Hugging Face's logo

Datasets:
akashnaren
/
agent-ui-efficiency-scores

Tasks:
Tabular Regression
Tabular Classification
Modalities:
Tabular
Text
Formats:
json
Languages:
English
Size:
< 1K
Tags:
agents
tool-use
ui
efficiency
Synthetic
lab-synthetic
Libraries:
Datasets
pandas
Polars
License:
Dataset card Data Studio Files
xet
 Community
Dataset Viewer
Auto-converted to Parquet Duplicate
trace_id
string
task
string
ui_mode
string
tokens
int64
turns
int64
recoveries
int64
latency_ms
int64
schema_errors
int64
ok
bool
source_note
string
tr-cli-9
restart_worker
cli
310
2
0
1,400
0
true
lab-synthetic; values match fiction in published agent-ui-sft aui-cli-004 tool payload
tr-cli-100
metrics_export
cli
470
1
0
1,753
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `metrics_export`); NOT production telemetry
tr-cli-104
metrics_query
cli
274
2
1
1,082
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `metrics_query`); NOT production telemetry
tr-cli-108
workflow_cancel
cli
322
2
0
1,404
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `workflow_cancel`); NOT production telemetry
tr-cli-112
workflow_signal
cli
329
3
0
1,331
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `workflow_signal`); NOT production telemetry
tr-cli-116
workflow_list
cli
244
1
0
856
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `workflow_list`); NOT production telemetry
tr-cli-120
telemetry_diagnose
cli
515
3
0
2,252
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `telemetry_diagnose`); NOT production telemetry
tr-cli-124
form_fill
cli
672
4
0
2,740
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `form_fill`); NOT production telemetry
tr-cli-128
form_validate
cli
539
3
2
2,361
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `form_validate`); NOT production telemetry
tr-cli-132
dom_navigation
cli
864
5
2
3,872
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `dom_navigation`); NOT production telemetry
tr-cli-136
selector_recovery
cli
728
4
0
3,096
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `selector_recovery`); NOT production telemetry
tr-cli-140
worker_restart_ui
cli
329
2
0
1,415
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `worker_restart_ui`); NOT production telemetry
tr-cli-144
ui_efficiency_score
cli
391
2
1
1,638
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `ui_efficiency_score`); NOT production telemetry
tr-cli-148
metrics_ui
cli
418
2
1
1,699
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `metrics_ui`); NOT production telemetry
tr-cli-152
workflow_describe
cli
300
3
1
1,200
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `workflow_describe`); NOT production telemetry
tr-cli-bun24
ui_efficiency_score
cli
640
4
0
2,800
0
true
lab-synthetic; bun-24 appears as a high-turn fiction id in agent-ui-sft aui-cli-025; this row is a staged CLI counterpart (NOT the raw aggregate)
tr-api-3
restart_worker
structured_api
520
3
0
2,100
0
true
lab-synthetic; values match fiction in published agent-ui-sft aui-cli-010 tool payload
tr-api-101
metrics_export
structured_api
680
3
0
2,400
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `metrics_export`); NOT production telemetry
tr-api-105
metrics_query
structured_api
499
1
1
1,958
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `metrics_query`); NOT production telemetry
tr-api-109
workflow_cancel
structured_api
489
2
0
1,930
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `workflow_cancel`); NOT production telemetry
tr-api-113
workflow_signal
structured_api
474
3
1
1,888
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `workflow_signal`); NOT production telemetry
tr-api-117
workflow_list
structured_api
404
1
0
1,512
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `workflow_list`); NOT production telemetry
tr-api-121
telemetry_diagnose
structured_api
720
2
1
2,800
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `telemetry_diagnose`); NOT production telemetry
tr-api-125
form_fill
structured_api
816
2
1
3,048
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `form_fill`); NOT production telemetry
tr-api-129
form_validate
structured_api
840
3
1
3,178
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `form_validate`); NOT production telemetry
tr-api-133
dom_navigation
structured_api
1,034
3
1
4,284
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `dom_navigation`); NOT production telemetry
tr-api-137
selector_recovery
structured_api
972
3
1
3,724
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `selector_recovery`); NOT production telemetry
tr-api-141
worker_restart_ui
structured_api
540
4
0
2,200
1
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `worker_restart_ui`); NOT production telemetry
tr-api-145
ui_efficiency_score
structured_api
604
2
1
2,447
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `ui_efficiency_score`); NOT production telemetry
tr-api-149
metrics_ui
structured_api
611
2
0
2,380
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `metrics_ui`); NOT production telemetry
tr-api-153
workflow_describe
structured_api
441
2
1
1,664
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `workflow_describe`); NOT production telemetry
tr-form-2
restart_worker
form
1,670
5
0
8,800
0
true
lab-synthetic; values match fiction in published agent-ui-sft aui-cli-017 tool payload
tr-ui-104
restart_worker
form
1,840
6
1
9,200
0
true
lab-synthetic; id appears in published agent-ui-sft aui-cli-001 tool payload
tr-form-102
metrics_export
form
1,995
7
1
10,080
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `metrics_export`); NOT production telemetry
tr-form-106
metrics_query
form
1,736
6
0
8,548
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `metrics_query`); NOT production telemetry
tr-form-110
workflow_cancel
form
1,938
5
1
9,753
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `workflow_cancel`); NOT production telemetry
tr-form-114
workflow_signal
form
1,820
4
1
9,288
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `workflow_signal`); NOT production telemetry
tr-form-118
workflow_list
form
1,456
4
0
7,430
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `workflow_list`); NOT production telemetry
tr-form-122
telemetry_diagnose
form
2,616
7
1
12,864
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `telemetry_diagnose`); NOT production telemetry
tr-form-126
form_fill
form
1,122
4
0
5,588
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `form_fill`); NOT production telemetry
tr-form-130
form_validate
form
1,248
5
1
6,292
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `form_validate`); NOT production telemetry
tr-form-134
dom_navigation
form
1,792
5
0
8,768
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `dom_navigation`); NOT production telemetry
tr-form-138
selector_recovery
form
1,710
6
1
9,120
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `selector_recovery`); NOT production telemetry
tr-form-142
worker_restart_ui
form
1,634
5
1
8,544
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `worker_restart_ui`); NOT production telemetry
tr-form-146
ui_efficiency_score
form
1,605
4
0
8,659
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `ui_efficiency_score`); NOT production telemetry
tr-form-150
metrics_ui
form
2,180
6
1
10,720
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `metrics_ui`); NOT production telemetry
tr-form-154
workflow_describe
form
1,464
4
0
7,560
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `workflow_describe`); NOT production telemetry
tr-dom-9
restart_worker
dom_click
4,120
11
2
19,000
0
true
lab-synthetic; values match fiction in published agent-ui-sft aui-cli-004 tool payload
tr-dom-103
metrics_export
dom_click
5,720
14
3
25,920
0
false
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `metrics_export`); NOT production telemetry
tr-dom-107
metrics_query
dom_click
3,762
10
2
16,864
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `metrics_query`); NOT production telemetry
tr-dom-111
workflow_cancel
dom_click
4,680
11
2
21,672
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `workflow_cancel`); NOT production telemetry
tr-dom-115
workflow_signal
dom_click
4,343
12
3
20,160
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `workflow_signal`); NOT production telemetry
tr-dom-119
workflow_list
dom_click
3,584
9
1
16,440
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `workflow_list`); NOT production telemetry
tr-dom-123
telemetry_diagnose
dom_click
6,466
15
3
29,344
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `telemetry_diagnose`); NOT production telemetry
tr-dom-127
form_fill
dom_click
2,632
8
1
13,328
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `form_fill`); NOT production telemetry
tr-dom-131
form_validate
dom_click
3,300
9
3
16,740
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `form_validate`); NOT production telemetry
tr-dom-135
dom_navigation
dom_click
2,442
5
0
9,792
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `dom_navigation`); NOT production telemetry
tr-dom-139
selector_recovery
dom_click
3,920
10
4
24,112
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `selector_recovery`); NOT production telemetry
tr-dom-143
worker_restart_ui
dom_click
4,495
11
2
20,128
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `worker_restart_ui`); NOT production telemetry
tr-dom-147
ui_efficiency_score
dom_click
3,708
9
1
16,384
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `ui_efficiency_score`); NOT production telemetry
tr-dom-151
metrics_ui
dom_click
5,136
14
2
24,288
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `metrics_ui`); NOT production telemetry
tr-dom-155
workflow_describe
dom_click
3,604
10
2
16,244
0
true
lab-synthetic bakeoff row; task/mode vocabulary aligned with published agent-ui-sft (task_type-like label `workflow_describe`); NOT production telemetry
tr-dom-bun24
ui_efficiency_score
dom_click
6,400
32
5
48,000
0
true
lab-synthetic; tokens/turns styled after bun-24 fiction in agent-ui-sft aui-cli-025 tool payload; NOT live measurements

Agent UI Efficiency Scores

Flat lab-synthetic bakeoff table for the public question in akashnaren/agent-ui-metrics: which agent UI is cheapest for a given lab task?

Author Akash Premkumar (akashnaren)
License Apache-2.0
Hub files train.jsonl (63), test.jsonl (14), optional scores.jsonl (77 full)
Related agent-ui-sft, agent-ui-human, ui-mode-router, agent-ui-mode-pairs

Scope / disclaimer (read first)

Rows are lab-authored staged scores (tokens, turns, recoveries, latency). They are:

  • NOT Tesla telemetry
  • NOT customer data
  • NOT production / employer system measurements

Identifiers such as tr-cli-9, tr-dom-9, tr-ui-104, and wf-synth-44 are harness fiction. A few numeric fields match JSON already published inside akashnaren/agent-ui-sft tool payloads (checkable on Hub); the rest use the same vocabulary with clearly labeled synthetic bakeoff rows.

Why this exists

agent-ui-sft embeds score-like JSON inside multi-turn chat. A flat table is easier for:

  • plotting tokens/turns by ui_mode
  • teaching a tiny regressor / ranker for “cheap UI”
  • documenting the question repo with a concrete, citable artifact

Pair with ui-mode-router (text→mode) vs this set (mode→staged cost).

Schema

column type required meaning
trace_id string yes Lab trace id (e.g. tr-cli-9)
task string yes Lab job name (restart_worker, metrics_export, …)
ui_mode string enum yes cli | structured_api | dom_click | form
tokens int yes Staged token count for the trace
turns int yes Staged tool/chat turns
recoveries int yes Staged recovery count (selector miss, validation retry, …)
latency_ms int yes Staged wall time (ms)
schema_errors int yes Staged API schema errors (usually 0)
ok bool yes Whether the lab job completed
source_note string yes Provenance; marks synthetic / derived-from-SFT fiction

Split / coverage (checkable on Hub)

split rows
train 63
test 14
total 77

Approximate balance by ui_mode (full set): cli 20, structured_api 18, form 19, dom_click 20.

Tasks include: restart_worker, metrics_export, metrics_query, workflow_cancel, workflow_signal, workflow_list, workflow_start, workflow_describe, workflow_cancel_ui, telemetry_diagnose, telemetry_ui, form_fill, form_validate, dom_navigation, selector_recovery, worker_restart_ui, ui_efficiency_score, metrics_ui — labels aligned with task_type / tool fiction in published agent-ui-sft.

Rows tied to published SFT fiction (examples)

These numbers match or extend staged fiction already visible in published SFT tool payloads:

trace_id task ui_mode tokens turns recoveries latency_ms SFT pointer
tr-cli-9 restart_worker cli 310 2 0 1400 aui-cli-004
tr-api-3 restart_worker structured_api 520 3 0 2100 aui-cli-010
tr-form-2 restart_worker form 1670 5 0 8800 aui-cli-017
tr-dom-9 restart_worker dom_click 4120 11 2 19000 aui-cli-004
tr-ui-104 restart_worker form 1840 6 1 9200 aui-cli-001

They are not live-cluster measurements.

Intended use

  • Tables / charts for the agent-ui-metrics write-up
  • Toy ranking: given task, predict cheapest ui_mode by tokens or turns
  • Eval companion to agent-ui-mode-pairs

Limitations

  • Synthetic / staged; not production telemetry
  • Small (n<1K); English task names only
  • Must not be marketed as measured efficiency on real worker fleets
  • No claimed SOTA; no production accuracy claims

Citation / links

Downloads last month
54
Number of rows:
77
Total file size:
57.5 kB

Collection including akashnaren/agent-ui-efficiency-scores