Dataset Viewer
Auto-converted to Parquet Duplicate
row_id
string
trial_id
string
setup_id
string
setup
string
family
string
branch
string
task
string
task_source
string
task_version
string
attempt
int32
reward
float64
regrade
string
exception
string
started_at
string
duration_minutes
float64
cost_usd
float64
n_agents
int32
models
list
instruction
string
agents
list
patch
string
verifier_output
string
advisor-opus-high-luna/happy-dom-deterministic-intersec__SGwb5UT
happy-dom-deterministic-intersec__SGwb5UT
advisor-opus-high-luna
Advisor + Luna (Opus high)
advisor
null
happy-dom-deterministic-intersectionobserver
DeepSWE 1.1
tasks_v2
2
1
null
null
2026-10-23T17:57:48.934018Z
63.4
2.345485
8
[ "claude-opus-5-5", "gpt-6-luna" ]
"Implement a real IntersectionObserver engine in Happy DOM with deterministic geometry handling and (...TRUNCATED)
[{"role":"worker","label":"call 1","model":"gpt-6-luna","harness":"pi","started_at":"2026-10-23T17:5(...TRUNCATED)
"diff --git a/packages/happy-dom/src/intersection-observer/IIntersectionObserverInit.ts b/packages/h(...TRUNCATED)
"[verifier] model.patch applied (51271 bytes)\n[verifier] Resetting files touched by test.patch\n[ve(...TRUNCATED)
advisor-opus-high-luna/happy-dom-deterministic-intersec__3dDWQD3
happy-dom-deterministic-intersec__3dDWQD3
advisor-opus-high-luna
Advisor + Luna (Opus high)
advisor
null
happy-dom-deterministic-intersectionobserver
DeepSWE 1.1
tasks_v2
1
1
null
null
2026-10-23T15:37:29.015920Z
36.2
1.245309
5
[ "claude-opus-5-5", "gpt-6-luna" ]
"Implement a real IntersectionObserver engine in Happy DOM with deterministic geometry handling and (...TRUNCATED)
[{"role":"worker","label":"call 1","model":"gpt-6-luna","harness":"pi","started_at":"2026-10-23T15:3(...TRUNCATED)
"diff --git a/packages/happy-dom/src/PropertySymbol.ts b/packages/happy-dom/src/PropertySymbol.ts\ni(...TRUNCATED)
"[verifier] model.patch applied (44363 bytes)\n[verifier] Resetting files touched by test.patch\n[ve(...TRUNCATED)
advisor-opus-high-luna/happy-dom-deterministic-intersec__tBZv43D
happy-dom-deterministic-intersec__tBZv43D
advisor-opus-high-luna
Advisor + Luna (Opus high)
advisor
null
happy-dom-deterministic-intersectionobserver
DeepSWE 1.1
tasks_v2
3
1
null
null
2026-10-23T20:31:08.852112Z
47.3
1.878545
6
[ "claude-opus-5-5", "gpt-6-luna" ]
"Implement a real IntersectionObserver engine in Happy DOM with deterministic geometry handling and (...TRUNCATED)
[{"role":"worker","label":"call 1","model":"gpt-6-luna","harness":"pi","started_at":"2026-10-23T20:3(...TRUNCATED)
"diff --git a/packages/happy-dom/src/intersection-observer/IIntersectionObserverInit.ts b/packages/h(...TRUNCATED)
"[verifier] model.patch applied (48873 bytes)\n[verifier] Resetting files touched by test.patch\n[ve(...TRUNCATED)
advisor-opus-high-luna/agentjido__req_llm-908__GWRWe6f
agentjido__req_llm-908__GWRWe6f
advisor-opus-high-luna
Advisor + Luna (Opus high)
advisor
null
agentjido__req_llm-908
Longtail-SWE
tasks_v2
3
1
null
AgentTimeoutError
2026-10-23T19:48:15.143329Z
91.2
null
8
[ "claude-opus-5-5", "gpt-6-luna" ]
"# Task statement\n\nV1: Add opt-in provider ownership for file references\nParent tracker: #829\nRo(...TRUNCATED)
[{"role":"advisor","label":"advisor","model":"claude-opus-5-5","harness":"claude -p (no tools)","sta(...TRUNCATED)
null
"test command exit code: 0\n * test decode_stream_event/2 decodes incomplete event [L#2126]\n * te(...TRUNCATED)
advisor-opus-high-luna/agentjido__req_llm-908__Fmite8c
agentjido__req_llm-908__Fmite8c
advisor-opus-high-luna
Advisor + Luna (Opus high)
advisor
null
agentjido__req_llm-908
Longtail-SWE
tasks_v2
2
0
null
null
2026-10-23T17:23:26.478452Z
53.2
1.103021
3
[ "claude-opus-5-5", "gpt-6-luna" ]
"V1: Add opt-in provider ownership for file references\nParent tracker: #829\nRoadmap: [ReqLLM 1.x r(...TRUNCATED)
[{"role":"worker","label":"call 1","model":"gpt-6-luna","harness":"pi","started_at":"2026-10-23T17:2(...TRUNCATED)
null
"test command exit code: 2\n * test decode_stream_event/2 decodes incomplete event [L#2126]\n * te(...TRUNCATED)
advisor-opus-high-luna/agentjido__req_llm-908__8jHNGzL
agentjido__req_llm-908__8jHNGzL
advisor-opus-high-luna
Advisor + Luna (Opus high)
advisor
null
agentjido__req_llm-908
Longtail-SWE
tasks_v2
1
0
null
null
2026-10-23T15:20:18.846915Z
54.3
1.841954
4
[ "claude-opus-5-5", "gpt-6-luna" ]
"V1: Add opt-in provider ownership for file references\nParent tracker: #829\nRoadmap: [ReqLLM 1.x r(...TRUNCATED)
[{"role":"worker","label":"call 1","model":"gpt-6-luna","harness":"pi","started_at":"2026-10-23T15:2(...TRUNCATED)
null
"test command exit code: 2\n * test decode_stream_event/2 decodes incomplete event [L#2126]\n * te(...TRUNCATED)
advisor-opus-high-luna/termenv-preserve-ansi-resets__6bEcP2P
termenv-preserve-ansi-resets__6bEcP2P
advisor-opus-high-luna
Advisor + Luna (Opus high)
advisor
null
termenv-preserve-ansi-resets
DeepSWE 1.1
tasks_v2
2
0
null
null
2026-10-23T19:23:11.110393Z
25.1
1.356751
5
[ "claude-opus-5-5", "gpt-6-luna" ]
"Add preserve-resets and ANSI-safe truncation to termenv. Create an ansi subpackage exporting: Token(...TRUNCATED)
[{"role":"worker","label":"call 1","model":"gpt-6-luna","harness":"pi","started_at":"2026-10-23T19:2(...TRUNCATED)
"diff --git a/ansi/ansi.go b/ansi/ansi.go\nnew file mode 100644\nindex 0000000..e9fe80d\n--- /dev/nu(...TRUNCATED)
"[verifier] model.patch applied (30618 bytes)\n[verifier] Resetting files touched by test.patch\n[ve(...TRUNCATED)
advisor-opus-high-luna/termenv-preserve-ansi-resets__oexAkAp
termenv-preserve-ansi-resets__oexAkAp
advisor-opus-high-luna
Advisor + Luna (Opus high)
advisor
null
termenv-preserve-ansi-resets
DeepSWE 1.1
tasks_v2
3
1
contract-fix
null
2026-10-23T21:45:48.749262Z
18.5
1.031129
5
[ "claude-opus-5-5", "gpt-6-luna" ]
"Add preserve-resets and ANSI-safe truncation to termenv. Create an ansi subpackage exporting: Token(...TRUNCATED)
[{"role":"worker","label":"call 1","model":"gpt-6-luna","harness":"pi","started_at":"2026-10-23T21:4(...TRUNCATED)
"diff --git a/ansi.go b/ansi.go\nnew file mode 100644\nindex 0000000..f8bdca0\n--- /dev/null\n+++ b/(...TRUNCATED)
"[verifier] model.patch applied (30628 bytes)\n[verifier] Resetting files touched by test.patch\n[ve(...TRUNCATED)
advisor-opus-high-luna/termenv-preserve-ansi-resets__ZwcXgf4
termenv-preserve-ansi-resets__ZwcXgf4
advisor-opus-high-luna
Advisor + Luna (Opus high)
advisor
null
termenv-preserve-ansi-resets
DeepSWE 1.1
tasks_v2
1
1
contract-fix
null
2026-10-23T17:06:23.173962Z
24.6
0.84026
4
[ "claude-opus-5-5", "gpt-6-luna" ]
"Add preserve-resets and ANSI-safe truncation to termenv. Create an ansi subpackage exporting: Token(...TRUNCATED)
[{"role":"worker","label":"call 1","model":"gpt-6-luna","harness":"pi","started_at":"2026-10-23T17:0(...TRUNCATED)
"diff --git a/ansi/ansi.go b/ansi/ansi.go\nnew file mode 100644\nindex 0000000..236da61\n--- /dev/nu(...TRUNCATED)
"[verifier] model.patch applied (31632 bytes)\n[verifier] Resetting files touched by test.patch\n[ve(...TRUNCATED)
advisor-opus-high-luna/sqlfmt-create-table-ddl-formatti__yT5a2EV
sqlfmt-create-table-ddl-formatti__yT5a2EV
advisor-opus-high-luna
Advisor + Luna (Opus high)
advisor
null
sqlfmt-create-table-ddl-formatting
DeepSWE 1.1
tasks_v2
1
1
sqlfmt-fixture
null
2026-10-23T16:37:31.019759Z
77.4
3.194233
6
[ "claude-opus-5-5", "gpt-6-luna" ]
"This task has two deliverables: (1) the formatting behavior defined by requirements 1-8; and (2) th(...TRUNCATED)
[{"role":"worker","label":"call 1","model":"gpt-6-luna","harness":"pi","started_at":"2026-10-23T16:3(...TRUNCATED)
"diff --git a/CHANGELOG.md b/CHANGELOG.md\nindex d76fd0c..590a5dc 100644\n--- a/CHANGELOG.md\n+++ b/(...TRUNCATED)
"[verifier] model.patch applied (63123 bytes)\n[verifier] Resetting files touched by test.patch\n[ve(...TRUNCATED)
End of preview. Expand in Data Studio

Hard12 agent trajectories

Full trajectories of solo and multi-agent coding setups on Hard12 (12 hard SWE tasks, 3 attempts each, run with Harbor). Leaderboard: https://ibragim.dev/leaderboard-v2/

One row = one trial of one setup. reward (0/1) is the final score; regrade names the verifier fix that changed it, if any. agents lists every agent call (role, model, harness, cost, full messages) in order. Review setups are branches of shared runs, so a trial_id can appear in several setups.

setup_id pass@1 rows
sol-6.1-high-codex 87.5% (8 tasks) 24
sol-6.1-medium-codex 86.1% 36
terra-6.1-high-pi 86.1% 36
opus-5.5-medium 83.3% 36
terra-6.1-medium-pi 83.3% 36
sol-6.1-low-codex 80.6% 36
kimi-k3.1-pro-pi 80.6% 36
fable-5.2-medium 77.8% 36
luna-6-pi 69.4% 36
opus-5.5-low 63.9% 36
deepseek-v4.1-flash-1017-opencode 63.6% (11 tasks) 32
kimi-k3-pi 61.1% 36
deepseek-v4.1-flash-opencode 47.2% 36
luna-6-codex 37.5% (8 tasks) 24
deepseek-v4-flash-0731-pi 20.8% (8 tasks) 24
qwen3.8-27b-pi 12.5% (8 tasks) 24
reflection-beam-pi 8.3% (8 tasks) 24
qwen3.8-27b-team-harness 30.6% (12 tasks) 35
orch-opus-qwen 86.1% 36
soft-opus-ds1017-neutral 86.1% 36
orch-opus-ds1017 83.3% 36
orch-opus-dsflash 80.6% 36
porch-opus-ds1017 80.6% 36
orch-opus-luna 75.0% 36
soft-opus-ds1017-delegate-first 72.2% 36
orch-opus-haiku 66.7% 36
pipeline-opus-high-luna 58.3% 36
pipeline-opus-medium-luna 58.3% 36
advisor-opus-high-luna 61.1% 36
advisor-opus-medium-luna 58.3% (12 tasks) 35
pause-opus-high-luna 63.9% 36
pause-opus-high-qwen 50.0% 36
review-sol-medium-base 88.9% 36
review-sol-medium-self 86.1% 36
review-sol-medium-cross-opus 75.0% 36
review-opus-low-cross-sol 72.2% 36
review-opus-low-base 61.1% 36
review-opus-low-self 61.1% 36
loop-ds1017-sol-review 81.8% (11 tasks) 30
loop-ds1017-base 63.6% (11 tasks) 30
loop-qwen-sol-review 44.4% 36
loop-qwen-base 33.3% 36

Removed: trials whose logs contained the operator's API keys (16 rows) and one solution-leak trial. Setups marked with fewer than 12 tasks lost the other tasks' data. Secrets are redacted. Task content keeps the licenses of the source repositories.

Downloads last month
-