@dsh-enhanced/assistant-evaluation 0.1.23 → 0.1.30

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 (132) hide show
  1. package/README.md +109 -3
  2. package/bin/dsh-benchmark.js +21 -0
  3. package/lib/benchmark/cli.d.ts +6 -0
  4. package/lib/benchmark/cli.d.ts.map +1 -0
  5. package/lib/benchmark/cli.js +332 -0
  6. package/lib/benchmark/cli.js.map +1 -0
  7. package/lib/benchmark/corpus.d.ts +28 -0
  8. package/lib/benchmark/corpus.d.ts.map +1 -0
  9. package/lib/benchmark/corpus.js +173 -0
  10. package/lib/benchmark/corpus.js.map +1 -0
  11. package/lib/benchmark/deepseek.d.ts +15 -0
  12. package/lib/benchmark/deepseek.d.ts.map +1 -0
  13. package/lib/benchmark/deepseek.js +143 -0
  14. package/lib/benchmark/deepseek.js.map +1 -0
  15. package/lib/benchmark/holdout-evidence.d.ts +126 -0
  16. package/lib/benchmark/holdout-evidence.d.ts.map +1 -0
  17. package/lib/benchmark/holdout-evidence.js +466 -0
  18. package/lib/benchmark/holdout-evidence.js.map +1 -0
  19. package/lib/benchmark/holdout-protocol.d.ts +68 -0
  20. package/lib/benchmark/holdout-protocol.d.ts.map +1 -0
  21. package/lib/benchmark/holdout-protocol.js +271 -0
  22. package/lib/benchmark/holdout-protocol.js.map +1 -0
  23. package/lib/benchmark/holdout-provider.d.ts +26 -0
  24. package/lib/benchmark/holdout-provider.d.ts.map +1 -0
  25. package/lib/benchmark/holdout-provider.js +382 -0
  26. package/lib/benchmark/holdout-provider.js.map +1 -0
  27. package/lib/benchmark/holdout.d.ts +83 -0
  28. package/lib/benchmark/holdout.d.ts.map +1 -0
  29. package/lib/benchmark/holdout.js +391 -0
  30. package/lib/benchmark/holdout.js.map +1 -0
  31. package/lib/benchmark/index.d.ts +9 -0
  32. package/lib/benchmark/index.d.ts.map +1 -0
  33. package/lib/benchmark/index.js +7 -0
  34. package/lib/benchmark/index.js.map +1 -0
  35. package/lib/benchmark/memory-corpus.d.ts +55 -0
  36. package/lib/benchmark/memory-corpus.d.ts.map +1 -0
  37. package/lib/benchmark/memory-corpus.js +153 -0
  38. package/lib/benchmark/memory-corpus.js.map +1 -0
  39. package/lib/benchmark/memory-runtime.d.ts +63 -0
  40. package/lib/benchmark/memory-runtime.d.ts.map +1 -0
  41. package/lib/benchmark/memory-runtime.js +75 -0
  42. package/lib/benchmark/memory-runtime.js.map +1 -0
  43. package/lib/benchmark/native.d.ts +46 -0
  44. package/lib/benchmark/native.d.ts.map +1 -0
  45. package/lib/benchmark/native.js +300 -0
  46. package/lib/benchmark/native.js.map +1 -0
  47. package/lib/benchmark/report.d.ts +4 -0
  48. package/lib/benchmark/report.d.ts.map +1 -0
  49. package/lib/benchmark/report.js +153 -0
  50. package/lib/benchmark/report.js.map +1 -0
  51. package/lib/benchmark/runner.d.ts +25 -0
  52. package/lib/benchmark/runner.d.ts.map +1 -0
  53. package/lib/benchmark/runner.js +119 -0
  54. package/lib/benchmark/runner.js.map +1 -0
  55. package/lib/benchmark/schema.d.ts +20 -0
  56. package/lib/benchmark/schema.d.ts.map +1 -0
  57. package/lib/benchmark/schema.js +183 -0
  58. package/lib/benchmark/schema.js.map +1 -0
  59. package/lib/benchmark/store.d.ts +18 -0
  60. package/lib/benchmark/store.d.ts.map +1 -0
  61. package/lib/benchmark/store.js +321 -0
  62. package/lib/benchmark/store.js.map +1 -0
  63. package/lib/benchmark/strategy-capabilities.d.ts +154 -0
  64. package/lib/benchmark/strategy-capabilities.d.ts.map +1 -0
  65. package/lib/benchmark/strategy-capabilities.js +300 -0
  66. package/lib/benchmark/strategy-capabilities.js.map +1 -0
  67. package/lib/benchmark/strategy-config.d.ts +26 -0
  68. package/lib/benchmark/strategy-config.d.ts.map +1 -0
  69. package/lib/benchmark/strategy-config.js +80 -0
  70. package/lib/benchmark/strategy-config.js.map +1 -0
  71. package/lib/benchmark/strategy-corpus.d.ts +21 -0
  72. package/lib/benchmark/strategy-corpus.d.ts.map +1 -0
  73. package/lib/benchmark/strategy-corpus.js +95 -0
  74. package/lib/benchmark/strategy-corpus.js.map +1 -0
  75. package/lib/benchmark/strategy-evidence.d.ts +161 -0
  76. package/lib/benchmark/strategy-evidence.d.ts.map +1 -0
  77. package/lib/benchmark/strategy-evidence.js +520 -0
  78. package/lib/benchmark/strategy-evidence.js.map +1 -0
  79. package/lib/benchmark/strategy-executor.d.ts +27 -0
  80. package/lib/benchmark/strategy-executor.d.ts.map +1 -0
  81. package/lib/benchmark/strategy-executor.js +178 -0
  82. package/lib/benchmark/strategy-executor.js.map +1 -0
  83. package/lib/benchmark/strategy-goal-runtime.d.ts +78 -0
  84. package/lib/benchmark/strategy-goal-runtime.d.ts.map +1 -0
  85. package/lib/benchmark/strategy-goal-runtime.js +324 -0
  86. package/lib/benchmark/strategy-goal-runtime.js.map +1 -0
  87. package/lib/benchmark/strategy-meter.d.ts +62 -0
  88. package/lib/benchmark/strategy-meter.d.ts.map +1 -0
  89. package/lib/benchmark/strategy-meter.js +236 -0
  90. package/lib/benchmark/strategy-meter.js.map +1 -0
  91. package/lib/benchmark/strategy-owner.d.ts +71 -0
  92. package/lib/benchmark/strategy-owner.d.ts.map +1 -0
  93. package/lib/benchmark/strategy-owner.js +282 -0
  94. package/lib/benchmark/strategy-owner.js.map +1 -0
  95. package/lib/benchmark/strategy-plan.d.ts +51 -0
  96. package/lib/benchmark/strategy-plan.d.ts.map +1 -0
  97. package/lib/benchmark/strategy-plan.js +71 -0
  98. package/lib/benchmark/strategy-plan.js.map +1 -0
  99. package/lib/benchmark/strategy-policy.d.ts +16 -0
  100. package/lib/benchmark/strategy-policy.d.ts.map +1 -0
  101. package/lib/benchmark/strategy-policy.js +77 -0
  102. package/lib/benchmark/strategy-policy.js.map +1 -0
  103. package/lib/benchmark/strategy.d.ts +17 -0
  104. package/lib/benchmark/strategy.d.ts.map +1 -0
  105. package/lib/benchmark/strategy.js +21 -0
  106. package/lib/benchmark/strategy.js.map +1 -0
  107. package/lib/benchmark/types.d.ts +133 -0
  108. package/lib/benchmark/types.d.ts.map +1 -0
  109. package/lib/benchmark/types.js +3 -0
  110. package/lib/benchmark/types.js.map +1 -0
  111. package/lib/benchmark/usage.d.ts +36 -0
  112. package/lib/benchmark/usage.d.ts.map +1 -0
  113. package/lib/benchmark/usage.js +84 -0
  114. package/lib/benchmark/usage.js.map +1 -0
  115. package/lib/service.d.ts +50 -1
  116. package/lib/service.d.ts.map +1 -1
  117. package/lib/service.js +556 -9
  118. package/lib/service.js.map +1 -1
  119. package/lib/sqlite.d.ts +1 -1
  120. package/lib/sqlite.d.ts.map +1 -1
  121. package/lib/sqlite.js +94 -2
  122. package/lib/sqlite.js.map +1 -1
  123. package/lib/store.d.ts +27 -5
  124. package/lib/store.d.ts.map +1 -1
  125. package/lib/store.js +324 -25
  126. package/lib/store.js.map +1 -1
  127. package/lib/types.d.ts +114 -5
  128. package/lib/types.d.ts.map +1 -1
  129. package/lib/types.js.map +1 -1
  130. package/lib/version.d.ts +1 -1
  131. package/lib/version.js +1 -1
  132. package/package.json +125 -18
package/package.json CHANGED
@@ -1,20 +1,44 @@
1
1
  {
2
2
  "name": "@dsh-enhanced/assistant-evaluation",
3
- "version": "0.1.23",
3
+ "version": "0.1.30",
4
4
  "description": "A bounded, append-only outcome evaluation ledger for the DSH personal assistant.",
5
5
  "license": "MIT",
6
6
  "type": "module",
7
7
  "main": "./lib/index.js",
8
8
  "types": "./lib/index.d.ts",
9
+ "bin": {
10
+ "dsh-benchmark": "./bin/dsh-benchmark.js"
11
+ },
9
12
  "exports": {
10
13
  ".": {
11
14
  "types": "./lib/index.d.ts",
12
15
  "default": "./lib/index.js"
13
16
  },
17
+ "./benchmark": {
18
+ "types": "./lib/benchmark/index.d.ts",
19
+ "default": "./lib/benchmark/index.js"
20
+ },
21
+ "./benchmark/native": {
22
+ "types": "./lib/benchmark/native.d.ts",
23
+ "default": "./lib/benchmark/native.js"
24
+ },
25
+ "./benchmark/holdout": {
26
+ "types": "./lib/benchmark/holdout.d.ts",
27
+ "default": "./lib/benchmark/holdout.js"
28
+ },
14
29
  "./cordis.patch.yml": "./cordis.patch.yml",
15
- "./package.json": "./package.json"
30
+ "./package.json": "./package.json",
31
+ "./benchmark/strategy": {
32
+ "types": "./lib/benchmark/strategy.d.ts",
33
+ "default": "./lib/benchmark/strategy.js"
34
+ },
35
+ "./benchmark/deepseek": {
36
+ "types": "./lib/benchmark/deepseek.d.ts",
37
+ "default": "./lib/benchmark/deepseek.js"
38
+ }
16
39
  },
17
40
  "files": [
41
+ "bin",
18
42
  "lib",
19
43
  "cordis.patch.yml",
20
44
  "README.md",
@@ -43,14 +67,46 @@
43
67
  },
44
68
  "peerDependencies": {
45
69
  "@deepseek-ai/cordis": "^4.0.1",
46
- "@deepseek-ai/dsh-agent": ">=0.1.0-rc.8 <0.2.0",
47
- "@deepseek-ai/dsh-skill": ">=0.1.0-rc.8 <0.2.0",
48
- "@deepseek-ai/dsh-tools": ">=0.1.0-rc.8 <0.2.0",
70
+ "@deepseek-ai/dsh-agent": ">=0.1.2-rc.1 <0.2.0",
71
+ "@deepseek-ai/dsh-agent-loop": ">=0.1.2-rc.1 <0.2.0",
72
+ "@deepseek-ai/dsh-llm": ">=0.1.2-rc.1 <0.2.0",
73
+ "@deepseek-ai/dsh-session": ">=0.1.2-rc.1 <0.2.0",
74
+ "@deepseek-ai/dsh-session-projection": ">=0.1.2-rc.1 <0.2.0",
75
+ "@deepseek-ai/dsh-system-prompt": ">=0.1.2-rc.1 <0.2.0",
76
+ "@deepseek-ai/dsh-skill": ">=0.1.2-rc.1 <0.2.0",
77
+ "@deepseek-ai/dsh-tools": ">=0.1.2-rc.1 <0.2.0",
49
78
  "@dsh-enhanced/assistant-automations": ">0.1.7 <0.2.0",
50
- "@dsh-enhanced/assistant-delivery": ">0.1.7 <0.2.0",
51
- "@dsh-enhanced/assistant-evolution": ">0.1.7 <0.2.0"
79
+ "@dsh-enhanced/assistant-delivery": ">0.1.24 <0.2.0",
80
+ "@dsh-enhanced/assistant-evolution": ">0.1.7 <0.2.0",
81
+ "@dsh-enhanced/personal-memory": ">=0.1.24 <0.2.0",
82
+ "@dsh-enhanced/assistant-policy": ">=0.1.24 <0.2.0",
83
+ "@deepseek-ai/dsh-session-persistence-jsonl": ">=0.1.2-rc.1 <0.2.0",
84
+ "@deepseek-ai/dsh-session-persistence": ">=0.1.2-rc.1 <0.2.0",
85
+ "@deepseek-ai/dsh-goal": ">=0.1.2-rc.1 <0.2.0",
86
+ "@deepseek-ai/dsh-tool-goal": ">=0.1.2-rc.1 <0.2.0",
87
+ "@deepseek-ai/dsh-goal-round-driver": ">=0.1.2-rc.1 <0.2.0",
88
+ "@deepseek-ai/dsh-subagent": ">=0.1.2-rc.1 <0.2.0",
89
+ "@dsh-enhanced/assistant-goals": ">0.1.24 <0.2.0",
90
+ "@dsh-enhanced/assistant-isolation": ">=0.1.0 <0.2.0",
91
+ "@dsh-enhanced/assistant-verifier": ">=0.1.0 <0.2.0",
92
+ "@dsh-enhanced/assistant-deepseek-budget": ">=0.1.0 <0.2.0"
52
93
  },
53
94
  "peerDependenciesMeta": {
95
+ "@deepseek-ai/dsh-agent-loop": {
96
+ "optional": true
97
+ },
98
+ "@deepseek-ai/dsh-llm": {
99
+ "optional": true
100
+ },
101
+ "@deepseek-ai/dsh-session": {
102
+ "optional": true
103
+ },
104
+ "@deepseek-ai/dsh-session-projection": {
105
+ "optional": true
106
+ },
107
+ "@deepseek-ai/dsh-system-prompt": {
108
+ "optional": true
109
+ },
54
110
  "@deepseek-ai/cordis": {
55
111
  "optional": true
56
112
  },
@@ -71,25 +127,76 @@
71
127
  },
72
128
  "@dsh-enhanced/assistant-evolution": {
73
129
  "optional": true
130
+ },
131
+ "@dsh-enhanced/personal-memory": {
132
+ "optional": true
133
+ },
134
+ "@dsh-enhanced/assistant-policy": {
135
+ "optional": true
136
+ },
137
+ "@deepseek-ai/dsh-session-persistence-jsonl": {
138
+ "optional": true
139
+ },
140
+ "@deepseek-ai/dsh-session-persistence": {
141
+ "optional": true
142
+ },
143
+ "@deepseek-ai/dsh-goal": {
144
+ "optional": true
145
+ },
146
+ "@deepseek-ai/dsh-tool-goal": {
147
+ "optional": true
148
+ },
149
+ "@deepseek-ai/dsh-goal-round-driver": {
150
+ "optional": true
151
+ },
152
+ "@deepseek-ai/dsh-subagent": {
153
+ "optional": true
154
+ },
155
+ "@dsh-enhanced/assistant-goals": {
156
+ "optional": true
157
+ },
158
+ "@dsh-enhanced/assistant-isolation": {
159
+ "optional": true
160
+ },
161
+ "@dsh-enhanced/assistant-verifier": {
162
+ "optional": true
163
+ },
164
+ "@dsh-enhanced/assistant-deepseek-budget": {
165
+ "optional": true
74
166
  }
75
167
  },
76
168
  "devDependencies": {
77
- "@deepseek-ai/cordis": "4.0.1",
78
- "@deepseek-ai/dsh-agent": "0.1.0-rc.8",
79
- "@deepseek-ai/dsh-llm": "0.1.0-rc.8",
80
- "@deepseek-ai/dsh-session": "0.1.0-rc.8",
81
- "@deepseek-ai/dsh-skill": "0.1.0-rc.8",
82
- "@deepseek-ai/dsh-system-prompt": "0.1.0-rc.8",
83
- "@deepseek-ai/dsh-tools": "0.1.0-rc.8",
169
+ "@deepseek-ai/dsh-agent-loop": "0.1.2-rc.1",
170
+ "@deepseek-ai/dsh-session-projection": "0.1.2-rc.1",
171
+ "@deepseek-ai/cordis": "4.0.2",
172
+ "@deepseek-ai/dsh-agent": "0.1.2-rc.1",
173
+ "@deepseek-ai/dsh-llm": "0.1.2-rc.1",
174
+ "@deepseek-ai/dsh-session": "0.1.2-rc.1",
175
+ "@deepseek-ai/dsh-skill": "0.1.2-rc.1",
176
+ "@deepseek-ai/dsh-system-prompt": "0.1.2-rc.1",
177
+ "@deepseek-ai/dsh-tools": "0.1.2-rc.1",
84
178
  "@types/node": "22.20.0",
85
179
  "typescript": "6.0.3",
86
180
  "vitest": "4.1.8",
87
- "@dsh-enhanced/assistant-automations": "0.1.23",
88
- "@dsh-enhanced/assistant-delivery": "0.1.23",
89
- "@dsh-enhanced/assistant-evolution": "0.1.23"
181
+ "@deepseek-ai/dsh-session-persistence-jsonl": "0.1.2-rc.1",
182
+ "@deepseek-ai/dsh-session-persistence": "0.1.2-rc.1",
183
+ "@deepseek-ai/dsh-goal": "0.1.2-rc.1",
184
+ "@deepseek-ai/dsh-tool-goal": "0.1.2-rc.1",
185
+ "@deepseek-ai/dsh-goal-round-driver": "0.1.2-rc.1",
186
+ "@deepseek-ai/dsh-subagent": "0.1.2-rc.1",
187
+ "@dsh-enhanced/assistant-delivery": "0.1.30",
188
+ "@dsh-enhanced/assistant-evolution": "0.1.30",
189
+ "@dsh-enhanced/assistant-verifier": "0.1.30",
190
+ "@dsh-enhanced/assistant-automations": "0.1.30",
191
+ "@dsh-enhanced/personal-memory": "0.1.30",
192
+ "@dsh-enhanced/assistant-policy": "0.1.30",
193
+ "@dsh-enhanced/assistant-deepseek-budget": "0.1.30",
194
+ "@dsh-enhanced/assistant-goals": "0.1.30",
195
+ "@dsh-enhanced/assistant-isolation": "0.1.30"
90
196
  },
91
197
  "dependencies": {
92
- "@deepseek-ai/schemastery": "3.18.1"
198
+ "@deepseek-ai/schemastery": "3.18.2",
199
+ "@dsh-enhanced/task-acceptance-contract": "0.1.30"
93
200
  },
94
201
  "scripts": {
95
202
  "build": "tsc -p tsconfig.build.json",