dist/index.js +36 -69
  6. package/package.json +10 -3
  7. package/packages/doctor/src/checks/local-memory.js +2 -2
  8. package/packages/memory/README.md +2 -2
  9. package/packages/memory/openclaw-plugin/README.md +2 -2
  10. package/packages/memory/openclaw-plugin/openclaw.plugin.json +1 -1
  11. package/packages/memory/package-lock.json +49 -33
  12. package/packages/memory/package.json +4 -1
  13. package/packages/memory/src/__tests__/engine.test.js +40 -5
  14. package/packages/memory/src/engine.js +38 -3
  15. package/packages/memory/src/server.js +2 -2
  16. package/packages/memory-engine-v2/.env.example +30 -0
  17. package/packages/memory-engine-v2/README.md +125 -0
  18. package/packages/memory-engine-v2/compat/Dockerfile +11 -0
  19. package/packages/memory-engine-v2/compat/requirements.txt +6 -0
  20. package/packages/memory-engine-v2/compat/server.py +1047 -0
  21. package/packages/memory-engine-v2/docker-compose.aws.yml +78 -0
  22. package/packages/memory-engine-v2/docker-compose.yml +206 -0
  23. package/packages/memory-engine-v2/extractor-async/Dockerfile +14 -0
  24. package/packages/memory-engine-v2/extractor-async/confidence.py +62 -0
  25. package/packages/memory-engine-v2/extractor-async/noise_filter.py +144 -0
  26. package/packages/memory-engine-v2/extractor-async/requirements.txt +2 -0
  27. package/packages/memory-engine-v2/extractor-async/test_confidence.py +76 -0
  28. package/packages/memory-engine-v2/extractor-async/test_noise_filter.py +177 -0
  29. package/packages/memory-engine-v2/extractor-async/worker.py +797 -0
  30. package/packages/memory-engine-v2/extractor-sync/Dockerfile +11 -0
  31. package/packages/memory-engine-v2/extractor-sync/requirements.txt +4 -0
  32. package/packages/memory-engine-v2/extractor-sync/server.py +424 -0
  33. package/packages/memory-engine-v2/org-model/migrations/001_init.sql +390 -0
  34. package/packages/memory-engine-v2/tests/e2e_smoke.py +356 -0
  35. package/packages/memory-engine-v2/tests/fixtures/generate_synthetic_corpus.py +758 -0
  36. package/packages/memory-engine/.env.example +0 -13
  37. package/packages/memory-engine/MIGRATION.md +0 -219
  38. package/packages/memory-engine/README.md +0 -145
  39. package/packages/memory-engine/bench/README.md +0 -99
  40. package/packages/memory-engine/bench/scorecards-engine/agent-coding__pentatonic-baseline__20260427-142523.json +0 -1115
  41. package/packages/memory-engine/bench/scorecards-engine/chat-recall__pentatonic-baseline__20260427-142648.json +0 -819
  42. package/packages/memory-engine/bench/scorecards-engine/circular-economy__pentatonic-baseline__20260427-142757.json +0 -1278
  43. package/packages/memory-engine/bench/scorecards-engine/customer-support__pentatonic-baseline__20260427-142900.json +0 -1018
  44. package/packages/memory-engine/bench/scorecards-engine/marketplace-ops__pentatonic-baseline__20260427-142957.json +0 -1038
  45. package/packages/memory-engine/bench/scorecards-engine/product-catalogue__pentatonic-baseline__20260427-143122.json +0 -961
  46. package/packages/memory-engine/bench/scorecards-engine-via-docker/agent-coding__pentatonic-memory__20260427-161812.json +0 -1115
  47. package/packages/memory-engine/bench/scorecards-engine-via-docker/chat-recall__pentatonic-memory__20260427-161701.json +0 -819
  48. package/packages/memory-engine/bench/scorecards-engine-via-docker/circular-economy__pentatonic-memory__20260427-161713.json +0 -1278
  49. package/packages/memory-engine/bench/scorecards-engine-via-docker/customer-support__pentatonic-memory__20260427-161723.json +0 -1018
  50. package/packages/memory-engine/bench/scorecards-engine-via-docker/marketplace-ops__pentatonic-memory__20260427-161732.json +0 -1038
  51. package/packages/memory-engine/bench/scorecards-engine-via-docker/product-catalogue__pentatonic-memory__20260427-161741.json +0 -937
  52. package/packages/memory-engine/bench/scorecards-engine-via-l2-7-layer-populated/agent-coding__pentatonic-memory__20260427-184718.json +0 -1115
  53. package/packages/memory-engine/bench/scorecards-engine-via-l2-7-layer-populated/chat-recall__pentatonic-memory__20260427-184614.json +0 -819
  54. package/packages/memory-engine/bench/scorecards-engine-via-l2-7-layer-populated/circular-economy__pentatonic-memory__20260427-184809.json +0 -1278
  55. package/packages/memory-engine/bench/scorecards-engine-via-l2-7-layer-populated/customer-support__pentatonic-memory__20260427-184854.json +0 -1018
  56. package/packages/memory-engine/bench/scorecards-engine-via-l2-7-layer-populated/marketplace-ops__pentatonic-memory__20260427-184929.json +0 -1038
  57. package/packages/memory-engine/bench/scorecards-engine-via-l2-7-layer-populated/product-catalogue__pentatonic-memory__20260427-185015.json +0 -961
  58. package/packages/memory-engine/bench/scorecards-engine-via-l2-empty-layers/agent-coding__pentatonic-memory__20260427-175252.json +0 -1115
  59. package/packages/memory-engine/bench/scorecards-engine-via-l2-empty-layers/chat-recall__pentatonic-memory__20260427-175312.json +0 -819
  60. package/packages/memory-engine/bench/scorecards-engine-via-l2-empty-layers/circular-economy__pentatonic-memory__20260427-175335.json +0 -1278
  61. package/packages/memory-engine/bench/scorecards-engine-via-l2-empty-layers/customer-support__pentatonic-memory__20260427-175355.json +0 -1018
  62. package/packages/memory-engine/bench/scorecards-engine-via-l2-empty-layers/marketplace-ops__pentatonic-memory__20260427-175413.json +0 -1038
  63. package/packages/memory-engine/bench/scorecards-engine-via-l2-empty-layers/product-catalogue__pentatonic-memory__20260427-175430.json +0 -883
  64. package/packages/memory-engine/bench/scorecards-engine-via-shim/agent-coding__pentatonic-memory__20260427-155409.json +0 -1115
  65. package/packages/memory-engine/bench/scorecards-engine-via-shim/chat-recall__pentatonic-memory__20260427-155421.json +0 -819
  66. package/packages/memory-engine/bench/scorecards-engine-via-shim/circular-economy__pentatonic-memory__20260427-155433.json +0 -1278
  67. package/packages/memory-engine/bench/scorecards-engine-via-shim/customer-support__pentatonic-memory__20260427-155443.json +0 -1018
  68. package/packages/memory-engine/bench/scorecards-engine-via-shim/marketplace-ops__pentatonic-memory__20260427-155453.json +0 -1038
  69. package/packages/memory-engine/bench/scorecards-engine-via-shim/product-catalogue__pentatonic-memory__20260427-155503.json +0 -937
  70. package/packages/memory-engine/bench/scorecards-pentatonic-baseline/agent-coding__pentatonic-memory-latest__20260427-145103.json +0 -1115
  71. package/packages/memory-engine/bench/scorecards-pentatonic-baseline/agent-coding__pentatonic-memory__20260427-144909.json +0 -1115
  72. package/packages/memory-engine/bench/scorecards-pentatonic-baseline/chat-recall__pentatonic-memory-latest__20260427-145153.json +0 -819
  73. package/packages/memory-engine/bench/scorecards-pentatonic-baseline/chat-recall__pentatonic-memory__20260427-145120.json +0 -542
  74. package/packages/memory-engine/bench/scorecards-pentatonic-baseline/circular-economy__pentatonic-memory-latest__20260427-145313.json +0 -1278
  75. package/packages/memory-engine/bench/scorecards-pentatonic-baseline/circular-economy__pentatonic-memory__20260427-145207.json +0 -894
  76. package/packages/memory-engine/bench/scorecards-pentatonic-baseline/customer-support__pentatonic-memory-latest__20260427-145412.json +0 -1018
  77. package/packages/memory-engine/bench/scorecards-pentatonic-baseline/customer-support__pentatonic-memory__20260427-145327.json +0 -680
  78. package/packages/memory-engine/bench/scorecards-pentatonic-baseline/marketplace-ops__pentatonic-memory-latest__20260427-145517.json +0 -1038
  79. package/packages/memory-engine/bench/scorecards-pentatonic-baseline/marketplace-ops__pentatonic-memory__20260427-145422.json +0 -693
  80. package/packages/memory-engine/bench/scorecards-pentatonic-baseline/product-catalogue__pentatonic-memory-latest__20260427-145616.json +0 -961
  81. package/packages/memory-engine/bench/scorecards-pentatonic-baseline/product-catalogue__pentatonic-memory__20260427-145528.json +0 -727
  82. package/packages/memory-engine/compat/Dockerfile +0 -22
  83. package/packages/memory-engine/compat/server.py +0 -1255
  84. package/packages/memory-engine/docker-compose.test.yml +0 -59
  85. package/packages/memory-engine/docker-compose.yml +0 -240
  86. package/packages/memory-engine/engine/README.md +0 -52
  87. package/packages/memory-engine/engine/l2-hybridrag-proxy.py +0 -1543
  88. package/packages/memory-engine/engine/l5-comms-layer.py +0 -663
  89. package/packages/memory-engine/engine/l6-document-store.py +0 -1018
  90. package/packages/memory-engine/engine/services/_shared/__init__.py +0 -1
  91. package/packages/memory-engine/engine/services/_shared/embed_provider.py +0 -468
  92. package/packages/memory-engine/engine/services/l2/Dockerfile +0 -50
  93. package/packages/memory-engine/engine/services/l2/init_databases.py +0 -81
  94. package/packages/memory-engine/engine/services/l2/l2-hybridrag-proxy.py +0 -2721
  95. package/packages/memory-engine/engine/services/l5/Dockerfile +0 -11
  96. package/packages/memory-engine/engine/services/l5/l5-comms-layer.py +0 -808
  97. package/packages/memory-engine/engine/services/l6/Dockerfile +0 -30
  98. package/packages/memory-engine/engine/services/l6/l6-document-store.py +0 -1221
  99. package/packages/memory-engine/engine/services/nv-embed/Dockerfile +0 -28
  100. package/packages/memory-engine/engine/services/nv-embed/server.py +0 -152
  101. package/packages/memory-engine/pme_memory/__init__.py +0 -0
  102. package/packages/memory-engine/pme_memory/__main__.py +0 -129
  103. package/packages/memory-engine/pme_memory/artifacts.py +0 -95
  104. package/packages/memory-engine/pme_memory/embed.py +0 -74
  105. package/packages/memory-engine/pme_memory/health.py +0 -36
  106. package/packages/memory-engine/pme_memory/hygiene.py +0 -159
  107. package/packages/memory-engine/pme_memory/indexer.py +0 -200
  108. package/packages/memory-engine/pme_memory/needs.py +0 -55
  109. package/packages/memory-engine/pme_memory/provenance.py +0 -80
  110. package/packages/memory-engine/pme_memory/scoring.py +0 -168
  111. package/packages/memory-engine/pme_memory/search.py +0 -52
  112. package/packages/memory-engine/pme_memory/store.py +0 -86
  113. package/packages/memory-engine/pme_memory/synthesis.py +0 -114
  114. package/packages/memory-engine/pyproject.toml +0 -65
  115. package/packages/memory-engine/scripts/kg-extractor.py +0 -557
  116. package/packages/memory-engine/scripts/kg-preflexor-v2.py +0 -738
  117. package/packages/memory-engine/scripts/wipe-legacy-l3-entities.py +0 -128
  118. package/packages/memory-engine/tests/e2e_arena.sh +0 -259
  119. package/packages/memory-engine/tests/embed_stub/Dockerfile +0 -13
  120. package/packages/memory-engine/tests/embed_stub/server.py +0 -80
  121. package/packages/memory-engine/tests/test_aggregate.py +0 -333
  122. package/packages/memory-engine/tests/test_api_contract.sh +0 -57
  123. package/packages/memory-engine/tests/test_arena_safety.py +0 -232
  124. package/packages/memory-engine/tests/test_channel_stat_reader.py +0 -437
  125. package/packages/memory-engine/tests/test_channel_stat_rollups.py +0 -308
  126. package/packages/memory-engine/tests/test_compat_nv_embed_probe.py +0 -48
  127. package/packages/memory-engine/tests/test_embed_provider.py +0 -492
  128. package/packages/memory-engine/tests/test_l2_qmd_vec_search.py +0 -280
  129. package/packages/memory-engine/tests/test_l3_arena_isolation.py +0 -412
  130. package/packages/memory-engine/tests/test_l6_module_load.py +0 -84
  131. package/packages/memory-engine/tests/test_people_list_reader.py +0 -432
@@ -1,680 +0,0 @@
1
- {
2
- "bench": "customer-support",
3
- "stack": "pentatonic-memory",
4
- "n_tasks": 20,
5
- "n_correct": 5,
6
- "accuracy": 0.25,
7
- "mean_score": 0.25,
8
- "p50_search_ms": 30.443045994616114,
9
- "p95_search_ms": 37.8251028523664,
10
- "total_tokens_in": 0,
11
- "total_tokens_out": 0,
12
- "total_usd": 0.0,
13
- "by_tag": {
14
- "factoid": {
15
- "n": 10,
16
- "mean_score": 0.3,
17
- "accuracy": 0.3
18
- },
19
- "customer": {
20
- "n": 8,
21
- "mean_score": 0.375,
22
- "accuracy": 0.375
23
- },
24
- "multi-doc": {
25
- "n": 6,
26
- "mean_score": 0.16666666666666666,
27
- "accuracy": 0.16666666666666666
28
- },
29
- "rma": {
30
- "n": 3,
31
- "mean_score": 0.3333333333333333,
32
- "accuracy": 0.3333333333333333
33
- },
34
- "policy": {
35
- "n": 5,
36
- "mean_score": 0.2,
37
- "accuracy": 0.2
38
- },
39
- "escalation": {
40
- "n": 4,
41
- "mean_score": 0.25,
42
- "accuracy": 0.25
43
- },
44
- "rubric": {
45
- "n": 3,
46
- "mean_score": 0.3333333333333333,
47
- "accuracy": 0.3333333333333333
48
- },
49
- "multi-hop": {
50
- "n": 1,
51
- "mean_score": 0.0,
52
- "accuracy": 0.0
53
- },
54
- "entity": {
55
- "n": 2,
56
- "mean_score": 0.0,
57
- "accuracy": 0.0
58
- }
59
- },
60
- "extra": {
61
- "ingest_ms": 11268.10247899266,
62
- "grading": "substring",
63
- "limit": 3,
64
- "tokens": {
65
- "corpus_tokens": 1227,
66
- "query_tokens": 283,
67
- "context_tokens": 278,
68
- "retrieval_tokens": 561,
69
- "naive_tokens": 24823,
70
- "saved_tokens": 24262,
71
- "reduction_pct": 0.9773999919429561,
72
- "mean_retrieval_tokens_per_task": 28.05,
73
- "tokenizer": "cl100k_base",
74
- "per_task": {
75
- "order-mina-count": {
76
- "query": 11,
77
- "context": 0,
78
- "retrieval": 11,
79
- "judge_in": 0,
80
- "judge_out": 0,
81
- "judge_latency_ms": 0.0
82
- },
83
- "order-mina-latest": {
84
- "query": 13,
85
- "context": 0,
86
- "retrieval": 13,
87
- "judge_in": 0,
88
- "judge_out": 0,
89
- "judge_latency_ms": 0.0
90
- },
91
- "rma-mina-sleeve-reason": {
92
- "query": 17,
93
- "context": 0,
94
- "retrieval": 17,
95
- "judge_in": 0,
96
- "judge_out": 0,
97
- "judge_latency_ms": 0.0
98
- },
99
- "rma-mina-lid-resolution": {
100
- "query": 11,
101
- "context": 57,
102
- "retrieval": 68,
103
- "judge_in": 0,
104
- "judge_out": 0,
105
- "judge_latency_ms": 0.0
106
- },
107
- "jareth-harness-bar-followup": {
108
- "query": 15,
109
- "context": 0,
110
- "retrieval": 15,
111
- "judge_in": 0,
112
- "judge_out": 0,
113
- "judge_latency_ms": 0.0
114
- },
115
- "jareth-second-order": {
116
- "query": 11,
117
- "context": 0,
118
- "retrieval": 11,
119
- "judge_in": 0,
120
- "judge_out": 0,
121
- "judge_latency_ms": 0.0
122
- },
123
- "priya-custom-status": {
124
- "query": 11,
125
- "context": 0,
126
- "retrieval": 11,
127
- "judge_in": 0,
128
- "judge_out": 0,
129
- "judge_latency_ms": 0.0
130
- },
131
- "priya-goodwill-offered": {
132
- "query": 9,
133
- "context": 0,
134
- "retrieval": 9,
135
- "judge_in": 0,
136
- "judge_out": 0,
137
- "judge_latency_ms": 0.0
138
- },
139
- "policy-custom-return": {
140
- "query": 11,
141
- "context": 0,
142
- "retrieval": 11,
143
- "judge_in": 0,
144
- "judge_out": 0,
145
- "judge_latency_ms": 0.0
146
- },
147
- "policy-40day-return": {
148
- "query": 18,
149
- "context": 46,
150
- "retrieval": 64,
151
- "judge_in": 0,
152
- "judge_out": 0,
153
- "judge_latency_ms": 0.0
154
- },
155
- "escalation-400gbp-full-refund": {
156
- "query": 20,
157
- "context": 0,
158
- "retrieval": 20,
159
- "judge_in": 0,
160
- "judge_out": 0,
161
- "judge_latency_ms": 0.0
162
- },
163
- "escalation-goodwill-20pct-tier": {
164
- "query": 15,
165
- "context": 49,
166
- "retrieval": 64,
167
- "judge_in": 0,
168
- "judge_out": 0,
169
- "judge_latency_ms": 0.0
170
- },
171
- "mina-vip-status": {
172
- "query": 10,
173
- "context": 42,
174
- "retrieval": 52,
175
- "judge_in": 0,
176
- "judge_out": 0,
177
- "judge_latency_ms": 0.0
178
- },
179
- "mina-preferences": {
180
- "query": 11,
181
- "context": 42,
182
- "retrieval": 53,
183
- "judge_in": 0,
184
- "judge_out": 0,
185
- "judge_latency_ms": 0.0
186
- },
187
- "priya-agent-guidance": {
188
- "query": 16,
189
- "context": 0,
190
- "retrieval": 16,
191
- "judge_in": 0,
192
- "judge_out": 0,
193
- "judge_latency_ms": 0.0
194
- },
195
- "rubric-mina-warranty-recommendation": {
196
- "query": 18,
197
- "context": 42,
198
- "retrieval": 60,
199
- "judge_in": 374,
200
- "judge_out": 44,
201
- "judge_latency_ms": 945.222895026207
202
- },
203
- "rubric-jareth-escalation": {
204
- "query": 30,
205
- "context": 0,
206
- "retrieval": 30,
207
- "judge_in": 357,
208
- "judge_out": 43,
209
- "judge_latency_ms": 868.3296079933643
210
- },
211
- "rubric-priya-delayed-custom": {
212
- "query": 16,
213
- "context": 0,
214
- "retrieval": 16,
215
- "judge_in": 310,
216
- "judge_out": 54,
217
- "judge_latency_ms": 939.1167230010033
218
- },
219
- "entity-mina-orders": {
220
- "query": 10,
221
- "context": 0,
222
- "retrieval": 10,
223
- "judge_in": 0,
224
- "judge_out": 0,
225
- "judge_latency_ms": 0.0
226
- },
227
- "entity-all-rmas": {
228
- "query": 10,
229
- "context": 0,
230
- "retrieval": 10,
231
- "judge_in": 0,
232
- "judge_out": 0,
233
- "judge_latency_ms": 0.0
234
- }
235
- },
236
- "judge_tokens_in": 1041,
237
- "judge_tokens_out": 141,
238
- "judge_calls": 3,
239
- "judge_mean_latency_ms": 917.5564086735249
240
- },
241
- "cost_usd": {
242
- "assumed_completion_tokens_per_task": 100,
243
- "rates": {
244
- "input_per_1k": 0.0025,
245
- "output_per_1k": 0.01,
246
- "model": "gpt-4o"
247
- },
248
- "retrieval_usd_in": 0.0014025,
249
- "retrieval_usd_out": 0.02,
250
- "retrieval_usd_total": 0.0214025,
251
- "naive_usd_total": 0.0820575,
252
- "saved_usd": 0.060655,
253
- "saved_usd_per_1k_tasks": 3.03275
254
- }
255
- },
256
- "task_results": [
257
- {
258
- "task_id": "order-mina-count",
259
- "query": "How many orders has Mina Okafor placed?",
260
- "answer": "",
261
- "hits": [],
262
- "correct": false,
263
- "score": 0.0,
264
- "grading_notes": "missing 1/1: ['2']",
265
- "search_time_ms": 34.31391599588096,
266
- "generation_time_ms": 0.0,
267
- "tokens_in": 0,
268
- "tokens_out": 0,
269
- "retrieval_tokens": 11,
270
- "query_tokens": 11,
271
- "context_tokens": 0,
272
- "judge_tokens_in": 0,
273
- "judge_tokens_out": 0,
274
- "judge_latency_ms": 0.0
275
- },
276
- {
277
- "task_id": "order-mina-latest",
278
- "query": "What was in Mina Okafor's most recent order?",
279
- "answer": "",
280
- "hits": [],
281
- "correct": false,
282
- "score": 0.0,
283
- "grading_notes": "missing 2/2: ['Luna', 'Loop']",
284
- "search_time_ms": 29.60854201228358,
285
- "generation_time_ms": 0.0,
286
- "tokens_in": 0,
287
- "tokens_out": 0,
288
- "retrieval_tokens": 13,
289
- "query_tokens": 13,
290
- "context_tokens": 0,
291
- "judge_tokens_in": 0,
292
- "judge_tokens_out": 0,
293
- "judge_latency_ms": 0.0
294
- },
295
- {
296
- "task_id": "rma-mina-sleeve-reason",
297
- "query": "Why did Mina open an RMA on order 2026-0142?",
298
- "answer": "",
299
- "hits": [],
300
- "correct": false,
301
- "score": 0.0,
302
- "grading_notes": "missing 2/2: ['wrong colour', 'Oat']",
303
- "search_time_ms": 32.31730399420485,
304
- "generation_time_ms": 0.0,
305
- "tokens_in": 0,
306
- "tokens_out": 0,
307
- "retrieval_tokens": 17,
308
- "query_tokens": 17,
309
- "context_tokens": 0,
310
- "judge_tokens_in": 0,
311
- "judge_tokens_out": 0,
312
- "judge_latency_ms": 0.0
313
- },
314
- {
315
- "task_id": "rma-mina-lid-resolution",
316
- "query": "How was Mina's Luna bottle lid complaint resolved?",
317
- "answer": "[Chat 2026-04-10] Mina: Hey, the Luna bottle lid is leaking \u2014 is that covered? Agent: Yes, the lid has a 2-year warranty. I'll ship you a replacement lid free of charge. Mina: Great, thanks!",
318
- "hits": [
319
- {
320
- "text": "[Chat 2026-04-10] Mina: Hey, the Luna bottle lid is leaking \u2014 is that covered? Agent: Yes, the lid has a 2-year warranty. I'll ship you a replacement lid free of charge. Mina: Great, thanks!",
321
- "score": 0.5323119854064589,
322
- "source": "pentatonic-memory",
323
- "doc_id": "chat-mina-2026-04-10"
324
- }
325
- ],
326
- "correct": true,
327
- "score": 1.0,
328
- "grading_notes": "all substrings matched",
329
- "search_time_ms": 28.685988974757493,
330
- "generation_time_ms": 0.0,
331
- "tokens_in": 0,
332
- "tokens_out": 0,
333
- "retrieval_tokens": 68,
334
- "query_tokens": 11,
335
- "context_tokens": 57,
336
- "judge_tokens_in": 0,
337
- "judge_tokens_out": 0,
338
- "judge_latency_ms": 0.0
339
- },
340
- {
341
- "task_id": "jareth-harness-bar-followup",
342
- "query": "Did Jareth's kite harness work with his bar out of the box?",
343
- "answer": "",
344
- "hits": [],
345
- "correct": false,
346
- "score": 0.0,
347
- "grading_notes": "missing 2/2: ['Duotone', 'adapter']",
348
- "search_time_ms": 34.421516000293195,
349
- "generation_time_ms": 0.0,
350
- "tokens_in": 0,
351
- "tokens_out": 0,
352
- "retrieval_tokens": 15,
353
- "query_tokens": 15,
354
- "context_tokens": 0,
355
- "judge_tokens_in": 0,
356
- "judge_tokens_out": 0,
357
- "judge_latency_ms": 0.0
358
- },
359
- {
360
- "task_id": "jareth-second-order",
361
- "query": "What did Jareth order after his initial harness purchase?",
362
- "answer": "",
363
- "hits": [],
364
- "correct": false,
365
- "score": 0.0,
366
- "grading_notes": "missing 2/2: ['adapter', 'NMD-ADPT-DC']",
367
- "search_time_ms": 25.77133697923273,
368
- "generation_time_ms": 0.0,
369
- "tokens_in": 0,
370
- "tokens_out": 0,
371
- "retrieval_tokens": 11,
372
- "query_tokens": 11,
373
- "context_tokens": 0,
374
- "judge_tokens_in": 0,
375
- "judge_tokens_out": 0,
376
- "judge_latency_ms": 0.0
377
- },
378
- {
379
- "task_id": "priya-custom-status",
380
- "query": "Can Priya change the text on her custom tote?",
381
- "answer": "",
382
- "hits": [],
383
- "correct": false,
384
- "score": 0.0,
385
- "grading_notes": "missing 2/2: [\"can't change\", 'production']",
386
- "search_time_ms": 27.974719007033855,
387
- "generation_time_ms": 0.0,
388
- "tokens_in": 0,
389
- "tokens_out": 0,
390
- "retrieval_tokens": 11,
391
- "query_tokens": 11,
392
- "context_tokens": 0,
393
- "judge_tokens_in": 0,
394
- "judge_tokens_out": 0,
395
- "judge_latency_ms": 0.0
396
- },
397
- {
398
- "task_id": "priya-goodwill-offered",
399
- "query": "What goodwill credit did Priya get offered?",
400
- "answer": "",
401
- "hits": [],
402
- "correct": false,
403
- "score": 0.0,
404
- "grading_notes": "missing 1/1: ['15%']",
405
- "search_time_ms": 26.681170013034716,
406
- "generation_time_ms": 0.0,
407
- "tokens_in": 0,
408
- "tokens_out": 0,
409
- "retrieval_tokens": 9,
410
- "query_tokens": 9,
411
- "context_tokens": 0,
412
- "judge_tokens_in": 0,
413
- "judge_tokens_out": 0,
414
- "judge_latency_ms": 0.0
415
- },
416
- {
417
- "task_id": "policy-custom-return",
418
- "query": "Can a custom-printed tote be returned for refund?",
419
- "answer": "",
420
- "hits": [],
421
- "correct": false,
422
- "score": 0.0,
423
- "grading_notes": "missing 1/1: ['final sale']",
424
- "search_time_ms": 29.563406016677618,
425
- "generation_time_ms": 0.0,
426
- "tokens_in": 0,
427
- "tokens_out": 0,
428
- "retrieval_tokens": 11,
429
- "query_tokens": 11,
430
- "context_tokens": 0,
431
- "judge_tokens_in": 0,
432
- "judge_tokens_out": 0,
433
- "judge_latency_ms": 0.0
434
- },
435
- {
436
- "task_id": "policy-40day-return",
437
- "query": "If an unused item was delivered 40 days ago, can I get a full refund?",
438
- "answer": "Returns policy: unused items returnable within 30 days of delivery for full refund. After 30 days but under 90 days: store credit only. Custom-printed and consumable items are final sale and non-returnable.",
439
- "hits": [
440
- {
441
- "text": "Returns policy: unused items returnable within 30 days of delivery for full refund. After 30 days but under 90 days: store credit only. Custom-printed and consumable items are final sale and non-returnable.",
442
- "score": 0.5349228375664262,
443
- "source": "pentatonic-memory",
444
- "doc_id": "policy-returns-30day"
445
- }
446
- ],
447
- "correct": false,
448
- "score": 0.0,
449
- "grading_notes": "all substrings matched; forbidden substring(s) present: ['full refund']",
450
- "search_time_ms": 34.331914997892454,
451
- "generation_time_ms": 0.0,
452
- "tokens_in": 0,
453
- "tokens_out": 0,
454
- "retrieval_tokens": 64,
455
- "query_tokens": 18,
456
- "context_tokens": 46,
457
- "judge_tokens_in": 0,
458
- "judge_tokens_out": 0,
459
- "judge_latency_ms": 0.0
460
- },
461
- {
462
- "task_id": "escalation-400gbp-full-refund",
463
- "query": "A Tier 1 agent wants to issue a full refund on a \u00a3400 order \u2014 can they?",
464
- "answer": "",
465
- "hits": [],
466
- "correct": false,
467
- "score": 0.0,
468
- "grading_notes": "missing 1/1: ['Tier 3']",
469
- "search_time_ms": 32.81780600082129,
470
- "generation_time_ms": 0.0,
471
- "tokens_in": 0,
472
- "tokens_out": 0,
473
- "retrieval_tokens": 20,
474
- "query_tokens": 20,
475
- "context_tokens": 0,
476
- "judge_tokens_in": 0,
477
- "judge_tokens_out": 0,
478
- "judge_latency_ms": 0.0
479
- },
480
- {
481
- "task_id": "escalation-goodwill-20pct-tier",
482
- "query": "What's the minimum agent tier needed to offer 20% goodwill credit?",
483
- "answer": "Goodwill credit: agents at Tier 1 may offer up to 10% off next order as goodwill. Tier 2 may offer up to 25%. Anything above 25% or any full refund outside policy requires Tier 3 approval.",
484
- "hits": [
485
- {
486
- "text": "Goodwill credit: agents at Tier 1 may offer up to 10% off next order as goodwill. Tier 2 may offer up to 25%. Anything above 25% or any full refund outside policy requires Tier 3 approval.",
487
- "score": 0.5517141571721631,
488
- "source": "pentatonic-memory",
489
- "doc_id": "policy-goodwill-credit"
490
- }
491
- ],
492
- "correct": true,
493
- "score": 1.0,
494
- "grading_notes": "all substrings matched",
495
- "search_time_ms": 33.02935999818146,
496
- "generation_time_ms": 0.0,
497
- "tokens_in": 0,
498
- "tokens_out": 0,
499
- "retrieval_tokens": 64,
500
- "query_tokens": 15,
501
- "context_tokens": 49,
502
- "judge_tokens_in": 0,
503
- "judge_tokens_out": 0,
504
- "judge_latency_ms": 0.0
505
- },
506
- {
507
- "task_id": "mina-vip-status",
508
- "query": "Is Mina Okafor a VIP customer?",
509
- "answer": "Customer note \u2014 Mina Okafor: VIP tier (5+ orders, \u00a3350+ lifetime spend). Preferences: matte finishes, neutrals over brights. Has flagged interest in lifetime-warranty items.",
510
- "hits": [
511
- {
512
- "text": "Customer note \u2014 Mina Okafor: VIP tier (5+ orders, \u00a3350+ lifetime spend). Preferences: matte finishes, neutrals over brights. Has flagged interest in lifetime-warranty items.",
513
- "score": 0.7720720221593629,
514
- "source": "pentatonic-memory",
515
- "doc_id": "customer-note-mina-vip"
516
- }
517
- ],
518
- "correct": true,
519
- "score": 1.0,
520
- "grading_notes": "all substrings matched",
521
- "search_time_ms": 27.029798016883433,
522
- "generation_time_ms": 0.0,
523
- "tokens_in": 0,
524
- "tokens_out": 0,
525
- "retrieval_tokens": 52,
526
- "query_tokens": 10,
527
- "context_tokens": 42,
528
- "judge_tokens_in": 0,
529
- "judge_tokens_out": 0,
530
- "judge_latency_ms": 0.0
531
- },
532
- {
533
- "task_id": "mina-preferences",
534
- "query": "What are Mina Okafor's stated preferences?",
535
- "answer": "Customer note \u2014 Mina Okafor: VIP tier (5+ orders, \u00a3350+ lifetime spend). Preferences: matte finishes, neutrals over brights. Has flagged interest in lifetime-warranty items.",
536
- "hits": [
537
- {
538
- "text": "Customer note \u2014 Mina Okafor: VIP tier (5+ orders, \u00a3350+ lifetime spend). Preferences: matte finishes, neutrals over brights. Has flagged interest in lifetime-warranty items.",
539
- "score": 0.5363281403698282,
540
- "source": "pentatonic-memory",
541
- "doc_id": "customer-note-mina-vip"
542
- }
543
- ],
544
- "correct": true,
545
- "score": 1.0,
546
- "grading_notes": "all substrings matched",
547
- "search_time_ms": 26.91432400024496,
548
- "generation_time_ms": 0.0,
549
- "tokens_in": 0,
550
- "tokens_out": 0,
551
- "retrieval_tokens": 53,
552
- "query_tokens": 11,
553
- "context_tokens": 42,
554
- "judge_tokens_in": 0,
555
- "judge_tokens_out": 0,
556
- "judge_latency_ms": 0.0
557
- },
558
- {
559
- "task_id": "priya-agent-guidance",
560
- "query": "When handling Priya's orders, what should the agent do before production starts?",
561
- "answer": "",
562
- "hits": [],
563
- "correct": false,
564
- "score": 0.0,
565
- "grading_notes": "missing 2/2: ['confirm', 'print text']",
566
- "search_time_ms": 31.542113021714613,
567
- "generation_time_ms": 0.0,
568
- "tokens_in": 0,
569
- "tokens_out": 0,
570
- "retrieval_tokens": 16,
571
- "query_tokens": 16,
572
- "context_tokens": 0,
573
- "judge_tokens_in": 0,
574
- "judge_tokens_out": 0,
575
- "judge_latency_ms": 0.0
576
- },
577
- {
578
- "task_id": "rubric-mina-warranty-recommendation",
579
- "query": "Mina is asking for product recommendations. Based on her history, what should we suggest?",
580
- "answer": "Customer note \u2014 Mina Okafor: VIP tier (5+ orders, \u00a3350+ lifetime spend). Preferences: matte finishes, neutrals over brights. Has flagged interest in lifetime-warranty items.",
581
- "hits": [
582
- {
583
- "text": "Customer note \u2014 Mina Okafor: VIP tier (5+ orders, \u00a3350+ lifetime spend). Preferences: matte finishes, neutrals over brights. Has flagged interest in lifetime-warranty items.",
584
- "score": 0.5268969216421068,
585
- "source": "pentatonic-memory",
586
- "doc_id": "customer-note-mina-vip"
587
- }
588
- ],
589
- "correct": true,
590
- "score": 1.0,
591
- "grading_notes": "judge=1.00 \u2014 The answer references Mina's VIP status and her stated interest in lifetime-warranty items, meeting the criteria for a s",
592
- "search_time_ms": 31.27754997694865,
593
- "generation_time_ms": 0.0,
594
- "tokens_in": 0,
595
- "tokens_out": 0,
596
- "retrieval_tokens": 60,
597
- "query_tokens": 18,
598
- "context_tokens": 42,
599
- "judge_tokens_in": 374,
600
- "judge_tokens_out": 44,
601
- "judge_latency_ms": 945.222895026207
602
- },
603
- {
604
- "task_id": "rubric-jareth-escalation",
605
- "query": "Jareth wants to return his kite harness for a full refund, claiming it 'didn't work' with his bar \u2014 how should we handle this?",
606
- "answer": "",
607
- "hits": [],
608
- "correct": false,
609
- "score": 0.0,
610
- "grading_notes": "judge=0.00 \u2014 The answer fails to provide any context or resolution regarding Jareth's claim, simply processing a refund without addre",
611
- "search_time_ms": 38.004239002475515,
612
- "generation_time_ms": 0.0,
613
- "tokens_in": 0,
614
- "tokens_out": 0,
615
- "retrieval_tokens": 30,
616
- "query_tokens": 30,
617
- "context_tokens": 0,
618
- "judge_tokens_in": 357,
619
- "judge_tokens_out": 43,
620
- "judge_latency_ms": 868.3296079933643
621
- },
622
- {
623
- "task_id": "rubric-priya-delayed-custom",
624
- "query": "Priya is frustrated about her custom tote delay. What's the best response?",
625
- "answer": "",
626
- "hits": [],
627
- "correct": false,
628
- "score": 0.0,
629
- "grading_notes": "judge=0.00 \u2014 The response does not acknowledge that custom items have a 10-12 working day lead time and does not mention goodwill. It",
630
- "search_time_ms": 33.17419299855828,
631
- "generation_time_ms": 0.0,
632
- "tokens_in": 0,
633
- "tokens_out": 0,
634
- "retrieval_tokens": 16,
635
- "query_tokens": 16,
636
- "context_tokens": 0,
637
- "judge_tokens_in": 310,
638
- "judge_tokens_out": 54,
639
- "judge_latency_ms": 939.1167230010033
640
- },
641
- {
642
- "task_id": "entity-mina-orders",
643
- "query": "List Mina Okafor's order IDs.",
644
- "answer": "",
645
- "hits": [],
646
- "correct": false,
647
- "score": 0.0,
648
- "grading_notes": "no expected_substrings set",
649
- "search_time_ms": 27.45368899195455,
650
- "generation_time_ms": 0.0,
651
- "tokens_in": 0,
652
- "tokens_out": 0,
653
- "retrieval_tokens": 10,
654
- "query_tokens": 10,
655
- "context_tokens": 0,
656
- "judge_tokens_in": 0,
657
- "judge_tokens_out": 0,
658
- "judge_latency_ms": 0.0
659
- },
660
- {
661
- "task_id": "entity-all-rmas",
662
- "query": "List all open and closed RMA case IDs.",
663
- "answer": "",
664
- "hits": [],
665
- "correct": false,
666
- "score": 0.0,
667
- "grading_notes": "no expected_substrings set",
668
- "search_time_ms": 28.635605005547404,
669
- "generation_time_ms": 0.0,
670
- "tokens_in": 0,
671
- "tokens_out": 0,
672
- "retrieval_tokens": 10,
673
- "query_tokens": 10,
674
- "context_tokens": 0,
675
- "judge_tokens_in": 0,
676
- "judge_tokens_out": 0,
677
- "judge_latency_ms": 0.0
678
- }
679
- ]
680
- }