SnowFX commited on
Commit
802b835
·
verified ·
1 Parent(s): d8cd3a3

Review fixes: reviewed/changed-row fields, answers without code notes, grading rules, sample seeds, citation

Browse files
Files changed (3) hide show
  1. README.md +89 -19
  2. simpleqa_200.jsonl +0 -0
  3. simpleqa_50.jsonl +0 -0
README.md CHANGED
@@ -30,6 +30,17 @@ questions. L.A.B. is an open-source web research agent built for small models: i
30
  in a real browser, and code checks every finding against the page it came from before it counts. Its answers are
31
  written only from those findings, each with its source.
32
 
 
 
 
 
 
 
 
 
 
 
 
33
  ## Results
34
 
35
  **`simpleqa_200`** (default): the 200 questions OpenAI's [simple-evals](https://github.com/openai/simple-evals) runs
@@ -41,18 +52,27 @@ with `--examples 200`, on L.A.B. 1.66 with Phi-4 through OpenRouter, November 16
41
  | Claude, judgement calls reviewed by the author | **170 (85%)** | 11 | 19 | **0.892** |
42
  | OpenAI's simple-evals grader (GPT-4.1) | 171 (85.5%) | 15 | 14 | 0.886 |
43
 
44
- When it answered, it was right 94% of the time. The two graders give the same grade on 194 of the 200.
 
 
45
 
46
- **`simpleqa_50`**: six runs on an earlier random 50 (seed 20261106), including the models alone with no browsing:
 
 
 
 
 
 
 
47
 
48
- | Run | Correct | Wrong | Not attempted | F-score | OpenAI's grader: correct |
49
- |---|---|---|---|---|---|
50
- | L.A.B. 1.66, Phi-4 | 45 | 4 | 1 | 0.909 | 46 |
51
- | L.A.B. 1.65.1, Phi-4 (Nov 15) | 39 | 5 | 6 | 0.830 | 39 |
52
- | L.A.B. 1.65.1, Phi-4 (Nov 11) | 38 | 6 | 6 | 0.809 | 38 |
53
- | Phi-4 alone, no browsing | 1 | 34 | 15 | 0.024 | 3 |
54
- | L.A.B. 1.64, Qwen2.5-7B, local on a CPU (no GPU) | 35 | 6 | 9 | 0.769 | 36 |
55
- | Qwen2.5-7B alone, no browsing | 0 | 13 | 37 | 0 | 0 |
56
 
57
  ## Fields
58
 
@@ -64,33 +84,83 @@ When it answered, it was right 94% of the time. The two graders give the same gr
64
  | `question_number` | The question's number in its sample (1 to 200, or 1 to 50) |
65
  | `simpleqa_index` | The question's row in SimpleQA's `simple_qa_test_set.csv` (0-based) |
66
  | `topic`, `question`, `reference_answer` | From SimpleQA |
67
- | `answer` | The final answer, exactly as written, with its citation numbers |
 
68
  | `cited_discoveries` | The discovery numbers the answer cites |
69
  | `discoveries` | Every finding the agent recorded: its number, the page's URL and the text taken from the page |
70
  | `grade` | CORRECT, INCORRECT or NOT_ATTEMPTED: graded by Claude, judgement calls reviewed by the author |
71
  | `grade_note` | Why, for every wrong answer, judgement call and borderline grade |
72
  | `grade_openai` | The grade from simple-evals' own grader prompt and model (GPT-4.1) |
73
-
74
- ## How it was run and graded
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
75
 
76
  - Each question was given to the agent as an ordinary objective, with 60 minutes and nothing carried over from the
77
  question before.
78
  - Sites that republish SimpleQA with its answers were blocked, so every answer had to be found on the open web.
79
  - Search: DuckDuckGo's free HTML results, not a paid search API.
80
- - Grading: SimpleQA's three grades. Claude graded each answer against the reference, one at a time, before seeing
81
- the automatic grades. The author then reviewed every judgement call and borderline grade (48 in all; 2 changed).
82
- OpenAI's grader is reported beside it as a second, automatic grade.
83
  - The questions, scripts, reports and grading notes are in the repository's
84
  [benchmarks/simpleqa](https://github.com/SnowFXXX/LocalAgentBrowser/tree/main/benchmarks/simpleqa) folder.
85
- `run.py sample --size 200 --simple-evals` draws the same 200 from the dataset file.
86
 
87
  ## Caveats
88
 
89
  - Every score here is self-reported by the project that built the agent.
90
- - 200 questions give a likely range of about 79% to 89% for 85%; 50 questions about 79% to 96% for 45 of 50.
91
- - Phi-4 is a 14B model run in the cloud for these runs. The local run is a different model, Qwen2.5-7B, on a CPU.
92
  - SimpleQA has been public since 2024, and some of its reference answers are known to be wrong.
93
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
94
  ## License
95
 
96
  MIT, like the LocalAgentBrowser repository. SimpleQA's questions and reference answers are © OpenAI under the MIT
 
30
  in a real browser, and code checks every finding against the page it came from before it counts. Its answers are
31
  written only from those findings, each with its source.
32
 
33
+ ## In short
34
+
35
+ - **85% of the standard 200, and right 94% of the time when it answered** (170 of 181), with Phi-4 (14B).
36
+ When it couldn't verify an answer, it said so: 19 not attempted, 11 wrong.
37
+ - **Two independent graders agree on 194 of the 200:** Claude, with the judgement calls reviewed by the author, and
38
+ OpenAI's own simple-evals grader (GPT-4.1). They count 170 and 171 correct.
39
+ - **The same model alone, with no browsing, answered 1 of 50; with L.A.B., 45 of the same 50.**
40
+ - **On a CPU with no GPU,** Qwen2.5-7B in L.A.B. answered 35 of 50. For comparison, the Fathom-DeepResearch paper
41
+ ([arXiv 2509.24107](https://arxiv.org/abs/2509.24107), Table 1) reports 50.8% for Qwen2.5-7B with plain search;
42
+ that is a different set of questions and grader, so it's a reference point, not a head-to-head result.
43
+
44
  ## Results
45
 
46
  **`simpleqa_200`** (default): the 200 questions OpenAI's [simple-evals](https://github.com/openai/simple-evals) runs
 
52
  | Claude, judgement calls reviewed by the author | **170 (85%)** | 11 | 19 | **0.892** |
53
  | OpenAI's simple-evals grader (GPT-4.1) | 171 (85.5%) | 15 | 14 | 0.886 |
54
 
55
+ With 200 questions, the likely range for 85% is about 79% to 89% (Wilson, 95%).
56
+
57
+ **`simpleqa_50`**: six runs on an earlier random 50, including the models alone with no browsing:
58
 
59
+ | Run | Correct | Wrong | Not attempted | F-score | OpenAI's grader: correct | Likely range |
60
+ |---|---|---|---|---|---|---|
61
+ | L.A.B. 1.66, Phi-4 | 45 | 4 | 1 | 0.909 | 46 | 79–96% |
62
+ | L.A.B. 1.65.1, Phi-4 (Nov 15) | 39 | 5 | 6 | 0.830 | 39 | 65–87% |
63
+ | L.A.B. 1.65.1, Phi-4 (Nov 11) | 38 | 6 | 6 | 0.809 | 38 | 63–86% |
64
+ | Phi-4 alone, no browsing | 1 | 34 | 15 | 0.024 | 3 | 0–11% |
65
+ | L.A.B. 1.64, Qwen2.5-7B, local on a CPU (no GPU) | 35 | 6 | 9 | 0.769 | 36 | 56–81% |
66
+ | Qwen2.5-7B alone, no browsing | 0 | 13 | 37 | 0 | 0 | 0–7% |
67
 
68
+ ## The questions
69
+
70
+ - **The 200** are exactly the draw simple-evals makes for `--examples 200`: `random.Random(0).sample` over the rows of
71
+ `simple_qa_test_set.csv`, in the order drawn (seed 0, simple-evals' own). Anyone running simple-evals with
72
+ `--examples 200` gets the same questions.
73
+ - **The 50** were drawn with `random.Random(20261106).sample` over the same rows, then sorted. One question is in both.
74
+ - `simpleqa_index` is each question's 0-based row in `simple_qa_test_set.csv`. The repository's `run.py sample`
75
+ redraws either set and checks the file's SHA-256 first.
76
 
77
  ## Fields
78
 
 
84
  | `question_number` | The question's number in its sample (1 to 200, or 1 to 50) |
85
  | `simpleqa_index` | The question's row in SimpleQA's `simple_qa_test_set.csv` (0-based) |
86
  | `topic`, `question`, `reference_answer` | From SimpleQA |
87
+ | `answer` | The final answer exactly as it was graded, including any notes L.A.B.'s code added (see below) |
88
+ | `answer_without_notes` | The same answer with those code notes removed, for reading |
89
  | `cited_discoveries` | The discovery numbers the answer cites |
90
  | `discoveries` | Every finding the agent recorded: its number, the page's URL and the text taken from the page |
91
  | `grade` | CORRECT, INCORRECT or NOT_ATTEMPTED: graded by Claude, judgement calls reviewed by the author |
92
  | `grade_note` | Why, for every wrong answer, judgement call and borderline grade |
93
  | `grade_openai` | The grade from simple-evals' own grader prompt and model (GPT-4.1) |
94
+ | `reviewed_by_author` | True for the 45 rows whose grade the author reviewed (48 review items; three rows came up twice) |
95
+ | `grade_changed_in_review` | True for the 2 rows the author changed: `simpleqa_200` q2 and q157 |
96
+ | `grade_before_review` | Claude's grade before the change, on those 2 rows |
97
+
98
+ ## How it was graded
99
+
100
+ - **SimpleQA's three grades.** CORRECT: contains the reference answer and contradicts nothing. INCORRECT: a different
101
+ answer, a contradiction, or a hedge between candidates ("22 or 23"). NOT_ATTEMPTED: says it couldn't find out, with
102
+ no wrong answer.
103
+ - **Partial answers** that give part of the reference and nothing wrong ("May 2019" for May 28, 2019; "May 2000" for
104
+ 2 May 2000) are NOT_ATTEMPTED in `grade`, following SimpleQA's definition. OpenAI's grader counts those two as
105
+ INCORRECT; both grades are kept.
106
+ - **Order and review.** Claude graded each answer against the reference, one at a time, before seeing the automatic
107
+ grades, with a note for every wrong answer, judgement call and borderline grade. The author reviewed every one of
108
+ them (18 in the 50-question runs, 30 in the 200), agreed with 46 and changed 2. OpenAI's grader ran afterwards on
109
+ the same `answer` text.
110
+ - **Why a second grader:** L.A.B.'s principle is not taking a model's word for it, so the grades shouldn't rest on one
111
+ model either. simple-evals' grader is the field's shared standard, and it agrees on 194 of the 200.
112
+
113
+ ## Notes from L.A.B.'s code inside answers
114
+
115
+ Some answers end with notes added by L.A.B.'s checks, not by the model:
116
+ - **"Left out, because no discovery gives these numbers (the pages read may still have them): …"**: a number or name
117
+ the model wrote was removed because no recorded finding contains it.
118
+ - **"Not found in the pages read: …"**: parts of the question no finding answered. On some single-answer questions
119
+ this list is garbled ("ten more (only none of the ten were found)") because the check read a number in the question
120
+ ("track 10") as a count. A known bug, being fixed; it changed no grade.
121
+ - **"> ⚠️ Numbers not found in the recorded discoveries"**: the same check, as a warning.
122
+
123
+ The answers are kept exactly as graded, notes included; `answer_without_notes` has them removed. Whether a number
124
+ inside a "Left out" note counts as the answer is the author's judgement call row by row: in q2 of the 200 it was
125
+ counted ("33 years" for 32, INCORRECT), in q88 it wasn't ("2020" for April 2019, NOT_ATTEMPTED). Each row's
126
+ `grade_note` says which.
127
+
128
+ ## How it was run
129
 
130
  - Each question was given to the agent as an ordinary objective, with 60 minutes and nothing carried over from the
131
  question before.
132
  - Sites that republish SimpleQA with its answers were blocked, so every answer had to be found on the open web.
133
  - Search: DuckDuckGo's free HTML results, not a paid search API.
 
 
 
134
  - The questions, scripts, reports and grading notes are in the repository's
135
  [benchmarks/simpleqa](https://github.com/SnowFXXX/LocalAgentBrowser/tree/main/benchmarks/simpleqa) folder.
 
136
 
137
  ## Caveats
138
 
139
  - Every score here is self-reported by the project that built the agent.
140
+ - "Small models" here means 7B on a CPU and Phi-4 (14B) in the cloud: small next to frontier models, but the 200 ran
141
+ on a cloud model, not on a CPU. A 200-question run on the local 7B is planned.
142
  - SimpleQA has been public since 2024, and some of its reference answers are known to be wrong.
143
 
144
+ ## Citation
145
+
146
+ ```bibtex
147
+ @misc{snowfx2026labsimpleqa,
148
+ author = {SnowFX},
149
+ title = {LocalAgentBrowser on SimpleQA: answers, citations and grades},
150
+ year = {2026},
151
+ publisher = {Hugging Face},
152
+ howpublished = {\url{https://huggingface.co/datasets/SnowFX/localagentbrowser-simpleqa}}
153
+ }
154
+
155
+ @misc{wei2024simpleqa,
156
+ author = {Wei, Jason and others},
157
+ title = {Measuring short-form factuality in large language models},
158
+ year = {2024},
159
+ eprint = {2411.04368},
160
+ archivePrefix = {arXiv}
161
+ }
162
+ ```
163
+
164
  ## License
165
 
166
  MIT, like the LocalAgentBrowser repository. SimpleQA's questions and reference answers are © OpenAI under the MIT
simpleqa_200.jsonl CHANGED
The diff for this file is too large to render. See raw diff
 
simpleqa_50.jsonl CHANGED
The diff for this file is too large to render. See raw diff