You need to agree to share your Hugging Face information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this dataset content.

web-benchmark · map_sep13 agent runs

Browser-agent runs on the Map Arena site, collected while building the map_sep13 task set (2026-09-14 to 2026-09-16). One row per run, with the task it attempted, the agent's answer, the grader verdict, and the full step-by-step trajectory.

  • Agent: browser-use, model openai/gpt-5.6-sol via OpenRouter, max 30 steps
  • Runs: 311 (runs voided because the API ran out of credits, or because a task wording could not be graded, are excluded)
  • Tasks referenced: 296
  • Screenshots are not included.

Files

file one row per contents
map_sep13/runs.jsonl run task id, level, template, intent, answer, verdict, per-check results, timings, tokens, cost, and trajectory: every step's thinking, evaluation, memory, next goal, actions, results, URL, and site clock time
map_sep13/conversations.jsonl run llm_conversation: the full text sent to the model at each step (system prompt, agent state, browser state)
map_sep13/tasks.jsonl task intent, answer checks (eval), and task metadata

Join the three on task_id; join runs and conversations on run_id.

Levels

  • L1: time-independent lookups, including Street View.
  • L2: answer only after an event happens.
  • L3: answer before a deadline.
  • L4: answer within an interval after an event.

"Route A" tasks run the site clock 6x faster with whole-minute ETAs; "route B" tasks run at 1x with per-second ETAs.

level runs passed pass rate
L1 105 49 47%
L2 75 21 28%
L3 70 26 37%
L4 61 24 39%

These are pass rates over the runs in this file, not over the final task set: pilots and reruns are included.

Batches

batch runs
L2 pilot, first run 14
L2 pilot, clean baseline 14
L3 pilot (latest ETA) 2
L3/L4 pilot batch 24
map_sep13 half sample 124
map_sep13 remaining half 111
street-view fix rerun 13
street-view navigation-hint rerun 9

Verdicts

passed is the verdict recorded at run time. passed_under_current_task_definition re-grades the same answer with the current task definition at the moment the agent submitted (null where the task was later deleted). They differ for 3 runs:

  • run 106 map_l2_19150: recorded fail, pass under the current definition
  • run 226 map_l1_20052: recorded fail, pass under the current definition
  • run 432 map_l1_20017: recorded fail, pass under the current definition

Run 106 was originally graded against a clock read after the browser closed; the other two belong to Street View tasks whose answer checks were corrected after the run (see below).

Known issues in the source tasks

  • Street View outside New York. 514 of the 533 panoramas are genuine New York imagery. The rest are single photos taken elsewhere (Indonesia, Copenhagen, an Alpine forest) placed at US city coordinates. Tasks that required the city name were changed on 2026-09-16 to grade only the photo content, and tasks that said "a Chicago dot" were pinned to one panorama's coordinates.
  • Brief events. Some route-B gap/lead tasks have a condition that holds for only a few seconds before reverting. They were kept deliberately (the page re-prices every second, so the condition is visible), but failures on them partly reflect chance. Their task metadata carries brief_event.
  • Tight L3 windows. In several L3 "latest value" tasks the gap between the scheduled ETA change and the deadline is shorter than the agent's median read-then-submit time (about 17 s).
  • Six passed L2 pilot tasks were deleted from the task set; their tasks.jsonl rows are recovered from the run conversations and have no answer checks.

Final experiments: map_sep26 (155 tasks) and stock_sep25 (136 tasks)

Agent runs of the final Map (155 tasks) and Stock-trading (136 tasks) task sets, four models × three time conditions, one attempt per task. Screenshots are not included.

Conditions

All runs use browser-use via OpenRouter, max_steps=30, and a simulated website clock reset before every task.

condition website clock in the prompt wait action
C1 no capped
C2 no uncapped (advances the simulated clock in full)
C3 yes, at second precision uncapped

Models: openai/gpt-5.6-sol, google/gemini-3.8-flash, deepseek/deepseek-v4-pro-0813, qwen/qwen3.5-122b-a10b.

Files

Each folder has the same four files. Join on run_id; join to tasks on task_id.

file one row per contents
tasks.jsonl task intent, answer checks (eval), level, template, task metadata
runs.jsonl run model, condition, verdict (passed, status), answer, per-check results, timing, tokens
conversations.jsonl run with a trajectory llm_conversation: the full prompt and reply of every model call
steps.jsonl run with a trajectory steps: per-step model output, actions, results, URL, website time

Coverage

map_sep26 — 1860 runs (155 tasks × 4 models × 3 conditions, complete).

model C1 runs / passed / with trajectory C2 C3
gpt-5.6-sol 155 / 34 / 155 155 / 32 / 155 155 / 22 / 152
gemini-3.8-flash 155 / 28 / 155 155 / 31 / 155 155 / 19 / 143
deepseek-v4-pro-0813 155 / 12 / 155 155 / 8 / 155 155 / 16 / 133
qwen3.5-122b-a10b 155 / 20 / 135 155 / 15 / 133 155 / 18 / 116

Every trajectory is matched to its own run (same task, step count, final answer and clock condition), not taken from a shared directory. trajectory_complete=false marks runs cut short by a timeout, whose trajectory stops early. 118 runs have no trajectory: their archived directories were empty (69), had been overwritten by a later run of another model on the same task (46), or never existed (3). Their results are still in runs.jsonl (has_trajectory=false).

stock_sep25 — 1575 runs; every run has its trajectory.

model C1 runs / passed C2 C3
gpt-5.6-sol 136 / 19 136 / 27 136 / 29
gemini-3.8-flash 136 / 23 136 / 19 136 / 20
deepseek-v4-pro-0813 136 / 19 136 / 22 136 / 21
qwen3.5-122b-a10b 130 / 18 91 / 6 130 / 18

qwen3.5-122b ran 130, 91 and 130 of the 136 tasks in C1, C2 and C3. status="error" marks runs that did not complete for infrastructure reasons; exclude them from pass rates.

Changes to map_sep26 (2026-11-10)

The earlier upload of map_sep26/ replaced hostnames with a pattern that also matched inside ordinary words ("required" → "requi__HOST__", "captured" → "captu__HOST__"; about 40,000 occurrences in steps.jsonl). This version is rebuilt from the raw run records with that fixed, and adds trajectories for gpt-5.6-sol C3 and other runs the earlier build had missed (1742 runs with a trajectory, up from 1667). Run rows now carry run_id, has_trajectory and trajectory_complete.

Downloads last month
731