-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy patheval.json
More file actions
237 lines (237 loc) · 7.66 KB
/
Copy patheval.json
File metadata and controls
237 lines (237 loc) · 7.66 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
{
"run": {
"measured_at": "2026-09-09",
"bundle": "knowledge",
"concepts": 4,
"index": "grounded-context-corpus",
"elasticsearch_configured": true,
"git_sha": "5675c04"
},
"totals": {
"pass": 19,
"known": 1,
"fail": 0
},
"cases": [
{
"id": "Q1",
"question": "What is the exact context window of claude-opus-5?",
"expected": "deterministic",
"actual": "deterministic",
"expected_route": "DETERMINISTIC",
"route": "DETERMINISTIC",
"rationale": "precision phrasing (\"context window\", \"exact\") plus a named model (claude-opus-5) \u2014 an exact fact must not be ranked",
"citations": 1,
"verdict": "PASS"
},
{
"id": "Q2",
"question": "What is the endpoint path for Anthropic's Messages API?",
"expected": "deterministic",
"actual": "deterministic",
"expected_route": "DETERMINISTIC",
"route": "DETERMINISTIC",
"rationale": "precision phrasing (\"endpoint\") \u2014 an exact fact must not be ranked",
"citations": 1,
"verdict": "PASS"
},
{
"id": "Q3",
"question": "Which of these models support vision?",
"expected": "mixed",
"actual": "mixed",
"expected_route": "BOTH",
"route": "BOTH",
"rationale": "no decisive signal \u2014 defaulting to BOTH, which is the safe side",
"citations": 8,
"verdict": "PASS"
},
{
"id": "Q4",
"question": "What is the max output tokens for claude-haiku-4-5?",
"expected": "deterministic",
"actual": "deterministic",
"expected_route": "DETERMINISTIC",
"route": "DETERMINISTIC",
"rationale": "precision phrasing (\"max output\") plus a named model (claude-haiku-4-5) \u2014 an exact fact must not be ranked",
"citations": 1,
"verdict": "PASS"
},
{
"id": "Q5",
"question": "How do I stream responses from the API?",
"expected": "semantic",
"actual": "semantic",
"expected_route": "SEMANTIC",
"route": "SEMANTIC",
"rationale": "exploratory phrasing (\"how do i\"), no exact field requested",
"citations": 5,
"verdict": "PASS"
},
{
"id": "Q6",
"question": "What's the recommended way to do hybrid search in Elasticsearch?",
"expected": "semantic",
"actual": "semantic",
"expected_route": "SEMANTIC",
"route": "SEMANTIC",
"rationale": "exploratory phrasing (\"recommended\"), no exact field requested",
"citations": 5,
"verdict": "PASS"
},
{
"id": "Q7",
"question": "How should I chunk documents for retrieval?",
"expected": "semantic",
"actual": "semantic",
"expected_route": "SEMANTIC",
"route": "SEMANTIC",
"rationale": "exploratory phrasing (\"how should i\", \"should i\"), no exact field requested",
"citations": 5,
"verdict": "PASS"
},
{
"id": "Q8",
"question": "What's the difference between BM25 and vector search?",
"expected": "semantic",
"actual": "semantic",
"expected_route": "SEMANTIC",
"route": "SEMANTIC",
"rationale": "exploratory phrasing (\"difference between\"), no exact field requested",
"citations": 5,
"verdict": "PASS"
},
{
"id": "Q9",
"question": "What does the rank_constant parameter do?",
"expected": "semantic",
"actual": "semantic",
"expected_route": "BOTH",
"route": "BOTH",
"rationale": "no decisive signal \u2014 defaulting to BOTH, which is the safe side",
"citations": 5,
"verdict": "PASS"
},
{
"id": "Q10",
"question": "Compare claude-opus-5 and claude-sonnet-5 on max output tokens.",
"expected": "mixed",
"actual": "mixed",
"expected_route": "BOTH",
"route": "BOTH",
"rationale": "cross-entity comparison (\"compare\") \u2014 exact values are required, so a deterministic miss refuses rather than falling back to passages",
"citations": 7,
"verdict": "PASS"
},
{
"id": "Q11",
"question": "What is the price per million tokens of GPT-5?",
"expected": "refusal",
"actual": "refusal",
"expected_route": "DETERMINISTIC",
"route": "DETERMINISTIC",
"rationale": "precision phrasing (\"price\") \u2014 an exact fact must not be ranked",
"citations": 0,
"verdict": "PASS"
},
{
"id": "Q12",
"question": "What is the exact context window of claude-haiku-4-5?",
"expected": "deterministic",
"actual": "deterministic",
"expected_route": "DETERMINISTIC",
"route": "DETERMINISTIC",
"rationale": "precision phrasing (\"context window\", \"exact\") plus a named model (claude-haiku-4-5) \u2014 an exact fact must not be ranked",
"citations": 1,
"verdict": "PASS"
},
{
"id": "Q13",
"question": "Whats the ctx window for opus",
"expected": "deterministic",
"actual": "deterministic",
"expected_route": "DETERMINISTIC",
"route": "DETERMINISTIC",
"rationale": "precision phrasing (\"ctx window\") \u2014 an exact fact must not be ranked",
"citations": 1,
"verdict": "PASS"
},
{
"id": "Q14",
"question": "How big is the Opus 5 context window?",
"expected": "deterministic",
"actual": "deterministic",
"expected_route": "DETERMINISTIC",
"route": "DETERMINISTIC",
"rationale": "precision phrasing (\"context window\") \u2014 an exact fact must not be ranked",
"citations": 1,
"verdict": "PASS"
},
{
"id": "Q15",
"question": "What is the context length of Sonnet 5",
"expected": "deterministic",
"actual": "deterministic",
"expected_route": "DETERMINISTIC",
"route": "DETERMINISTIC",
"rationale": "precision phrasing (\"context length\") \u2014 an exact fact must not be ranked",
"citations": 1,
"verdict": "PASS"
},
{
"id": "Q16",
"question": "Haiku 4.5 max output",
"expected": "deterministic",
"actual": "deterministic",
"expected_route": "DETERMINISTIC",
"route": "DETERMINISTIC",
"rationale": "precision phrasing (\"max output\") \u2014 an exact fact must not be ranked",
"citations": 1,
"verdict": "PASS"
},
{
"id": "Q17",
"question": "the Opus model endpoint",
"expected": "deterministic",
"actual": "deterministic",
"expected_route": "DETERMINISTIC",
"route": "DETERMINISTIC",
"rationale": "precision phrasing (\"endpoint\") \u2014 an exact fact must not be ranked",
"citations": 1,
"verdict": "PASS"
},
{
"id": "Q18",
"question": "Is Sonnet 5 cheaper than Opus 5?",
"expected": "refusal",
"actual": "refusal",
"expected_route": "BOTH",
"route": "BOTH",
"rationale": "cross-entity comparison (\"cheaper\") \u2014 exact values are required, so a deterministic miss refuses rather than falling back to passages",
"citations": 0,
"verdict": "PASS"
},
{
"id": "Q19",
"question": "How do I bake sourdough bread?",
"expected": "refusal",
"actual": "refusal",
"expected_route": "SEMANTIC",
"route": "SEMANTIC",
"rationale": "exploratory phrasing (\"how do i\"), no exact field requested",
"citations": 0,
"verdict": "PASS"
},
{
"id": "Q20",
"question": "What is the best way to train for a marathon?",
"expected": "refusal",
"actual": "semantic",
"expected_route": "SEMANTIC",
"route": "SEMANTIC",
"rationale": "exploratory phrasing (\"best way\"), no exact field requested",
"citations": 5,
"verdict": "KNOWN"
}
]
}