example_id
string | run_id
string | domain
string | task_type
string | difficulty
string | query
string | gold_answer
string | answer_tokens
int64 | is_correct
int64 | correctness_label
string | faithfulness_label
string | hallucination_flag
bool | retrieval_strategy
string | chunking_strategy
string | n_retrieved_chunks
int64 | top1_score
float64 | mean_retrieved_score
float64 | recall_at_5
float64 | recall_at_10
float64 | mrr_at_10
float64 | used_long_context_window
bool | context_window_tokens
int64 | latency_ms_retrieval
int64 | latency_ms_generation
int64 | total_latency_ms
int64 | embedding_model
string | reranker_model
string | doc_ids_used
string | chunk_ids_used
string | supervising_judge_label
string | eval_mode
string | user_feedback_label
string | created_at_utc
string | generator_model
string | temperature
float64 | top_p
float64 | max_new_tokens
int64 | stop_reason
string | prompt_tokens
int64 | total_cost_usd
float64 | scenario_id
string | has_answer_in_corpus
int64 | is_noanswer_probe
int64 | has_relevant_in_top5
int64 | has_relevant_in_top10
int64 | answered_without_retrieval
int64 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
QA000001
|
run_0
|
financial_reports
|
multi_hop
|
medium
|
Which segment contributed most to revenue growth?
|
The enterprise subscription segment contributed the largest share of growth.
| 175
| 1
|
correct
|
faithful
| false
|
hybrid
|
fixed_512_tokens
| 38
| 0.812784
| 0.298053
| 0
| 0
| 0
| false
| 4,096
| 220
| 849
| 1,108
|
all-minilm-l12-v2
|
colbert-v2
|
DOC0523|DOC0323|DOC0426|DOC0253
|
C04183|C02494|C03407|C01959
|
pass
|
human_eval
|
positive
|
2025-03-31T01:04:00Z
|
mistral-large
| 0.65
| 0.88
| 324
|
error
| 615
| 0.00054
|
SC007
| 1
| 0
| 0
| 0
| 1
|
QA000002
|
run_1
|
financial_reports
|
factoid
|
hard
|
Which segment contributed most to revenue growth?
|
The enterprise subscription segment contributed the largest share of growth.
| 149
| 0
|
incorrect
|
unknown
| false
|
bm25
|
sliding_window_256_overlap
| 22
| 0.181427
| 0.41487
| 0.375
| 1
| 0.5
| true
| 16,384
| 164
| 987
| 1,220
|
text-embedding-3-large
|
gte-qwen-reranker
|
DOC0502|DOC0631|DOC0206|DOC0240
|
C03969|C05043|C01579|C01864
|
fail
|
online_shadow
|
negative
|
2025-03-02T21:07:00Z
|
gpt-4o
| 0.49
| 0.78
| 337
|
eos
| 588
| 0.003118
|
SC008
| 1
| 0
| 1
| 1
| 0
|
QA000003
|
run_2
|
financial_reports
|
explanation
|
hard
|
Which segment contributed most to revenue growth?
|
The enterprise subscription segment contributed the largest share of growth.
| 249
| 0
|
incorrect
|
unfaithful
| true
|
dense
|
sliding_window_256_overlap
| 29
| 0.468516
| 0.306921
| 0
| 1
| 0.166667
| false
| 4,096
| 224
| 1,911
| 2,187
|
bge-m3
|
colbert-v2
|
DOC0587|DOC0371|DOC0533|DOC0323
|
C04650|C02964|C04250|C02491
|
borderline
|
human_eval
|
negative
|
2025-03-11T08:08:00Z
|
gpt-4o
| 0.55
| 0.88
| 351
|
eos
| 628
| 0.004167
|
SC036
| 1
| 0
| 0
| 1
| 0
|
QA000004
|
run_3
|
hr_policies
|
explanation
|
medium
|
What is the standard probation period for new hires?
|
The typical probation period is three to six months depending on role.
| 149
| 1
|
correct
|
faithful
| false
|
bm25
|
sliding_window_256_overlap
| 40
| 0.776064
| 0.309426
| 0.5
| 1
| 1
| true
| 8,192
| 271
| 525
| 822
|
e5-mistral-7b
|
jina-reranker-v1
|
DOC0118|DOC0329|DOC0356|DOC0573
|
C00875|C02553|C02859|C04538
|
pass
|
online_shadow
|
neutral
|
2025-03-26T06:15:00Z
|
llama-3.1-8b-instruct
| 0.61
| 0.72
| 206
|
eos
| 489
| 0.000084
|
SC043
| 1
| 0
| 1
| 1
| 0
|
QA000005
|
run_4
|
medical_guides
|
multi_hop
|
hard
|
Which symptoms indicate the need for immediate escalation?
|
Red-flag symptoms include chest pain, severe shortness of breath, and confusion.
| 193
| 1
|
correct
|
faithful
| false
|
hybrid
|
by_heading
| 37
| 0.784336
| 0.339827
| 0.444444
| 1
| 1
| true
| 16,384
| 198
| 679
| 938
|
text-embedding-3-small
|
bge-reranker-base
|
DOC0500|DOC0391|DOC0509|DOC0442
|
C03944|C03135|C04062|C03526
|
borderline
|
human_eval
|
positive
|
2025-03-03T02:23:00Z
|
gpt-4o-mini
| 0.29
| 0.76
| 318
|
eos
| 681
| 0.00023
|
SC010
| 1
| 0
| 1
| 1
| 0
|
QA000006
|
run_5
|
developer_docs
|
table_qa
|
medium
|
How do I authenticate using API keys?
|
API keys are passed as a bearer token in the Authorization header.
| 127
| 1
|
correct
|
faithful
| false
|
hybrid
|
fixed_512_tokens
| 28
| 0.848851
| 0.388729
| 0.444444
| 1
| 1
| false
| 4,096
| 152
| 862
| 1,120
|
gte-large
|
bge-reranker-base
|
DOC0515|DOC0583|DOC0076|DOC0406
|
C04136|C04613|C00548|C03238
|
pass
|
online_shadow
|
neutral
|
2025-03-16T23:20:00Z
|
llama-3.1-70b-instruct
| 0.85
| 0.86
| 222
|
eos
| 392
| 0.000369
|
SC005
| 1
| 0
| 1
| 1
| 0
|
QA000007
|
run_6
|
product_docs
|
temporal_reasoning
|
medium
|
How does the billing cycle work for annual plans?
|
Annual plans are billed once per year with proration on upgrades.
| 37
| 1
|
correct
|
faithful
| false
|
dense_then_rerank
|
semantic
| 27
| 1
| 0.442011
| 0.5
| 1
| 1
| false
| 4,096
| 165
| 296
| 544
|
bge-m3
|
gte-qwen-reranker
|
DOC0326|DOC0244|DOC0520|DOC0320
|
C02540|C01901|C04169|C02475
|
pass
|
human_eval
|
neutral
|
2025-03-22T15:28:00Z
|
claude-3.5-sonnet
| 0.38
| 0.7
| 96
|
eos
| 128
| 0.00093
|
SC003
| 1
| 0
| 1
| 1
| 0
|
QA000008
|
run_7
|
developer_docs
|
factoid
|
easy
|
What is the rate limit for write requests?
|
Write requests are limited per minute and per project ID.
| 68
| 1
|
correct
|
faithful
| false
|
bm25
|
fixed_512_tokens
| 29
| 0.625541
| 0.36997
| 0.625
| 1
| 1
| false
| 4,096
| 167
| 317
| 554
|
text-embedding-3-small
|
gte-qwen-reranker
|
DOC0331|DOC0600|DOC0410
|
C02570|C04779|C02578|C03291
|
pass
|
online_shadow
|
positive
|
2025-02-24T21:51:00Z
|
llama-3.1-8b-instruct
| 0.58
| 0.77
| 138
|
eos
| 165
| 0.000028
|
SC031
| 1
| 0
| 1
| 1
| 0
|
QA000009
|
run_8
|
financial_reports
|
explanation
|
easy
|
Which segment contributed most to revenue growth?
|
The enterprise subscription segment contributed the largest share of growth.
| 87
| 1
|
correct
|
faithful
| false
|
dense
|
fixed_512_tokens
| 24
| 0.68217
| 0.43631
| 0.555556
| 1
| 1
| false
| 8,192
| 119
| 254
| 450
|
e5-mistral-7b
|
bge-reranker-base
|
DOC0589|DOC0343|DOC0065|DOC0476
|
C04662|C02725|C00452|C03750
|
pass
|
offline_batch
|
neutral
|
2025-03-21T21:11:00Z
|
gpt-4o-mini
| 0.55
| 0.75
| 211
|
eos
| 287
| 0.000096
|
SC033
| 1
| 0
| 1
| 1
| 0
|
QA000010
|
run_9
|
financial_reports
|
instruction_following
|
medium
|
How did operating margin change year over year?
|
Operating margin improved by approximately two percentage points.
| 198
| 1
|
correct
|
faithful
| false
|
dense_then_rerank
|
semantic
| 34
| 0.526988
| 0.366279
| 0.5
| 1
| 0.5
| false
| 16,384
| 209
| 916
| 1,183
|
e5-mistral-7b
|
gte-qwen-reranker
|
DOC0065|DOC0575|DOC0586|DOC0601
|
C00467|C04556|C04627|C04798
|
pass
|
online_shadow
|
neutral
|
2025-02-08T14:17:00Z
|
llama-3.1-8b-instruct
| 0.78
| 0.8
| 383
|
eos
| 586
| 0.000104
|
SC034
| 1
| 0
| 1
| 1
| 0
|
QA000011
|
run_10
|
product_docs
|
explanation
|
hard
|
How does the billing cycle work for annual plans?
|
Annual plans are billed once per year with proration on upgrades.
| 151
| 1
|
correct
|
faithful
| false
|
bm25
|
by_heading
| 29
| 0.805435
| 0.347905
| 0.5
| 1
| 1
| false
| 4,096
| 157
| 1,033
| 1,225
|
text-embedding-3-small
|
jina-reranker-v1
|
DOC0036|DOC0647|DOC0447|DOC0394
|
C00281|C05175|C03560|C03162
|
fail
|
human_eval
|
very_positive
|
2025-03-08T10:55:00Z
|
claude-3.5-sonnet
| 0.19
| 0.96
| 247
|
eos
| 555
| 0.003857
|
SC020
| 1
| 0
| 1
| 1
| 0
|
QA000012
|
run_11
|
medical_guides
|
multi_hop
|
hard
|
What is the recommended dosage for adult patients?
|
Dosage is weight-adjusted and must follow local clinical guidelines.
| 82
| 0
|
incorrect
|
unfaithful
| true
|
bm25
|
semantic
| 34
| 0.880359
| 0.324304
| 0.5
| 1
| 1
| true
| 8,192
| 235
| 276
| 602
|
gte-large
|
gte-qwen-reranker
|
DOC0422|DOC0556|DOC0186|DOC0354
|
C03366|C04427|C01443|C02821
|
fail
|
human_eval
|
negative
|
2025-03-09T14:22:00Z
|
llama-3.1-8b-instruct
| 0.81
| 0.98
| 208
|
eos
| 179
| 0.000034
|
SC038
| 1
| 0
| 1
| 1
| 0
|
QA000013
|
run_12
|
support_faq
|
multi_hop
|
easy
|
Where can I check the status of my support ticket?
|
Ticket status is available on the support portal under 'My Requests'.
| 240
| 1
|
correct
|
faithful
| false
|
bm25
|
by_heading
| 14
| 0.703791
| 0.632208
| 0.555556
| 1
| 1
| true
| 16,384
| 146
| 984
| 1,248
|
e5-mistral-7b
|
cross-encoder-ms-marco
|
DOC0270|DOC0519|DOC0241|DOC0623
|
C02054|C04167|C01885|C04979
|
pass
|
human_eval
|
no_feedback
|
2025-03-09T02:59:00Z
|
mistral-large
| 0.21
| 0.75
| 313
|
eos
| 496
| 0.00061
|
SC002
| 1
| 0
| 1
| 1
| 0
|
QA000014
|
run_13
|
medical_guides
|
comparison
|
medium
|
Which symptoms indicate the need for immediate escalation?
|
Red-flag symptoms include chest pain, severe shortness of breath, and confusion.
| 108
| 1
|
correct
|
faithful
| false
|
bm25
|
by_heading
| 13
| 0.562793
| 0.612658
| 0.444444
| 1
| 0.5
| false
| 32,768
| 127
| 267
| 509
|
gte-large
|
jina-reranker-v1
|
DOC0354|DOC0199|DOC0191|DOC0186
|
C02831|C01549|C01477|C01434
|
pass
|
offline_batch
|
very_positive
|
2025-03-03T06:55:00Z
|
gpt-4o-mini
| 0.22
| 0.77
| 174
|
eos
| 296
| 0.000099
|
SC009
| 1
| 0
| 1
| 1
| 0
|
QA000015
|
run_14
|
financial_reports
|
summarization
|
medium
|
Which segment contributed most to revenue growth?
|
The enterprise subscription segment contributed the largest share of growth.
| 65
| 0
|
incorrect
|
unfaithful
| true
|
dense_then_rerank
|
by_heading
| 9
| 0.770268
| 0.761919
| 0
| 1
| 0.166667
| true
| 16,384
| 113
| 341
| 502
|
e5-mistral-7b
|
cross-encoder-ms-marco
|
DOC0343|DOC0333|DOC0218|DOC0070
|
C02723|C02590|C01688|C00505
|
borderline
|
human_eval
|
very_negative
|
2025-03-08T11:48:00Z
|
llama-3.1-70b-instruct
| 0.79
| 0.95
| 225
|
eos
| 192
| 0.000172
|
SC035
| 1
| 0
| 0
| 1
| 0
|
QA000016
|
run_15
|
hr_policies
|
instruction_following
|
medium
|
What is the standard probation period for new hires?
|
The typical probation period is three to six months depending on role.
| 217
| 0
|
partial
|
unfaithful
| true
|
hybrid
|
by_heading
| 7
| 0.755746
| 0.68227
| 0
| 1
| 0.166667
| true
| 8,192
| 96
| 1,434
| 1,602
|
e5-mistral-7b
|
jina-reranker-v1
|
DOC0593|DOC0118|DOC0139|DOC0229
|
C04707|C00864|C01091|C01776
|
borderline
|
offline_batch
|
very_negative
|
2025-02-17T04:02:00Z
|
gpt-4o
| 0.31
| 0.75
| 353
|
eos
| 712
| 0.004114
|
SC046
| 1
| 0
| 0
| 1
| 0
|
QA000017
|
run_16
|
hr_policies
|
instruction_following
|
medium
|
What is the standard probation period for new hires?
|
The typical probation period is three to six months depending on role.
| 137
| 1
|
correct
|
faithful
| false
|
dense
|
by_heading
| 34
| 0.824767
| 0.345688
| 0.5
| 1
| 1
| false
| 32,768
| 138
| 980
| 1,204
|
all-minilm-l12-v2
|
bge-reranker-base
|
DOC0431|DOC0279|DOC0149|DOC0560
|
C03446|C02129|C01161|C04453
|
pass
|
offline_batch
|
positive
|
2025-03-08T15:16:00Z
|
llama-3.1-70b-instruct
| 0.59
| 0.92
| 212
|
eos
| 305
| 0.00032
|
SC043
| 1
| 0
| 1
| 1
| 0
|
QA000018
|
run_17
|
policies
|
temporal_reasoning
|
medium
|
What is the data retention policy for customer logs?
|
Customer logs are retained for 180 days by default.
| 133
| 1
|
correct
|
faithful
| false
|
dense
|
fixed_512_tokens
| 19
| 0.23918
| 0.438487
| 0.285714
| 1
| 0.25
| false
| 32,768
| 151
| 835
| 1,102
|
bge-m3
|
cross-encoder-ms-marco
|
DOC0211|DOC0633|DOC0399|DOC0306
|
C01618|C05049|C03191|C02389
|
pass
|
offline_batch
|
no_feedback
|
2025-03-12T06:33:00Z
|
claude-3.5-sonnet
| 0.47
| 0.73
| 253
|
eos
| 422
| 0.003411
|
SC012
| 1
| 0
| 1
| 1
| 0
|
QA000019
|
run_18
|
policies
|
factoid
|
medium
|
How is user consent stored for analytics tracking?
|
User consent is stored as a hashed record with timestamps.
| 89
| 1
|
correct
|
faithful
| false
|
dense
|
sliding_window_256_overlap
| 42
| 0.696569
| 0.278163
| 0.555556
| 0.888889
| 1
| true
| 16,384
| 272
| 726
| 1,042
|
text-embedding-3-small
|
cross-encoder-ms-marco
|
DOC0305|DOC0328|DOC0592|DOC0633
|
C02377|C02552|C04697|C05083
|
pass
|
human_eval
|
positive
|
2025-02-15T13:16:00Z
|
claude-3.5-sonnet
| 0.85
| 0.89
| 220
|
eos
| 323
| 0.002134
|
SC048
| 1
| 0
| 1
| 1
| 0
|
QA000020
|
run_19
|
hr_policies
|
comparison
|
hard
|
How many days of annual leave are employees entitled to?
|
Employees receive a standard allocation of annual leave as defined in their contract.
| 64
| 0
|
partial
|
unfaithful
| true
|
hybrid
|
sliding_window_256_overlap
| 14
| 0.767621
| 0.459023
| 0.428571
| 1
| 1
| false
| 4,096
| 100
| 504
| 640
|
text-embedding-3-small
|
bge-reranker-base
|
DOC0118|DOC0529|DOC0058|DOC0265
|
C00864|C04224|C00413|C02022
|
borderline
|
offline_batch
|
very_negative
|
2025-02-03T15:42:00Z
|
gpt-4o
| 0.73
| 0.78
| 120
|
eos
| 244
| 0.00125
|
SC045
| 1
| 0
| 1
| 1
| 0
|
QA000021
|
run_20
|
developer_docs
|
comparison
|
easy
|
What is the rate limit for write requests?
|
Write requests are limited per minute and per project ID.
| 187
| 1
|
correct
|
faithful
| false
|
bm25_then_rerank
|
sliding_window_256_overlap
| 14
| 0.862509
| 0.584769
| 0.555556
| 1
| 1
| false
| 4,096
| 94
| 1,070
| 1,296
|
text-embedding-3-large
|
bge-reranker-base
|
DOC0157|DOC0210|DOC0410|DOC0194
|
C01231|C01611|C03286|C01502
|
pass
|
human_eval
|
positive
|
2025-03-28T13:12:00Z
|
mistral-large
| 0.37
| 0.92
| 278
|
eos
| 711
| 0.000573
|
SC032
| 1
| 0
| 1
| 1
| 0
|
QA000022
|
run_21
|
financial_reports
|
multi_hop
|
medium
|
How did operating margin change year over year?
|
Operating margin improved by approximately two percentage points.
| 73
| 0
|
incorrect
|
unknown
| false
|
hybrid
|
fixed_512_tokens
| 44
| 0.775494
| 0.236238
| 0.333333
| 1
| 1
| true
| 16,384
| 295
| 569
| 901
|
bge-m3
|
gte-qwen-reranker
|
DOC0171|DOC0253|DOC0456|DOC0240
|
C01323|C01963|C03617|C01873
|
borderline
|
online_shadow
|
negative
|
2025-02-20T15:16:00Z
|
gpt-4o
| 0.75
| 0.86
| 245
|
eos
| 222
| 0.00126
|
SC007
| 1
| 0
| 1
| 1
| 0
|
QA000023
|
run_22
|
product_docs
|
factoid
|
medium
|
Can I downgrade from enterprise to pro mid-cycle?
|
Downgrades are applied at the end of the current billing period.
| 231
| 1
|
correct
|
faithful
| false
|
dense
|
by_heading
| 16
| 0.978517
| 0.583121
| 0.5
| 0.9
| 1
| false
| 32,768
| 100
| 824
| 1,063
|
gte-large
|
bge-reranker-base
|
DOC0316|DOC0072|DOC0336|DOC0351
|
C02460|C00524|C02636|C02790
|
pass
|
offline_batch
|
very_positive
|
2025-03-13T00:10:00Z
|
gpt-4o-mini
| 0.28
| 0.75
| 331
|
eos
| 877
| 0.000286
|
SC004
| 1
| 0
| 1
| 1
| 0
|
QA000024
|
run_23
|
research_papers
|
table_qa
|
medium
|
How large is the evaluation dataset?
|
The evaluation dataset contains several tens of thousands of examples.
| 252
| 1
|
correct
|
unfaithful
| true
|
hybrid
|
semantic
| 31
| 0.806039
| 0.352838
| 0.444444
| 0.888889
| 1
| false
| 16,384
| 216
| 1,282
| 1,590
|
gte-large
|
jina-reranker-v1
|
DOC0085|DOC0517|DOC0120
|
C00642|C00646|C04152|C00912
|
pass
|
online_shadow
|
negative
|
2025-02-17T08:44:00Z
|
llama-3.1-70b-instruct
| 0.59
| 0.82
| 402
|
eos
| 775
| 0.000677
|
SC060
| 1
| 0
| 1
| 1
| 0
|
QA000025
|
run_24
|
financial_reports
|
summarization
|
easy
|
How did operating margin change year over year?
|
Operating margin improved by approximately two percentage points.
| 82
| 1
|
correct
|
faithful
| false
|
hybrid
|
semantic
| 40
| 0.82732
| 0.320824
| 0.555556
| 1
| 1
| true
| 32,768
| 265
| 244
| 589
|
text-embedding-3-large
|
gte-qwen-reranker
|
DOC0190|DOC0552|DOC0610|DOC0102
|
C01460|C04388|C04851|C00736
|
pass
|
online_shadow
|
very_positive
|
2025-02-13T16:26:00Z
|
gpt-4o-mini
| 0.7
| 0.85
| 217
|
eos
| 176
| 0.000079
|
SC033
| 1
| 0
| 1
| 1
| 0
|
QA000026
|
run_25
|
financial_reports
|
instruction_following
|
medium
|
How did operating margin change year over year?
|
Operating margin improved by approximately two percentage points.
| 47
| 1
|
correct
|
faithful
| false
|
bm25_then_rerank
|
fixed_512_tokens
| 44
| 0.665782
| 0.236789
| 0.333333
| 0.888889
| 1
| true
| 8,192
| 186
| 132
| 361
|
text-embedding-3-large
|
colbert-v2
|
DOC0065|DOC0601|DOC0622
|
C00464|C00451|C04798|C04942
|
pass
|
human_eval
|
neutral
|
2025-02-08T05:18:00Z
|
gpt-4o-mini
| 0.66
| 1
| 96
|
eos
| 186
| 0.000059
|
SC034
| 1
| 0
| 1
| 1
| 0
|
QA000027
|
run_26
|
developer_docs
|
summarization
|
medium
|
How do I authenticate using API keys?
|
API keys are passed as a bearer token in the Authorization header.
| 110
| 1
|
correct
|
faithful
| false
|
hybrid
|
by_heading
| 42
| 0.852941
| 0.308834
| 0.5
| 1
| 1
| true
| 32,768
| 250
| 603
| 976
|
text-embedding-3-large
|
colbert-v2
|
DOC0210|DOC0348|DOC0194|DOC0049
|
C01608|C02760|C01502|C00367
|
pass
|
online_shadow
|
positive
|
2025-02-23T00:40:00Z
|
llama-3.1-70b-instruct
| 0.63
| 0.84
| 294
|
eos
| 221
| 0.000229
|
SC025
| 1
| 0
| 1
| 1
| 0
|
QA000028
|
run_27
|
financial_reports
|
instruction_following
|
hard
|
Which segment contributed most to revenue growth?
|
The enterprise subscription segment contributed the largest share of growth.
| 172
| 1
|
correct
|
faithful
| false
|
bm25_then_rerank
|
semantic
| 6
| 0.394658
| 0.684046
| 0.8
| 1
| 0.5
| true
| 32,768
| 100
| 673
| 814
|
text-embedding-3-large
|
jina-reranker-v1
|
DOC0065|DOC0381|DOC0589|DOC0010
|
C00469|C03072|C04663|C00073
|
borderline
|
offline_batch
|
positive
|
2025-03-15T10:08:00Z
|
llama-3.1-8b-instruct
| 0.76
| 0.75
| 358
|
eos
| 442
| 0.000083
|
SC008
| 1
| 0
| 1
| 1
| 0
|
QA000029
|
run_28
|
policies
|
explanation
|
hard
|
How is user consent stored for analytics tracking?
|
User consent is stored as a hashed record with timestamps.
| 21
| 1
|
correct
|
faithful
| false
|
dense
|
sliding_window_256_overlap
| 8
| 0.758766
| 0.703424
| 0.714286
| 1
| 1
| true
| 32,768
| 99
| 83
| 250
|
all-minilm-l12-v2
|
colbert-v2
|
DOC0002|DOC0434|DOC0156|DOC0630
|
C00014|C03463|C01191|C05035
|
fail
|
offline_batch
|
very_positive
|
2025-03-04T06:47:00Z
|
llama-3.1-8b-instruct
| 0.54
| 0.87
| 165
|
eos
| 128
| 0.000016
|
SC049
| 1
| 0
| 1
| 1
| 0
|
QA000030
|
run_29
|
financial_reports
|
temporal_reasoning
|
hard
|
How did operating margin change year over year?
|
Operating margin improved by approximately two percentage points.
| 105
| 1
|
correct
|
faithful
| false
|
dense
|
sliding_window_256_overlap
| 14
| 0.887056
| 0.590421
| 0.375
| 0.875
| 1
| true
| 16,384
| 157
| 502
| 685
|
gte-large
|
jina-reranker-v1
|
DOC0465|DOC0333|DOC0601|DOC0552
|
C03675|C02592|C04791|C04387
|
fail
|
online_shadow
|
very_positive
|
2025-03-22T14:16:00Z
|
llama-3.1-8b-instruct
| 0.84
| 0.77
| 175
|
eos
| 324
| 0.000049
|
SC036
| 1
| 0
| 1
| 1
| 0
|
QA000031
|
run_30
|
financial_reports
|
explanation
|
easy
|
Which segment contributed most to revenue growth?
|
The enterprise subscription segment contributed the largest share of growth.
| 110
| 1
|
correct
|
faithful
| false
|
dense_then_rerank
|
fixed_512_tokens
| 17
| 0.94423
| 0.445159
| 0.571429
| 1
| 1
| true
| 32,768
| 122
| 598
| 743
|
gte-large
|
colbert-v2
|
DOC0065|DOC0371|DOC0010|DOC0218
|
C00453|C02964|C00069|C01686
|
pass
|
human_eval
|
positive
|
2025-02-06T04:56:00Z
|
llama-3.1-70b-instruct
| 0.22
| 0.96
| 279
|
eos
| 345
| 0.000275
|
SC033
| 1
| 0
| 1
| 1
| 0
|
QA000032
|
run_31
|
hr_policies
|
instruction_following
|
medium
|
How many days of annual leave are employees entitled to?
|
Employees receive a standard allocation of annual leave as defined in their contract.
| 212
| 0
|
incorrect
|
unknown
| false
|
bm25
|
semantic
| 17
| 0.898614
| 0.497971
| 0
| 0
| 0
| false
| 16,384
| 146
| 741
| 995
|
text-embedding-3-small
|
bge-reranker-base
|
DOC0368|DOC0144|DOC0118
|
C02918|C01139|C02928|C00863
|
borderline
|
human_eval
|
very_negative
|
2025-03-18T22:39:00Z
|
gpt-4o-mini
| 0.38
| 0.95
| 260
|
eos
| 694
| 0.000246
|
SC046
| 1
| 0
| 0
| 0
| 0
|
QA000033
|
run_32
|
financial_reports
|
temporal_reasoning
|
easy
|
How did operating margin change year over year?
|
Operating margin improved by approximately two percentage points.
| 90
| 1
|
correct
|
faithful
| false
|
hybrid
|
sliding_window_256_overlap
| 22
| 0.945405
| 0.46164
| 0.625
| 1
| 1
| true
| 16,384
| 181
| 546
| 772
|
text-embedding-3-small
|
colbert-v2
|
DOC0253|DOC0614|DOC0269|DOC0597
|
C01962|C04888|C02048|C04730
|
pass
|
offline_batch
|
positive
|
2025-03-22T20:01:00Z
|
gpt-4o
| 0.73
| 0.74
| 211
|
eos
| 205
| 0.001541
|
SC033
| 1
| 0
| 1
| 1
| 0
|
QA000034
|
run_33
|
financial_reports
|
instruction_following
|
hard
|
How did operating margin change year over year?
|
Operating margin improved by approximately two percentage points.
| 204
| 1
|
correct
|
faithful
| false
|
bm25
|
fixed_512_tokens
| 38
| 0.879493
| 0.33563
| 0.5
| 0.9
| 1
| true
| 16,384
| 264
| 870
| 1,226
|
gte-large
|
gte-qwen-reranker
|
DOC0218|DOC0381|DOC0343|DOC0065
|
C01687|C03057|C02719|C00461
|
borderline
|
offline_batch
|
neutral
|
2025-03-10T07:52:00Z
|
llama-3.1-8b-instruct
| 0.35
| 0.88
| 313
|
eos
| 415
| 0.000089
|
SC008
| 1
| 0
| 1
| 1
| 0
|
QA000035
|
run_34
|
developer_docs
|
factoid
|
easy
|
How do I authenticate using API keys?
|
API keys are passed as a bearer token in the Authorization header.
| 43
| 1
|
correct
|
faithful
| false
|
bm25
|
sliding_window_256_overlap
| 42
| 0.803723
| 0.256664
| 0.571429
| 1
| 1
| true
| 8,192
| 238
| 238
| 583
|
gte-large
|
cross-encoder-ms-marco
|
DOC0347|DOC0331|DOC0427|DOC0437
|
C02754|C02560|C03412|C03494
|
pass
|
online_shadow
|
very_positive
|
2025-02-05T14:00:00Z
|
llama-3.1-70b-instruct
| 0.45
| 0.91
| 191
|
eos
| 128
| 0.0001
|
SC031
| 1
| 0
| 1
| 1
| 0
|
QA000036
|
run_35
|
developer_docs
|
summarization
|
hard
|
How do I authenticate using API keys?
|
API keys are passed as a bearer token in the Authorization header.
| 159
| 0
|
partial
|
unfaithful
| true
|
dense
|
fixed_512_tokens
| 5
| 0.719979
| 0.575153
| 1
| 1
| 1
| true
| 8,192
| 99
| 565
| 743
|
all-minilm-l12-v2
|
cross-encoder-ms-marco
|
DOC0572|DOC0410|DOC0583|DOC0600
|
C04519|C03284|C04613|C04785
|
fail
|
human_eval
|
very_negative
|
2025-03-05T21:47:00Z
|
gpt-4o-mini
| 0.72
| 0.78
| 227
|
eos
| 334
| 0.00016
|
SC006
| 1
| 0
| 1
| 1
| 0
|
QA000037
|
run_36
|
product_docs
|
temporal_reasoning
|
easy
|
What are the limits for the enterprise tier?
|
Enterprise tier includes custom limits negotiated per contract.
| 130
| 1
|
correct
|
faithful
| false
|
dense
|
semantic
| 14
| 0.6772
| 0.567859
| 0.555556
| 1
| 1
| false
| 4,096
| 138
| 959
| 1,196
|
all-minilm-l12-v2
|
bge-reranker-base
|
DOC0367|DOC0423|DOC0004|DOC0026
|
C02916|C03382|C00027|C00194
|
pass
|
human_eval
|
very_positive
|
2025-03-20T08:29:00Z
|
gpt-4o
| 0.53
| 0.89
| 304
|
eos
| 269
| 0.002125
|
SC024
| 1
| 0
| 1
| 1
| 0
|
QA000038
|
run_37
|
financial_reports
|
temporal_reasoning
|
hard
|
Which segment contributed most to revenue growth?
|
The enterprise subscription segment contributed the largest share of growth.
| 257
| 0
|
incorrect
|
unfaithful
| true
|
hybrid
|
sliding_window_256_overlap
| 40
| 0.128407
| 0.275927
| 0.666667
| 1
| 0.5
| true
| 16,384
| 199
| 1,485
| 1,733
|
all-minilm-l12-v2
|
jina-reranker-v1
|
DOC0384|DOC0597|DOC0217|DOC0587
|
C03088|C04732|C01675|C04652
|
borderline
|
online_shadow
|
negative
|
2025-02-12T19:58:00Z
|
claude-3.5-sonnet
| 0.25
| 0.84
| 387
|
eos
| 650
| 0.005455
|
SC036
| 1
| 0
| 1
| 1
| 0
|
QA000039
|
run_38
|
financial_reports
|
comparison
|
easy
|
Which segment contributed most to revenue growth?
|
The enterprise subscription segment contributed the largest share of growth.
| 84
| 1
|
correct
|
faithful
| false
|
hybrid
|
fixed_512_tokens
| 27
| 0.78845
| 0.423484
| 0.555556
| 1
| 1
| false
| 4,096
| 144
| 491
| 753
|
all-minilm-l12-v2
|
colbert-v2
|
DOC0190|DOC0272|DOC0590|DOC0251
|
C01460|C02063|C04671|C01947
|
pass
|
human_eval
|
positive
|
2025-02-26T17:14:00Z
|
mistral-large
| 0.24
| 0.84
| 169
|
eos
| 336
| 0.000272
|
SC033
| 1
| 0
| 1
| 1
| 0
|
QA000040
|
run_39
|
developer_docs
|
temporal_reasoning
|
easy
|
How do I authenticate using API keys?
|
API keys are passed as a bearer token in the Authorization header.
| 39
| 1
|
correct
|
faithful
| false
|
hybrid
|
by_heading
| 37
| 0.805348
| 0.300722
| 0
| 0
| 0
| false
| 4,096
| 170
| 143
| 358
|
e5-mistral-7b
|
jina-reranker-v1
|
DOC0406|DOC0419|DOC0640|DOC0526
|
C03237|C03346|C05115|C04208
|
pass
|
human_eval
|
positive
|
2025-03-14T02:32:00Z
|
llama-3.1-8b-instruct
| 0.34
| 0.87
| 131
|
eos
| 141
| 0.000024
|
SC032
| 1
| 0
| 0
| 0
| 1
|
QA000041
|
run_40
|
research_papers
|
factoid
|
medium
|
What baseline models were used in the study?
|
The study compares transformer-based baselines with classical models.
| 170
| 1
|
correct
|
faithful
| false
|
hybrid
|
semantic
| 43
| 0.24514
| 0.287121
| 0.285714
| 1
| 0.333333
| false
| 16,384
| 289
| 995
| 1,375
|
bge-m3
|
cross-encoder-ms-marco
|
DOC0078|DOC0369|DOC0487|DOC0284
|
C00566|C02941|C03864|C02203
|
pass
|
offline_batch
|
very_positive
|
2025-03-04T22:24:00Z
|
mistral-large
| 0.34
| 0.87
| 316
|
eos
| 463
| 0.000455
|
SC062
| 1
| 0
| 1
| 1
| 0
|
QA000042
|
run_41
|
developer_docs
|
summarization
|
easy
|
How do I authenticate using API keys?
|
API keys are passed as a bearer token in the Authorization header.
| 159
| 1
|
correct
|
faithful
| false
|
dense_then_rerank
|
fixed_512_tokens
| 23
| 0.612878
| 0.44999
| 0
| 0
| 0
| true
| 8,192
| 122
| 896
| 1,138
|
e5-mistral-7b
|
gte-qwen-reranker
|
DOC0410|DOC0526|DOC0572|DOC0530
|
C03293|C04206|C04518|C04227
|
pass
|
human_eval
|
very_positive
|
2025-03-24T20:09:00Z
|
claude-3.5-sonnet
| 0.21
| 0.91
| 233
|
eos
| 438
| 0.004034
|
SC031
| 1
| 0
| 0
| 0
| 1
|
QA000043
|
run_42
|
product_docs
|
multi_hop
|
hard
|
How does the billing cycle work for annual plans?
|
Annual plans are billed once per year with proration on upgrades.
| 199
| 1
|
correct
|
faithful
| false
|
dense
|
fixed_512_tokens
| 15
| 0.427197
| 0.480014
| 0.428571
| 1
| 0.5
| false
| 4,096
| 125
| 706
| 946
|
text-embedding-3-large
|
gte-qwen-reranker
|
DOC0339|DOC0242|DOC0212|DOC0349
|
C02651|C01897|C01626|C02767
|
borderline
|
human_eval
|
positive
|
2025-03-02T13:12:00Z
|
llama-3.1-8b-instruct
| 0.77
| 0.84
| 289
|
eos
| 768
| 0.000118
|
SC021
| 1
| 0
| 1
| 1
| 0
|
QA000044
|
run_43
|
support_faq
|
instruction_following
|
easy
|
Where can I check the status of my support ticket?
|
Ticket status is available on the support portal under 'My Requests'.
| 227
| 1
|
correct
|
faithful
| false
|
dense_then_rerank
|
by_heading
| 33
| 0.769343
| 0.362448
| 0.555556
| 1
| 1
| true
| 8,192
| 194
| 621
| 841
|
bge-m3
|
bge-reranker-base
|
DOC0501|DOC0472|DOC0232|DOC0475
|
C03958|C03711|C01807|C03742
|
pass
|
human_eval
|
no_feedback
|
2025-02-08T15:16:00Z
|
gpt-4o-mini
| 0.33
| 0.92
| 416
|
eos
| 755
| 0.000242
|
SC013
| 1
| 0
| 1
| 1
| 0
|
QA000045
|
run_44
|
hr_policies
|
instruction_following
|
medium
|
What is the standard probation period for new hires?
|
The typical probation period is three to six months depending on role.
| 117
| 0
|
incorrect
|
unknown
| false
|
dense
|
fixed_512_tokens
| 26
| 0.19405
| 0.354776
| 0
| 1
| 0.142857
| false
| 4,096
| 127
| 503
| 694
|
e5-mistral-7b
|
gte-qwen-reranker
|
DOC0543|DOC0033|DOC0304|DOC0404
|
C04332|C00241|C02375|C03217
|
borderline
|
offline_batch
|
negative
|
2025-02-23T02:40:00Z
|
llama-3.1-8b-instruct
| 0.74
| 0.85
| 291
|
eos
| 430
| 0.000062
|
SC043
| 1
| 0
| 0
| 1
| 0
|
QA000046
|
run_45
|
developer_docs
|
explanation
|
hard
|
How do I authenticate using API keys?
|
API keys are passed as a bearer token in the Authorization header.
| 233
| 1
|
correct
|
faithful
| false
|
bm25
|
sliding_window_256_overlap
| 39
| 0.673199
| 0.274938
| 0.571429
| 1
| 1
| false
| 4,096
| 234
| 1,825
| 2,181
|
text-embedding-3-small
|
jina-reranker-v1
|
DOC0457|DOC0322|DOC0491|DOC0437
|
C03628|C02483|C03890|C03494
|
fail
|
online_shadow
|
neutral
|
2025-02-14T23:12:00Z
|
claude-3.5-sonnet
| 0.88
| 0.9
| 328
|
eos
| 869
| 0.005619
|
SC028
| 1
| 0
| 1
| 1
| 0
|
QA000047
|
run_46
|
developer_docs
|
factoid
|
hard
|
How do I authenticate using API keys?
|
API keys are passed as a bearer token in the Authorization header.
| 231
| 1
|
correct
|
faithful
| false
|
dense
|
by_heading
| 43
| 0.225479
| 0.244806
| 0.428571
| 1
| 0.5
| true
| 32,768
| 191
| 1,458
| 1,771
|
text-embedding-3-large
|
bge-reranker-base
|
DOC0331|DOC0168|DOC0159|DOC0605
|
C02560|C01299|C01238|C04813
|
borderline
|
human_eval
|
positive
|
2025-02-01T05:02:00Z
|
llama-3.1-70b-instruct
| 0.37
| 0.78
| 359
|
eos
| 540
| 0.00053
|
SC006
| 1
| 0
| 1
| 1
| 0
|
QA000048
|
run_47
|
financial_reports
|
factoid
|
medium
|
Which segment contributed most to revenue growth?
|
The enterprise subscription segment contributed the largest share of growth.
| 176
| 1
|
correct
|
faithful
| false
|
dense_then_rerank
|
semantic
| 40
| 0.832401
| 0.30927
| 0.5
| 1
| 1
| false
| 8,192
| 180
| 1,016
| 1,292
|
bge-m3
|
colbert-v2
|
DOC0134|DOC0465|DOC0142|DOC0625
|
C01067|C03677|C01118|C04989
|
pass
|
offline_batch
|
neutral
|
2025-02-05T04:34:00Z
|
gpt-4o
| 0.88
| 0.86
| 283
|
eos
| 358
| 0.002563
|
SC035
| 1
| 0
| 1
| 1
| 0
|
QA000049
|
run_48
|
support_faq
|
instruction_following
|
medium
|
What should I do if the app keeps crashing?
|
Collect logs and contact support with device details and app version.
| 144
| 1
|
correct
|
faithful
| false
|
dense_then_rerank
|
by_heading
| 32
| 0.8002
| 0.32736
| 0
| 0
| 0
| false
| 4,096
| 173
| 981
| 1,187
|
e5-mistral-7b
|
cross-encoder-ms-marco
|
DOC0129|DOC0274|DOC0455|DOC0122
|
C00996|C02074|C03616|C00938
|
pass
|
human_eval
|
positive
|
2025-03-18T14:18:00Z
|
claude-3.5-sonnet
| 0.25
| 0.84
| 220
|
eos
| 365
| 0.003195
|
SC001
| 1
| 0
| 0
| 0
| 1
|
QA000050
|
run_49
|
developer_docs
|
comparison
|
medium
|
How do I authenticate using API keys?
|
API keys are passed as a bearer token in the Authorization header.
| 85
| 0
|
partial
|
unfaithful
| true
|
bm25_then_rerank
|
sliding_window_256_overlap
| 23
| 0.902492
| 0.429996
| 0
| 1
| 0.166667
| false
| 4,096
| 116
| 509
| 694
|
e5-mistral-7b
|
gte-qwen-reranker
|
DOC0600|DOC0223|DOC0210|DOC0395
|
C04776|C01739|C01612|C03167
|
borderline
|
offline_batch
|
no_feedback
|
2025-03-03T04:42:00Z
|
llama-3.1-70b-instruct
| 0.62
| 0.79
| 175
|
eos
| 271
| 0.000251
|
SC026
| 1
| 0
| 0
| 1
| 0
|
QA000051
|
run_50
|
developer_docs
|
multi_hop
|
medium
|
What is the rate limit for write requests?
|
Write requests are limited per minute and per project ID.
| 79
| 1
|
correct
|
faithful
| false
|
hybrid
|
by_heading
| 13
| 0.341705
| 0.594223
| 0.444444
| 1
| 0.5
| false
| 8,192
| 114
| 224
| 473
|
all-minilm-l12-v2
|
gte-qwen-reranker
|
DOC0427|DOC0605|DOC0167|DOC0221
|
C03425|C04815|C01295|C01716
|
pass
|
online_shadow
|
positive
|
2025-03-08T01:49:00Z
|
gpt-4o-mini
| 0.89
| 0.72
| 142
|
eos
| 302
| 0.000087
|
SC027
| 1
| 0
| 1
| 1
| 0
|
QA000052
|
run_51
|
developer_docs
|
instruction_following
|
easy
|
What is the rate limit for write requests?
|
Write requests are limited per minute and per project ID.
| 179
| 1
|
correct
|
faithful
| false
|
dense_then_rerank
|
fixed_512_tokens
| 43
| 0.722239
| 0.267463
| 0.625
| 1
| 1
| true
| 8,192
| 172
| 1,213
| 1,418
|
gte-large
|
colbert-v2
|
DOC0457|DOC0013|DOC0382|DOC0515
|
C03629|C00100|C03078|C04135
|
pass
|
online_shadow
|
positive
|
2025-03-05T07:22:00Z
|
claude-3.5-sonnet
| 0.75
| 0.85
| 347
|
eos
| 468
| 0.003985
|
SC032
| 1
| 0
| 1
| 1
| 0
|
QA000053
|
run_52
|
hr_policies
|
summarization
|
medium
|
What is the standard probation period for new hires?
|
The typical probation period is three to six months depending on role.
| 239
| 1
|
correct
|
faithful
| false
|
bm25
|
by_heading
| 43
| 0.82738
| 0.277702
| 0.5
| 1
| 1
| false
| 4,096
| 212
| 1,424
| 1,752
|
text-embedding-3-large
|
cross-encoder-ms-marco
|
DOC0644|DOC0038|DOC0311|DOC0404
|
C05146|C00291|C02418|C03218
|
pass
|
online_shadow
|
neutral
|
2025-02-10T12:01:00Z
|
llama-3.1-70b-instruct
| 0.28
| 0.7
| 322
|
error
| 884
| 0.000781
|
SC046
| 1
| 0
| 1
| 1
| 0
|
QA000054
|
run_53
|
product_docs
|
comparison
|
medium
|
What are the limits for the enterprise tier?
|
Enterprise tier includes custom limits negotiated per contract.
| 171
| 0
|
incorrect
|
unfaithful
| true
|
hybrid
|
sliding_window_256_overlap
| 38
| 0.941251
| 0.27953
| 0.571429
| 1
| 1
| false
| 4,096
| 258
| 873
| 1,201
|
all-minilm-l12-v2
|
cross-encoder-ms-marco
|
DOC0494|DOC0370|DOC0091|DOC0598
|
C03904|C02962|C00670|C04766
|
borderline
|
offline_batch
|
neutral
|
2025-03-01T15:07:00Z
|
mistral-large
| 0.31
| 0.82
| 247
|
eos
| 568
| 0.000545
|
SC019
| 1
| 0
| 1
| 1
| 0
|
QA000055
|
run_54
|
financial_reports
|
factoid
|
hard
|
Which segment contributed most to revenue growth?
|
The enterprise subscription segment contributed the largest share of growth.
| 114
| 1
|
correct
|
faithful
| false
|
bm25
|
fixed_512_tokens
| 25
| 0.404856
| 0.335071
| 0
| 0
| 0
| false
| 32,768
| 188
| 840
| 1,137
|
all-minilm-l12-v2
|
jina-reranker-v1
|
DOC0047|DOC0243|DOC0251|DOC0225
|
C00357|C01899|C01953|C01758
|
borderline
|
offline_batch
|
very_positive
|
2025-03-21T00:54:00Z
|
claude-3.5-sonnet
| 0.31
| 0.81
| 253
|
eos
| 377
| 0.003013
|
SC008
| 1
| 0
| 0
| 0
| 1
|
QA000056
|
run_55
|
support_faq
|
temporal_reasoning
|
medium
|
Where can I check the status of my support ticket?
|
Ticket status is available on the support portal under 'My Requests'.
| 101
| 1
|
correct
|
faithful
| false
|
dense_then_rerank
|
semantic
| 37
| 0.809483
| 0.298056
| 0.5
| 1
| 1
| false
| 4,096
| 229
| 625
| 881
|
text-embedding-3-small
|
colbert-v2
|
DOC0129|DOC0567|DOC0054|DOC0298
|
C01002|C04490|C00391|C02323
|
pass
|
human_eval
|
neutral
|
2025-03-31T22:54:00Z
|
gpt-4o
| 0.85
| 0.89
| 167
|
eos
| 347
| 0.001834
|
SC014
| 1
| 0
| 1
| 1
| 0
|
QA000057
|
run_56
|
research_papers
|
summarization
|
medium
|
How large is the evaluation dataset?
|
The evaluation dataset contains several tens of thousands of examples.
| 243
| 0
|
partial
|
unknown
| false
|
bm25
|
by_heading
| 17
| 0.819838
| 0.522649
| 0.625
| 1
| 1
| false
| 8,192
| 164
| 1,410
| 1,595
|
all-minilm-l12-v2
|
gte-qwen-reranker
|
DOC0108|DOC0035|DOC0542|DOC0369
|
C00785|C00267|C04326|C02945
|
borderline
|
online_shadow
|
very_negative
|
2025-02-28T18:46:00Z
|
mistral-large
| 0.3
| 1
| 428
|
eos
| 735
| 0.000632
|
SC060
| 1
| 0
| 1
| 1
| 0
|
QA000058
|
run_57
|
hr_policies
|
temporal_reasoning
|
easy
|
What is the standard probation period for new hires?
|
The typical probation period is three to six months depending on role.
| 136
| 1
|
correct
|
unfaithful
| true
|
hybrid
|
sliding_window_256_overlap
| 28
| 0.673546
| 0.365516
| 0.571429
| 1
| 1
| false
| 16,384
| 169
| 462
| 687
|
e5-mistral-7b
|
jina-reranker-v1
|
DOC0560|DOC0118|DOC0543|DOC0463
|
C04452|C00866|C04334|C03669
|
pass
|
offline_batch
|
very_negative
|
2025-02-15T04:55:00Z
|
gpt-4o-mini
| 0.51
| 0.96
| 310
|
eos
| 455
| 0.000159
|
SC011
| 1
| 0
| 1
| 1
| 0
|
QA000059
|
run_58
|
medical_guides
|
multi_hop
|
hard
|
What is the recommended dosage for adult patients?
|
Dosage is weight-adjusted and must follow local clinical guidelines.
| 141
| 0
|
incorrect
|
unknown
| false
|
bm25
|
by_heading
| 30
| 0.285484
| 0.310386
| 0.6
| 1
| 0.333333
| true
| 32,768
| 229
| 786
| 1,139
|
text-embedding-3-large
|
gte-qwen-reranker
|
DOC0090|DOC0208|DOC0302|DOC0315
|
C00668|C01594|C02355|C02452
|
fail
|
online_shadow
|
negative
|
2025-02-07T01:46:00Z
|
mistral-large
| 0.38
| 0.99
| 248
|
eos
| 506
| 0.000457
|
SC039
| 1
| 0
| 1
| 1
| 0
|
QA000060
|
run_59
|
hr_policies
|
explanation
|
easy
|
What is the standard probation period for new hires?
|
The typical probation period is three to six months depending on role.
| 83
| 1
|
correct
|
faithful
| false
|
dense
|
by_heading
| 9
| 0.840726
| 0.719537
| 0.625
| 1
| 1
| true
| 32,768
| 92
| 466
| 660
|
all-minilm-l12-v2
|
cross-encoder-ms-marco
|
DOC0020|DOC0018|DOC0383|DOC0309
|
C00154|C00136|C03082|C02411
|
pass
|
online_shadow
|
no_feedback
|
2025-03-16T18:17:00Z
|
mistral-large
| 0.39
| 0.99
| 157
|
eos
| 264
| 0.000249
|
SC044
| 1
| 0
| 1
| 1
| 0
|
QA000061
|
run_60
|
financial_reports
|
temporal_reasoning
|
easy
|
How did operating margin change year over year?
|
Operating margin improved by approximately two percentage points.
| 168
| 1
|
correct
|
faithful
| false
|
dense
|
semantic
| 9
| 0.719547
| 0.72377
| 0.555556
| 1
| 1
| true
| 32,768
| 97
| 849
| 1,039
|
bge-m3
|
cross-encoder-ms-marco
|
DOC0401|DOC0597|DOC0025|DOC0364
|
C03205|C04737|C00187|C02890
|
pass
|
offline_batch
|
positive
|
2025-03-08T20:59:00Z
|
mistral-large
| 0.43
| 0.72
| 236
|
eos
| 349
| 0.000395
|
SC033
| 1
| 0
| 1
| 1
| 0
|
QA000062
|
run_61
|
medical_guides
|
explanation
|
hard
|
What is the recommended dosage for adult patients?
|
Dosage is weight-adjusted and must follow local clinical guidelines.
| 251
| 0
|
partial
|
unfaithful
| true
|
dense
|
fixed_512_tokens
| 42
| 0.819858
| 0.301758
| 0.428571
| 1
| 1
| false
| 32,768
| 229
| 1,482
| 1,851
|
gte-large
|
gte-qwen-reranker
|
DOC0046|DOC0178|DOC0315|DOC0140
|
C00339|C01387|C02450|C01103
|
fail
|
offline_batch
|
no_feedback
|
2025-02-27T04:35:00Z
|
mistral-large
| 0.38
| 0.78
| 422
|
eos
| 758
| 0.000726
|
SC010
| 1
| 0
| 1
| 1
| 0
|
QA000063
|
run_62
|
product_docs
|
temporal_reasoning
|
medium
|
Can I downgrade from enterprise to pro mid-cycle?
|
Downgrades are applied at the end of the current billing period.
| 168
| 0
|
incorrect
|
unknown
| false
|
dense_then_rerank
|
semantic
| 8
| 0.857916
| 0.820229
| 0.625
| 1
| 1
| true
| 16,384
| 107
| 795
| 964
|
all-minilm-l12-v2
|
bge-reranker-base
|
DOC0351|DOC0405|DOC0036|DOC0550
|
C02781|C03231|C00277|C04365
|
borderline
|
online_shadow
|
neutral
|
2025-03-03T05:35:00Z
|
mistral-large
| 0.58
| 0.79
| 279
|
eos
| 537
| 0.000503
|
SC022
| 1
| 0
| 1
| 1
| 0
|
QA000064
|
run_63
|
financial_reports
|
comparison
|
hard
|
How did operating margin change year over year?
|
Operating margin improved by approximately two percentage points.
| 188
| 0
|
incorrect
|
unfaithful
| true
|
dense_then_rerank
|
fixed_512_tokens
| 30
| 0.472154
| 0.33776
| 0
| 0.8
| 0.166667
| false
| 4,096
| 235
| 546
| 903
|
e5-mistral-7b
|
jina-reranker-v1
|
DOC0381|DOC0219|DOC0010|DOC0217
|
C03058|C01697|C00072|C01661
|
borderline
|
human_eval
|
negative
|
2025-02-02T09:38:00Z
|
gpt-4o-mini
| 0.44
| 0.72
| 319
|
eos
| 705
| 0.000235
|
SC036
| 1
| 0
| 0
| 1
| 0
|
QA000065
|
run_64
|
hr_policies
|
summarization
|
easy
|
How many days of annual leave are employees entitled to?
|
Employees receive a standard allocation of annual leave as defined in their contract.
| 174
| 1
|
correct
|
faithful
| false
|
bm25_then_rerank
|
semantic
| 7
| 0.929169
| 0.791163
| 0.714286
| 1
| 1
| false
| 8,192
| 108
| 588
| 729
|
bge-m3
|
colbert-v2
|
DOC0577|DOC0469|DOC0309|DOC0644
|
C04573|C03695|C02408|C05141
|
pass
|
offline_batch
|
very_positive
|
2025-02-11T21:34:00Z
|
gpt-4o-mini
| 0.45
| 0.95
| 239
|
eos
| 350
| 0.000146
|
SC011
| 1
| 0
| 1
| 1
| 0
|
QA000066
|
run_65
|
developer_docs
|
temporal_reasoning
|
hard
|
What is the rate limit for write requests?
|
Write requests are limited per minute and per project ID.
| 24
| 1
|
correct
|
faithful
| false
|
dense_then_rerank
|
semantic
| 8
| 0.235051
| 0.624551
| 0.6
| 1
| 0.333333
| true
| 16,384
| 73
| 193
| 306
|
text-embedding-3-small
|
colbert-v2
|
DOC0616|DOC0347|DOC0504|DOC0511
|
C04914|C02753|C03994|C04089
|
borderline
|
human_eval
|
neutral
|
2025-02-16T03:11:00Z
|
gpt-4o
| 0.5
| 0.97
| 190
|
eos
| 128
| 0.000534
|
SC028
| 1
| 0
| 1
| 1
| 0
|
QA000067
|
run_66
|
support_faq
|
explanation
|
easy
|
How can I reset my account password?
|
You can reset your password from the sign-in page using the forgot password link.
| 204
| 1
|
correct
|
faithful
| false
|
bm25
|
sliding_window_256_overlap
| 42
| 0.719401
| 0.283582
| 0
| 0
| 0
| true
| 16,384
| 337
| 999
| 1,413
|
gte-large
|
gte-qwen-reranker
|
DOC0307|DOC0196|DOC0615|DOC0519
|
C02395|C01526|C04892|C04165
|
pass
|
human_eval
|
very_positive
|
2025-02-26T12:30:00Z
|
mistral-large
| 0.38
| 0.88
| 326
|
eos
| 558
| 0.000539
|
SC015
| 1
| 0
| 0
| 0
| 1
|
QA000068
|
run_67
|
research_papers
|
comparison
|
easy
|
What baseline models were used in the study?
|
The study compares transformer-based baselines with classical models.
| 37
| 1
|
correct
|
faithful
| false
|
dense
|
semantic
| 11
| 0.844666
| 0.648045
| 0.571429
| 1
| 1
| true
| 8,192
| 132
| 169
| 436
|
gte-large
|
cross-encoder-ms-marco
|
DOC0137|DOC0120|DOC0284
|
C01084|C00911|C00910|C02172
|
pass
|
offline_batch
|
positive
|
2025-02-01T10:15:00Z
|
llama-3.1-8b-instruct
| 0.7
| 0.97
| 96
|
eos
| 133
| 0.000022
|
SC059
| 1
| 0
| 1
| 1
| 0
|
QA000069
|
run_68
|
developer_docs
|
table_qa
|
medium
|
What is the rate limit for write requests?
|
Write requests are limited per minute and per project ID.
| 105
| 1
|
correct
|
faithful
| false
|
dense
|
sliding_window_256_overlap
| 18
| 0.439213
| 0.525985
| 0.375
| 1
| 0.5
| false
| 4,096
| 153
| 442
| 699
|
text-embedding-3-small
|
cross-encoder-ms-marco
|
DOC0159|DOC0410|DOC0530|DOC0492
|
C01241|C03298|C04227|C03892
|
pass
|
offline_batch
|
positive
|
2025-03-17T10:12:00Z
|
llama-3.1-8b-instruct
| 0.5
| 0.91
| 163
|
eos
| 312
| 0.000056
|
SC029
| 1
| 0
| 1
| 1
| 0
|
QA000070
|
run_69
|
medical_guides
|
explanation
|
hard
|
What is the recommended dosage for adult patients?
|
Dosage is weight-adjusted and must follow local clinical guidelines.
| 106
| 0
|
partial
|
unknown
| false
|
dense_then_rerank
|
semantic
| 35
| 1
| 0.382046
| 0
| 1
| 0.166667
| false
| 16,384
| 280
| 304
| 618
|
text-embedding-3-small
|
bge-reranker-base
|
DOC0098|DOC0386|DOC0485|DOC0531
|
C00715|C03106|C03819|C04237
|
borderline
|
offline_batch
|
negative
|
2025-02-04T10:28:00Z
|
gpt-4o-mini
| 0.51
| 0.83
| 292
|
eos
| 277
| 0.0001
|
SC038
| 1
| 0
| 0
| 1
| 0
|
QA000071
|
run_70
|
product_docs
|
summarization
|
medium
|
Can I downgrade from enterprise to pro mid-cycle?
|
Downgrades are applied at the end of the current billing period.
| 125
| 0
|
incorrect
|
unknown
| false
|
hybrid
|
by_heading
| 35
| 0.79133
| 0.280967
| 0
| 1
| 0.166667
| true
| 32,768
| 176
| 581
| 892
|
text-embedding-3-small
|
colbert-v2
|
DOC0481|DOC0578|DOC0351
|
C03778|C04587|C02796|C03780
|
borderline
|
offline_batch
|
neutral
|
2025-03-31T00:52:00Z
|
llama-3.1-8b-instruct
| 0.27
| 0.74
| 181
|
eos
| 423
| 0.000067
|
SC023
| 1
| 0
| 0
| 1
| 0
|
QA000072
|
run_71
|
policies
|
factoid
|
medium
|
What is the data retention policy for customer logs?
|
Customer logs are retained for 180 days by default.
| 167
| 1
|
correct
|
faithful
| false
|
dense
|
fixed_512_tokens
| 12
| 0.120821
| 0.557525
| 0.333333
| 1
| 0.333333
| true
| 8,192
| 127
| 502
| 741
|
all-minilm-l12-v2
|
gte-qwen-reranker
|
DOC0439|DOC0510|DOC0150|DOC0156
|
C03508|C04075|C01164|C01214
|
pass
|
human_eval
|
positive
|
2025-03-09T00:30:00Z
|
gpt-4o-mini
| 0.12
| 0.99
| 258
|
eos
| 502
| 0.000179
|
SC050
| 1
| 0
| 1
| 1
| 0
|
QA000073
|
run_72
|
developer_docs
|
multi_hop
|
medium
|
How do I authenticate using API keys?
|
API keys are passed as a bearer token in the Authorization header.
| 43
| 1
|
correct
|
faithful
| false
|
dense
|
semantic
| 17
| 0.441462
| 0.513555
| 0.375
| 0.875
| 0.333333
| true
| 16,384
| 149
| 343
| 600
|
bge-m3
|
gte-qwen-reranker
|
DOC0492|DOC0114|DOC0406|DOC0331
|
C03892|C00816|C03242|C02572
|
pass
|
online_shadow
|
positive
|
2025-02-13T06:22:00Z
|
claude-3.5-sonnet
| 0.75
| 0.91
| 96
|
eos
| 135
| 0.00098
|
SC030
| 1
| 0
| 1
| 1
| 0
|
QA000074
|
run_73
|
support_faq
|
factoid
|
hard
|
Where can I check the status of my support ticket?
|
Ticket status is available on the support portal under 'My Requests'.
| 69
| 1
|
correct
|
unfaithful
| true
|
dense_then_rerank
|
semantic
| 5
| 0.123642
| 0.551853
| 1
| 1
| 0.5
| true
| 8,192
| 90
| 388
| 564
|
e5-mistral-7b
|
jina-reranker-v1
|
DOC0462|DOC0455|DOC0324|DOC0054
|
C03663|C03609|C02498|C00396
|
borderline
|
offline_batch
|
very_negative
|
2025-03-28T00:33:00Z
|
mistral-large
| 0.78
| 0.81
| 125
|
eos
| 197
| 0.0002
|
SC001
| 1
| 0
| 1
| 1
| 0
|
QA000075
|
run_74
|
medical_guides
|
factoid
|
medium
|
What is the recommended dosage for adult patients?
|
Dosage is weight-adjusted and must follow local clinical guidelines.
| 141
| 1
|
correct
|
faithful
| false
|
bm25
|
sliding_window_256_overlap
| 38
| 0.858417
| 0.365384
| 0.5
| 1
| 1
| true
| 8,192
| 252
| 473
| 792
|
e5-mistral-7b
|
gte-qwen-reranker
|
DOC0008|DOC0325|DOC0023|DOC0438
|
C00056|C02512|C00175|C03497
|
pass
|
offline_batch
|
neutral
|
2025-02-28T00:52:00Z
|
llama-3.1-8b-instruct
| 0.29
| 0.74
| 306
|
eos
| 486
| 0.000076
|
SC037
| 1
| 0
| 1
| 1
| 0
|
QA000076
|
run_75
|
developer_docs
|
table_qa
|
hard
|
How do I authenticate using API keys?
|
API keys are passed as a bearer token in the Authorization header.
| 73
| 0
|
partial
|
unfaithful
| true
|
hybrid
|
fixed_512_tokens
| 6
| 0.741348
| 0.631448
| 1
| 1
| 1
| false
| 32,768
| 65
| 419
| 517
|
text-embedding-3-small
|
bge-reranker-base
|
DOC0378|DOC0511|DOC0348|DOC0419
|
C03016|C04089|C02761|C03350
|
fail
|
online_shadow
|
no_feedback
|
2025-03-02T23:30:00Z
|
mistral-large
| 0.39
| 0.94
| 231
|
eos
| 148
| 0.00019
|
SC006
| 1
| 0
| 1
| 1
| 0
|
QA000077
|
run_76
|
financial_reports
|
multi_hop
|
medium
|
How did operating margin change year over year?
|
Operating margin improved by approximately two percentage points.
| 100
| 1
|
correct
|
faithful
| false
|
hybrid
|
fixed_512_tokens
| 24
| 0.908235
| 0.449913
| 0.444444
| 1
| 1
| true
| 8,192
| 172
| 605
| 871
|
bge-m3
|
bge-reranker-base
|
DOC0552|DOC0586|DOC0171
|
C04387|C04628|C01322|C01320
|
pass
|
human_eval
|
very_positive
|
2025-02-21T07:36:00Z
|
gpt-4o
| 0.39
| 0.94
| 195
|
eos
| 200
| 0.001515
|
SC007
| 1
| 0
| 1
| 1
| 0
|
QA000078
|
run_77
|
developer_docs
|
multi_hop
|
medium
|
What is the rate limit for write requests?
|
Write requests are limited per minute and per project ID.
| 97
| 1
|
correct
|
faithful
| false
|
hybrid
|
semantic
| 41
| 0.804606
| 0.356969
| 0
| 0
| 0
| true
| 8,192
| 223
| 764
| 1,022
|
gte-large
|
jina-reranker-v1
|
DOC0167|DOC0041|DOC0049|DOC0450
|
C01295|C00306|C00370|C03589
|
pass
|
online_shadow
|
positive
|
2025-02-20T18:23:00Z
|
gpt-4o
| 0.65
| 0.86
| 170
|
eos
| 326
| 0.001904
|
SC005
| 1
| 0
| 0
| 0
| 1
|
QA000079
|
run_78
|
product_docs
|
factoid
|
medium
|
Can I downgrade from enterprise to pro mid-cycle?
|
Downgrades are applied at the end of the current billing period.
| 205
| 1
|
correct
|
faithful
| false
|
bm25_then_rerank
|
by_heading
| 34
| 0.149795
| 0.327371
| 0.375
| 1
| 0.5
| true
| 32,768
| 221
| 904
| 1,216
|
gte-large
|
cross-encoder-ms-marco
|
DOC0405|DOC0370|DOC0326|DOC0064
|
C03221|C02955|C02539|C00439
|
pass
|
human_eval
|
neutral
|
2025-02-07T21:37:00Z
|
llama-3.1-8b-instruct
| 0.25
| 1
| 288
|
eos
| 421
| 0.000076
|
SC003
| 1
| 0
| 1
| 1
| 0
|
QA000080
|
run_79
|
financial_reports
|
temporal_reasoning
|
medium
|
How did operating margin change year over year?
|
Operating margin improved by approximately two percentage points.
| 22
| 1
|
correct
|
faithful
| false
|
bm25_then_rerank
|
by_heading
| 29
| 0.717319
| 0.393447
| 0.5
| 1
| 1
| true
| 32,768
| 180
| 145
| 410
|
gte-large
|
colbert-v2
|
DOC0100|DOC0253|DOC0070
|
C00724|C01959|C01968|C00515
|
pass
|
human_eval
|
neutral
|
2025-02-26T23:25:00Z
|
llama-3.1-70b-instruct
| 0.88
| 0.82
| 120
|
eos
| 128
| 0.000093
|
SC034
| 1
| 0
| 1
| 1
| 0
|
QA000081
|
run_80
|
financial_reports
|
multi_hop
|
medium
|
Which segment contributed most to revenue growth?
|
The enterprise subscription segment contributed the largest share of growth.
| 48
| 1
|
correct
|
faithful
| false
|
hybrid
|
sliding_window_256_overlap
| 44
| 0.208075
| 0.270796
| 0.375
| 1
| 0.5
| true
| 32,768
| 313
| 393
| 798
|
bge-m3
|
gte-qwen-reranker
|
DOC0568|DOC0253|DOC0552|DOC0622
|
C04494|C01966|C04383|C04947
|
pass
|
offline_batch
|
neutral
|
2025-03-10T17:44:00Z
|
gpt-4o
| 0.15
| 0.98
| 207
|
eos
| 131
| 0.000794
|
SC035
| 1
| 0
| 1
| 1
| 0
|
QA000082
|
run_81
|
medical_guides
|
temporal_reasoning
|
medium
|
Which symptoms indicate the need for immediate escalation?
|
Red-flag symptoms include chest pain, severe shortness of breath, and confusion.
| 142
| 1
|
correct
|
faithful
| false
|
dense_then_rerank
|
semantic
| 31
| 0.497136
| 0.353913
| 0.428571
| 1
| 0.333333
| true
| 16,384
| 189
| 443
| 751
|
gte-large
|
bge-reranker-base
|
DOC0485|DOC0620|DOC0186|DOC0444
|
C03834|C04932|C01441|C03540
|
pass
|
offline_batch
|
positive
|
2025-02-24T19:43:00Z
|
gpt-4o-mini
| 0.88
| 0.89
| 275
|
eos
| 458
| 0.000144
|
SC041
| 1
| 0
| 1
| 1
| 0
|
QA000083
|
run_82
|
product_docs
|
explanation
|
easy
|
Can I downgrade from enterprise to pro mid-cycle?
|
Downgrades are applied at the end of the current billing period.
| 159
| 1
|
correct
|
faithful
| false
|
dense
|
sliding_window_256_overlap
| 25
| 0.733893
| 0.433916
| 0.555556
| 1
| 1
| true
| 8,192
| 175
| 595
| 839
|
e5-mistral-7b
|
colbert-v2
|
DOC0326|DOC0091|DOC0247|DOC0372
|
C02539|C00673|C01920|C02979
|
pass
|
offline_batch
|
very_positive
|
2025-02-14T18:29:00Z
|
llama-3.1-8b-instruct
| 0.16
| 0.92
| 270
|
eos
| 564
| 0.000089
|
SC024
| 1
| 0
| 1
| 1
| 0
|
QA000084
|
run_83
|
policies
|
comparison
|
easy
|
How is user consent stored for analytics tracking?
|
User consent is stored as a hashed record with timestamps.
| 62
| 1
|
correct
|
faithful
| false
|
dense_then_rerank
|
by_heading
| 39
| 0.800721
| 0.343952
| 0.555556
| 1
| 1
| false
| 4,096
| 292
| 450
| 814
|
all-minilm-l12-v2
|
gte-qwen-reranker
|
DOC0156|DOC0341|DOC0104|DOC0172
|
C01220|C02668|C00755|C01342
|
pass
|
online_shadow
|
neutral
|
2025-03-27T22:43:00Z
|
gpt-4o
| 0.35
| 0.98
| 119
|
eos
| 226
| 0.001244
|
SC012
| 1
| 0
| 1
| 1
| 0
|
QA000085
|
run_84
|
product_docs
|
summarization
|
medium
|
Can I downgrade from enterprise to pro mid-cycle?
|
Downgrades are applied at the end of the current billing period.
| 33
| 1
|
correct
|
faithful
| false
|
bm25_then_rerank
|
fixed_512_tokens
| 16
| 0.72166
| 0.508511
| 0.5
| 1
| 1
| true
| 16,384
| 166
| 107
| 300
|
text-embedding-3-large
|
cross-encoder-ms-marco
|
DOC0637|DOC0351|DOC0550|DOC0072
|
C05101|C02779|C04362|C00520
|
pass
|
human_eval
|
positive
|
2025-03-27T08:05:00Z
|
gpt-4o-mini
| 0.3
| 0.95
| 97
|
eos
| 128
| 0.000039
|
SC004
| 1
| 0
| 1
| 1
| 0
|
QA000086
|
run_85
|
developer_docs
|
comparison
|
medium
|
How do I authenticate using API keys?
|
API keys are passed as a bearer token in the Authorization header.
| 213
| 1
|
correct
|
faithful
| false
|
dense_then_rerank
|
fixed_512_tokens
| 31
| 0.830347
| 0.342438
| 0.428571
| 1
| 1
| true
| 16,384
| 191
| 604
| 850
|
bge-m3
|
colbert-v2
|
DOC0382|DOC0347|DOC0616|DOC0605
|
C03076|C02756|C04913|C04817
|
pass
|
online_shadow
|
positive
|
2025-02-18T16:00:00Z
|
gpt-4o-mini
| 0.31
| 0.79
| 290
|
eos
| 495
| 0.000215
|
SC025
| 1
| 0
| 1
| 1
| 0
|
QA000087
|
run_86
|
developer_docs
|
explanation
|
hard
|
What is the rate limit for write requests?
|
Write requests are limited per minute and per project ID.
| 103
| 1
|
correct
|
unfaithful
| true
|
bm25
|
sliding_window_256_overlap
| 39
| 0.732375
| 0.310016
| 0.444444
| 0.888889
| 1
| false
| 32,768
| 245
| 650
| 925
|
bge-m3
|
bge-reranker-base
|
DOC0297|DOC0003|DOC0395|DOC0388
|
C02312|C00022|C03169|C03126
|
borderline
|
human_eval
|
negative
|
2025-03-08T07:53:00Z
|
claude-3.5-sonnet
| 0.9
| 0.96
| 181
|
eos
| 408
| 0.002654
|
SC028
| 1
| 0
| 1
| 1
| 0
|
QA000088
|
run_87
|
research_papers
|
explanation
|
medium
|
What baseline models were used in the study?
|
The study compares transformer-based baselines with classical models.
| 192
| 1
|
correct
|
faithful
| false
|
bm25
|
semantic
| 25
| 0.796926
| 0.45634
| 0
| 0
| 0
| false
| 4,096
| 193
| 574
| 804
|
gte-large
|
gte-qwen-reranker
|
DOC0284|DOC0105|DOC0435|DOC0120
|
C02205|C00765|C03474|C00910
|
pass
|
online_shadow
|
positive
|
2025-03-07T02:21:00Z
|
gpt-4o-mini
| 0.47
| 0.86
| 310
|
eos
| 401
| 0.000168
|
SC062
| 1
| 0
| 0
| 0
| 1
|
QA000089
|
run_88
|
medical_guides
|
factoid
|
medium
|
Which symptoms indicate the need for immediate escalation?
|
Red-flag symptoms include chest pain, severe shortness of breath, and confusion.
| 215
| 1
|
correct
|
faithful
| false
|
hybrid
|
semantic
| 34
| 0.602632
| 0.339147
| 0.444444
| 1
| 1
| false
| 4,096
| 246
| 1,745
| 2,062
|
text-embedding-3-small
|
gte-qwen-reranker
|
DOC0474|DOC0500|DOC0259|DOC0482
|
C03731|C03938|C01992|C03798
|
pass
|
human_eval
|
positive
|
2025-03-18T15:40:00Z
|
gpt-4o
| 0.2
| 0.71
| 387
|
eos
| 780
| 0.003827
|
SC009
| 1
| 0
| 1
| 1
| 0
|
QA000090
|
run_89
|
medical_guides
|
summarization
|
medium
|
Which symptoms indicate the need for immediate escalation?
|
Red-flag symptoms include chest pain, severe shortness of breath, and confusion.
| 203
| 1
|
correct
|
faithful
| false
|
dense
|
fixed_512_tokens
| 44
| 0.942356
| 0.288003
| 0
| 0
| 0
| true
| 8,192
| 288
| 1,495
| 1,852
|
text-embedding-3-small
|
colbert-v2
|
DOC0199|DOC0507|DOC0191|DOC0354
|
C01542|C04037|C01487|C02825
|
pass
|
online_shadow
|
positive
|
2025-03-14T16:22:00Z
|
gpt-4o
| 0.89
| 0.9
| 311
|
eos
| 414
| 0.003239
|
SC037
| 1
| 0
| 0
| 0
| 1
|
QA000091
|
run_90
|
research_papers
|
factoid
|
easy
|
How large is the evaluation dataset?
|
The evaluation dataset contains several tens of thousands of examples.
| 177
| 1
|
correct
|
faithful
| false
|
bm25
|
semantic
| 40
| 0.878118
| 0.336505
| 0.555556
| 1
| 1
| false
| 4,096
| 163
| 1,436
| 1,731
|
all-minilm-l12-v2
|
bge-reranker-base
|
DOC0166|DOC0342|DOC0035|DOC0085
|
C01285|C02683|C00260|C00636
|
pass
|
offline_batch
|
neutral
|
2025-02-02T17:53:00Z
|
gpt-4o
| 0.29
| 0.85
| 272
|
eos
| 526
| 0.002868
|
SC059
| 1
| 0
| 1
| 1
| 0
|
QA000092
|
run_91
|
product_docs
|
explanation
|
medium
|
How does the billing cycle work for annual plans?
|
Annual plans are billed once per year with proration on upgrades.
| 41
| 1
|
correct
|
faithful
| false
|
bm25_then_rerank
|
fixed_512_tokens
| 44
| 0.84185
| 0.275575
| 0.5
| 1
| 1
| true
| 32,768
| 210
| 244
| 488
|
gte-large
|
cross-encoder-ms-marco
|
DOC0066|DOC0447|DOC0539|DOC0344
|
C00473|C03558|C04283|C02732
|
pass
|
human_eval
|
positive
|
2025-02-02T10:33:00Z
|
llama-3.1-70b-instruct
| 0.16
| 0.86
| 169
|
eos
| 135
| 0.000106
|
SC019
| 1
| 0
| 1
| 1
| 0
|
QA000093
|
run_92
|
developer_docs
|
temporal_reasoning
|
medium
|
How do I authenticate using API keys?
|
API keys are passed as a bearer token in the Authorization header.
| 159
| 1
|
correct
|
faithful
| false
|
dense_then_rerank
|
by_heading
| 14
| 0.842897
| 0.580682
| 0.5
| 1
| 1
| true
| 32,768
| 104
| 562
| 704
|
bge-m3
|
gte-qwen-reranker
|
DOC0605|DOC0378|DOC0428
|
C04817|C03017|C03008|C03429
|
pass
|
offline_batch
|
no_feedback
|
2025-02-18T08:42:00Z
|
gpt-4o-mini
| 0.12
| 0.82
| 217
|
eos
| 470
| 0.00015
|
SC026
| 1
| 0
| 1
| 1
| 0
|
QA000094
|
run_93
|
hr_policies
|
factoid
|
medium
|
How many days of annual leave are employees entitled to?
|
Employees receive a standard allocation of annual leave as defined in their contract.
| 165
| 1
|
correct
|
unfaithful
| true
|
hybrid
|
sliding_window_256_overlap
| 29
| 0.793725
| 0.39767
| 0.444444
| 1
| 1
| true
| 8,192
| 180
| 553
| 843
|
text-embedding-3-small
|
jina-reranker-v1
|
DOC0088|DOC0362|DOC0119|DOC0128
|
C00661|C02885|C00883|C00983
|
pass
|
human_eval
|
very_negative
|
2025-02-26T01:25:00Z
|
gpt-4o-mini
| 0.47
| 0.72
| 317
|
eos
| 516
| 0.000172
|
SC043
| 1
| 0
| 1
| 1
| 0
|
QA000095
|
run_94
|
financial_reports
|
explanation
|
medium
|
How did operating margin change year over year?
|
Operating margin improved by approximately two percentage points.
| 70
| 1
|
correct
|
faithful
| false
|
hybrid
|
by_heading
| 36
| 0.839514
| 0.307903
| 0.444444
| 1
| 1
| true
| 32,768
| 241
| 461
| 725
|
gte-large
|
jina-reranker-v1
|
DOC0240|DOC0364|DOC0608|DOC0142
|
C01865|C02893|C04836|C01113
|
pass
|
online_shadow
|
positive
|
2025-03-08T16:14:00Z
|
claude-3.5-sonnet
| 0.83
| 0.75
| 125
|
eos
| 144
| 0.001463
|
SC007
| 1
| 0
| 1
| 1
| 0
|
QA000096
|
run_95
|
policies
|
multi_hop
|
easy
|
How is user consent stored for analytics tracking?
|
User consent is stored as a hashed record with timestamps.
| 165
| 1
|
correct
|
faithful
| false
|
dense
|
sliding_window_256_overlap
| 30
| 0.699103
| 0.352213
| 0
| 0
| 0
| true
| 8,192
| 175
| 808
| 1,081
|
bge-m3
|
jina-reranker-v1
|
DOC0633|DOC0183|DOC0446|DOC0328
|
C05067|C01411|C03556|C02549
|
pass
|
offline_batch
|
positive
|
2025-03-20T09:44:00Z
|
mistral-large
| 0.11
| 0.93
| 263
|
eos
| 381
| 0.000412
|
SC048
| 1
| 0
| 0
| 0
| 1
|
QA000097
|
run_96
|
developer_docs
|
temporal_reasoning
|
medium
|
How do I authenticate using API keys?
|
API keys are passed as a bearer token in the Authorization header.
| 62
| 0
|
incorrect
|
unfaithful
| true
|
dense_then_rerank
|
sliding_window_256_overlap
| 24
| 0.825663
| 0.377376
| 0
| 1
| 0.142857
| true
| 16,384
| 145
| 520
| 775
|
gte-large
|
bge-reranker-base
|
DOC0124|DOC0491|DOC0168|DOC0297
|
C00958|C03888|C01300|C02315
|
borderline
|
offline_batch
|
negative
|
2025-02-01T17:57:00Z
|
gpt-4o
| 0.68
| 0.77
| 247
|
eos
| 193
| 0.001161
|
SC027
| 1
| 0
| 0
| 1
| 0
|
QA000098
|
run_97
|
product_docs
|
explanation
|
hard
|
How does the billing cycle work for annual plans?
|
Annual plans are billed once per year with proration on upgrades.
| 144
| 0
|
partial
|
unfaithful
| true
|
dense_then_rerank
|
by_heading
| 23
| 0.703499
| 0.378932
| 0
| 0.833333
| 0.166667
| false
| 4,096
| 164
| 510
| 769
|
text-embedding-3-small
|
bge-reranker-base
|
DOC0162|DOC0367|DOC0520|DOC0061
|
C01255|C02913|C04171|C00429
|
borderline
|
online_shadow
|
very_negative
|
2025-03-11T13:52:00Z
|
gpt-4o-mini
| 0.31
| 0.89
| 250
|
eos
| 526
| 0.000162
|
SC020
| 1
| 0
| 0
| 1
| 0
|
QA000099
|
run_98
|
financial_reports
|
multi_hop
|
medium
|
Which segment contributed most to revenue growth?
|
The enterprise subscription segment contributed the largest share of growth.
| 168
| 1
|
correct
|
faithful
| false
|
bm25_then_rerank
|
sliding_window_256_overlap
| 22
| 0.650668
| 0.445058
| 0.333333
| 1
| 0.5
| true
| 16,384
| 138
| 810
| 1,003
|
text-embedding-3-large
|
cross-encoder-ms-marco
|
DOC0586|DOC0024|DOC0190|DOC0506
|
C04630|C00180|C01461|C04020
|
pass
|
offline_batch
|
neutral
|
2025-03-17T08:54:00Z
|
mistral-large
| 0.28
| 0.86
| 323
|
eos
| 520
| 0.00043
|
SC034
| 1
| 0
| 1
| 1
| 0
|
QA000100
|
run_99
|
financial_reports
|
comparison
|
hard
|
How did operating margin change year over year?
|
Operating margin improved by approximately two percentage points.
| 204
| 0
|
incorrect
|
unfaithful
| true
|
bm25_then_rerank
|
by_heading
| 18
| 0.810017
| 0.548982
| 0.5
| 1
| 1
| false
| 4,096
| 185
| 1,387
| 1,604
|
gte-large
|
gte-qwen-reranker
|
DOC0595|DOC0133|DOC0272
|
C04725|C01064|C02061|C01055
|
fail
|
human_eval
|
negative
|
2025-03-14T03:14:00Z
|
llama-3.1-70b-instruct
| 0.66
| 0.77
| 291
|
eos
| 612
| 0.000542
|
SC008
| 1
| 0
| 1
| 1
| 0
|
π§ RAG QA Evaluation Logs & Corpus
Multi-Table RAG Telemetry for Quality, Hallucinations, Latency, and Cost
A multi-table dataset modeling a production-style RAG (Retrieval-Augmented Generation) system end-to-end:
- a document corpus
- a chunk-level index
- retrieval events per query
- QA evaluation runs with correctness, hallucination, latency, and cost signals
- scenario templates for QA use cases
- a data dictionary documenting every column
All records are fully synthetic but realistic system logs and corpus data β designed to look and behave like real RAG telemetry while remaining safe to share and experiment with.
The dataset is intended for RAG quality analysis, meta-modeling, hallucination risk scoring, and dashboard-style telemetry.
π Privacy & Synthetic Data
This dataset is fully synthetic.
- No real users, customers, patients, or organisations are represented.
- No personally identifiable information (PII) is included.
- All IDs, queries, documents, and logs were programmatically generated to mimic realistic RAG system behaviour while preserving privacy.
The design aims to balance realism (for meaningful analysis and modeling) with strong privacy guarantees, making it suitable for open research, teaching, demos, and public dashboards.
π Dataset Overview
| Field | Description |
|---|---|
| Files | rag_corpus_documents.csv, rag_corpus_chunks.csv, rag_qa_eval_runs.csv, rag_retrieval_events.csv, rag_qa_scenarios.csv, rag_qa_data_dictionary.csv |
| Tables | 6 (documents, chunks, QA runs, retrieval events, data dictionary) |
| Total rows (approx.) | ~103K across all tables (103,273 rows in total) |
| Main targets | is_correct, hallucination_flag, faithfulness_label |
| Type | Multi-table tabular logs + short text fields |
Tables are linked by stable identifiers such as doc_id, chunk_id, run_id, example_id, and scenario_id, making joins explicit and reliable.
π Files
rag_corpus_documents.csvβ document-level corpus metadatarag_corpus_chunks.csvβ chunk-level index and text contentrag_qa_eval_runs.csvβ QA runs with labels, metrics, and configurationsrag_retrieval_events.csvβ per-chunk retrieval telemetryrag_qa_scenarios.csvβ scenario-level QA templates and use casesrag_qa_data_dictionary.csvβ column-level documentation for all tables
All files are in CSV format, use snake_case column names, and are designed to be ML- and analytics-ready.
π Table Summary
Approximate size per table:
| Table | Rows | Columns | Granularity |
|--------------------------|--------|---------|-------------------------------------------------------|
| rag_corpus_documents | 658 | 19 | One row per document in the RAG corpus |
| rag_corpus_chunks | 5,237 | 6 | One row per text chunk derived from a document |
| rag_qa_eval_runs | 3,824 | 46 | One row per QA evaluation example |
| rag_retrieval_events | 93,375 | 9 | One row per retrieved chunk for a given QA example |
| rag_qa_scenarios | 88 | 11 | One row per scenario-level QA template / use case |
| rag_qa_data_dictionary | 91 | 5 | One row per column definition across all tables |
π§± Table Structure
1οΈβ£ Document Corpus β rag_corpus_documents.csv
High-level view of the RAG knowledge base. Granularity: 1 row = 1 document Key fields:
doc_idβ unique document identifierdomainβ e.g.support_faq,hr_policies,product_docs,developer_docs,policies,financial_reports,medical_guides,research_papers,customer_success,data_platform_docs,mlops_docs,marketing_analyticstitleβ document titlesource_typeβ e.g.pdf_manual,spreadsheet,wiki_pagelanguageβ currentlyenn_sections,n_tokens,n_chunks,avg_chunk_tokensβ structural and size indicatorscreated_at_utc,last_updated_at_utcβ lifecycle timestampsis_active,contains_tablesβ operational flagspii_risk_level,security_tierβ risk and access levelowner_team,embedding_model,search_index,top_keywordsβ ownership and indexing metadata
Use this table to understand what kind of corpus the RAG system is built on and how corpus properties relate to downstream performance.
2οΈβ£ Chunk Corpus β rag_corpus_chunks.csv
What the retriever actually βseesβ.
Granularity: 1 row = 1 chunk
Key fields:
chunk_idβ unique chunk identifierdoc_idβ foreign key torag_corpus_documents.doc_iddomainβ propagated from the parent documentchunk_indexβ 0-based position of the chunk within the documentestimated_tokensβ approximate token lengthchunk_textβ the text content used for retrieval and ranking
Use this table to rebuild retrieval candidates, inspect chunking strategies, and study how content structure affects retrieval.
3οΈβ£ QA Evaluation Runs β rag_qa_eval_runs.csv
End-to-end evaluation records for questionβanswer runs, including labels and metrics. Granularity: 1 row = 1 QA example (one query, one answer, one configuration) Key fields: Context & content
example_id,run_idβ unique identifiersdomain,task_type,difficultyβ scenario and question typescenario_idβ link torag_qa_scenariosqueryβ user-style question textgold_answerβ reference answerhas_answer_in_corpusβ whether the corpus actually contains sufficient evidence
Quality & hallucination signals
is_correctβ main binary correctness flagcorrectness_labelβ descriptive view of correctness (e.g. correct / partial / incorrect)faithfulness_labelβ e.g. faithful / unfaithful / unknownhallucination_flagβ binary hallucination indicatoruser_feedback_labelβ simplified user-style feedbacksupervising_judge_labelβ synthetic βexpertβ judgementis_noanswer_probeβ marks deliberately unanswerable queries
Retrieval metrics
retrieval_strategy,chunking_strategyn_retrieved_chunkstop1_score,mean_retrieved_scorerecall_at_5,recall_at_10,mrr_at_10has_relevant_in_top5,has_relevant_in_top10
Latency & resource usage
latency_ms_retrieval,latency_ms_generation,total_latency_msused_long_context_window,context_window_tokens
Configuration & cost
embedding_model,reranker_model,generator_modeltemperature,top_p,max_new_tokens,stop_reasonprompt_tokens,answer_tokens,total_cost_usd
Supervision & usage
doc_ids_used,chunk_ids_usedeval_mode,created_at_utc
This table is the main entry point for meta-modeling, risk scoring, and latencyβcostβquality tradeoff analysis.
4οΈβ£ Retrieval Events β rag_retrieval_events.csv
Per-chunk retrieval telemetry for each QA example.
Granularity: 1 row = 1 retrieved chunk for an example
Key fields:
run_id,example_idβ link back torag_qa_eval_runschunk_idβ link torag_corpus_chunksrankβ rank position (1 = top)retrieval_scoreβ retriever scoreis_relevantβ relevance label for this chunkdomain,difficulty,retrieval_strategyβ redundant context fields for easier analysis
Use this table to reconstruct retrieval lists, compute custom ranking metrics, and explore how ranking quality influences final answers.
5οΈβ£ Scenarios β rag_qa_scenarios.csv
Scenario-level templates and use cases for QA runs.
Granularity: 1 row = 1 scenario
Key fields:
scenario_idβ links torag_qa_eval_runs.scenario_iddomainβ scenario domain (support, HR, finance, medical, developer docs, etc.)primary_doc_idβ anchor document for the scenarioquery,gold_answerβ canonical scenario-level QA pairdifficulty_levelβ e.g. easy / medium / hardscenario_typeβ e.g. factual QA, policy lookup, multi-hop reasoninguse_caseβ short description of the business or product scenariohas_answer_in_corpusβ whether the scenario is designed to be answerable from the corpusn_eval_examples,is_used_in_evalβ how many QA examples were generated per scenario and whether it appears in the evaluation runs
This table adds a narrative layer on top of the logs, making it easier to build dashboards, teaching materials, or explainability reports.
6οΈβ£ Data Dictionary β rag_qa_data_dictionary.csv
Column-level documentation across all tables. Granularity: 1 row = 1 column definition Key fields:
table_nameβ name of the table the column belongs tocolumn_nameβ column name in snake_casedtypeβ high-level type (int, float, bool, category, datetime, text)descriptionβ human-readable explanationallowed_valuesβ expected values or ranges where applicable
Use this file as a single source of truth when exploring or building models on top of the dataset.
π― Targets & Tasks
Typical learning targets:
is_correctβ classification: did the system answer correctly?hallucination_flagβ classification: is the answer hallucinated?faithfulness_labelβ multi-class view of answer faithfulness
Paired with rich system signals (retrieval metrics, latency, cost, configuration), these enable:- Meta-models that estimate answer quality before showing it to users
- Risk scores driving block / escalate / rerun decisions
- Policy design for when to switch retrieval strategy or model configuration
π Example Usage
Using plain pandas with the raw files:
python import pandas as pd base_path = "path/to/data" # or the local dataset path docs = pd.read_csv(f"{base_path}/rag_corpus_documents.csv") chunks = pd.read_csv(f"{base_path}/rag_corpus_chunks.csv") qa_runs = pd.read_csv(f"{base_path}/rag_qa_eval_runs.csv") retrieval_events = pd.read_csv(f"{base_path}/rag_retrieval_events.csv") scenarios = pd.read_csv(f"{base_path}/rag_qa_scenarios.csv") dictionary = pd.read_csv(f"{base_path}/rag_qa_data_dictionary.csv")
Example join: attach the top-ranked retrieved chunk to each QA example:
python top = ( retrieval_events.query("rank == 1") .merge(chunks[["chunk_id", "chunk_text"]], on="chunk_id", how="left") ) qa_with_top_chunk = qa_runs.merge( top[["run_id", "chunk_text"]], on="run_id", how="left", suffixes=("", "_top_chunk"), )
You can then train a simple meta-model on qa_with_top_chunk to predict is_correct or hallucination_flag from retrieval and configuration features.
π¬ Research & Applications
RAG meta-modeling
- Predict correctness or hallucination risk from retrieval and latency metrics
- Build guardrails that decide when to block, escalate, or rerun answers
Retrieval & ranking analysis
- Compare retrieval strategies across domains and difficulty levels
- Explore how rank, score, and recall relate to final correctness
Latency & cost trade-offs
- Study how
total_latency_ms,context_window_tokens, andtotal_cost_usdinteract with answer quality - Prototype βfast vs carefulβ modes for RAG systems
- Study how
Teaching & dashboards
- Demonstrate a realistic RAG pipeline without exposing real logs
- Build dashboards that visualise quality, latency, cost, and configuration over time
π§ Ethical Considerations
- All records are fully synthetic system logs and corpus content, not collected from real users or organisations.
- No personally identifiable information (PII) is included.
- The dataset is intended for research, teaching, benchmarking, and prototyping,
not for validating real-world systems in high-stakes domains (e.g. clinical, legal, financial decisions).
π Citation
When using this dataset in research, demos, or teaching material, please cite the dataset URL on Hugging Face and:
βRAG QA Evaluation Logs & Corpus β Synthetic Multi-Table Benchmark by Tarek Masryoβ
π License
CC BY 4.0 (Attribution Required)
You are free to use, share, and modify this dataset, provided that appropriate credit is given.
- Downloads last month
- 7