conversations listlengths 2 208 | agent stringclasses 1
value | model stringclasses 1
value | model_provider stringclasses 1
value | date stringlengths 27 27 | task stringlengths 9 9 | episode stringclasses 71
values | run_id stringlengths 18 18 | trial_name stringlengths 18 18 | result stringclasses 5
values | instruction stringlengths 1 29.9k | verifier_output stringlengths 48 82.5M ⌀ |
|---|---|---|---|---|---|---|---|---|---|---|---|
[
{
"content": "",
"role": "user"
},
{
"content": "1. 读取输入数据，获取时钟数量 N 和每个时钟的周期 T_i。\n2. 计算所有时钟周期的最小公倍数（LCM），这将给出所有时钟再次同时指向正上方所需的秒数。\n3. 使用辗转相除法（欧几里得算法）计算最大公约数（GCD），然后用公式 LCM(a,b) = a * b / GCD(a,b) 来计算最小公倍数。\n4. 对于多个数的 LCM，可以逐步计算：先算前两个数的 LCM，再将结果与第三个数计算 LCM，依此类推。\n5. 将最终结果写入文件 `/app/solution.py`。\n\n\... | opencode | hosted_vllm/Qwen3-Coder-30B-A3B-Instruct | hosted_vllm | 2026-07-27T02:17:36.971962Z | taco-5724 | episode-4 | taco-5724__eKizZPe | taco-5724__eKizZPe | 0.0 | 1. 读取输入数据，获取时钟数量 N 和每个时钟的周期 T_i。
2. 计算所有时钟周期的最小公倍数（LCM），这将给出所有时钟再次同时指向正上方所需的秒数。
3. 使用辗转相除法（欧几里得算法）计算最大公约数（GCD），然后用公式 LCM(a,b) = a * b / GCD(a,b) 来计算最小公倍数。
4. 对于多个数的 LCM，可以逐步计算：先算前两个数的 LCM，再将结果与第三个数计算 LCM，依此类推。
5. 将最终结果写入文件 `/app/solution.py`。 | Running test 0...
Test 0: FAILED
Expected: 6
Actual:
Running test 1...
Test 1: FAILED
Expected: 1000000000000000000
Actual:
Running test 10...
Test 10: FAILED
Expected: 276539701556344345
Actual:
Running test 100...
Test 100: FAILED
Expected: 1012103002011322440420121023023011000
Actual:
Running test 101...
Test 10... |
[
{
"content": "",
"role": "user"
},
{
"content": "I'll solve this step by step.\n\nFirst, let me understand what we're looking for:\n1. We need to find the largest contiguous subarray where all elements are congruent modulo some value m ≥ 2\n2. This means for a subarray, all elements must have the sa... | opencode | hosted_vllm/Qwen3-Coder-30B-A3B-Instruct | hosted_vllm | 2026-07-27T00:18:06.622560Z | taco-8889 | episode-13 | taco-8889__ks8ht7m | taco-8889__ks8ht7m | 0.0 | I'll solve this step by step.
First, let me understand what we're looking for:
1. We need to find the largest contiguous subarray where all elements are congruent modulo some value m ≥ 2
2. This means for a subarray, all elements must have the same remainder when divided by some m ≥ 2
Let me examine the examples to c... | Running test 0...
Test 0: PASSED
Running test 1...
Test 1: FAILED
Expected: 1
1
6
2
3
Actual: 1\n1\n6\n2\n3
Running test 10...
Test 10: PASSED
Running test 100...
Test 100: PASSED
Running test 101...
Test 101: FAILED
Expected: 3
3
2
6
Actual: 3\n3\n2\n6
Running test 102...
Test 102: PASSED
Running test 103...
Test 103:... |
[{"content":"","role":"user"},{"content":"I'll solve this step by step.\n\nFirst, let's understand t(...TRUNCATED) | opencode | hosted_vllm/Qwen3-Coder-30B-A3B-Instruct | hosted_vllm | 2026-07-27T15:22:42.446980Z | taco-5828 | episode-16 | taco-5828__ZjcBsnx | taco-5828__ZjcBsnx | 0.0 | "I'll solve this step by step.\n\nFirst, let's understand the problem:\n1. We have n points on a 2D (...TRUNCATED) | "Running test 0...\nTest 0: FAILED\nExpected: 281\nActual: File \"/app/solution.py\", line 45\n (...TRUNCATED) |
[{"content":"","role":"user"},{"content":"Looking at this problem, I need to:\n\n1. Parse input reco(...TRUNCATED) | opencode | hosted_vllm/Qwen3-Coder-30B-A3B-Instruct | hosted_vllm | 2026-07-27T14:01:21.912029Z | taco-9815 | episode-7 | taco-9815__XB5dCcx | taco-9815__XB5dCcx | 0.0 | "Looking at this problem, I need to:\n\n1. Parse input records of players and their scores\n2. Find (...TRUNCATED) | "Running test 0...\nTest 0: FAILED\nExpected: 10\na noob\nb noob\nc noob\nd noob\ne random\nf random(...TRUNCATED) |
[{"content":"","role":"user"},{"content":"I'll solve this step by step.\n\nFirst, let me understand (...TRUNCATED) | opencode | hosted_vllm/Qwen3-Coder-30B-A3B-Instruct | hosted_vllm | 2026-07-27T18:45:50.493238Z | taco-7883 | episode-0 | taco-7883__UWhJTgt | taco-7883__UWhJTgt | 0.0 | "I'll solve this step by step.\n\nFirst, let me understand the problem requirements:\n1. Find a time(...TRUNCATED) | "Running test 0...\nTest 0: FAILED\nExpected: 99\nActual: python3: can't open file '/app/solution.py(...TRUNCATED) |
[{"content":"","role":"user"},{"content":"[{\"id\":\"call_1722082162160431551340\",\"function\":{\"a(...TRUNCATED) | opencode | hosted_vllm/Qwen3-Coder-30B-A3B-Instruct | hosted_vllm | 2026-07-27T01:25:26.686629Z | taco-1762 | episode-0 | taco-1762__4VVcTC2 | taco-1762__4VVcTC2 | 0.0 | "[{\"id\":\"call_1722082162160431551340\",\"function\":{\"arguments\":\"{\\\"url\\\":\\\"https://ope(...TRUNCATED) | "Running test 0...\nTest 0: FAILED\nExpected: {}\nActual: python3: can't open file '/app/solution.py(...TRUNCATED) |
[{"content":"","role":"user"},{"content":"(tool use)\n<tool_call>\n{\"name\": \"glob\", \"arguments\(...TRUNCATED) | opencode | hosted_vllm/Qwen3-Coder-30B-A3B-Instruct | hosted_vllm | 2026-07-27T01:49:49.891814Z | taco-5867 | episode-14 | taco-5867__FSiEN7L | taco-5867__FSiEN7L | 1.0 | (tool use) | "Running test 0...\nTest 0: PASSED\nRunning test 1...\nTest 1: PASSED\nRunning test 10...\nTest 10: (...TRUNCATED) |
[{"content":"","role":"user"},{"content":"We'll solve this step-by-step:\n\n1. First, let's understa(...TRUNCATED) | opencode | hosted_vllm/Qwen3-Coder-30B-A3B-Instruct | hosted_vllm | 2026-07-27T18:04:35.076850Z | taco-5366 | episode-0 | taco-5366__U4BvhXw | taco-5366__U4BvhXw | 0.0 | "We'll solve this step-by-step:\n\n1. First, let's understand the problem:\n - We have two arrays:(...TRUNCATED) | "Running test 0...\nTest 0: FAILED\nExpected: 2\nActual: python3: can't open file '/app/solution.py'(...TRUNCATED) |
[{"content":"","role":"user"},{"content":"(tool use)\n<tool_call>\n{\"name\": \"write\", \"arguments(...TRUNCATED) | opencode | hosted_vllm/Qwen3-Coder-30B-A3B-Instruct | hosted_vllm | 2026-07-27T00:53:16.605477Z | taco-8020 | episode-1 | taco-8020__Z9RhfAv | taco-8020__Z9RhfAv | 0.0 | (tool use) | "Running test 0...\nTest 0: FAILED\nExpected: 2\nActual: 0\nRunning test 1...\nTest 1: FAILED\nExpec(...TRUNCATED) |
[{"content":"","role":"user"},{"content":"Let me analyze the coding task carefully.\n\n## Problem Un(...TRUNCATED) | opencode | hosted_vllm/Qwen3-Coder-30B-A3B-Instruct | hosted_vllm | 2026-07-27T03:27:50.696770Z | taco-6888 | episode-29 | taco-6888__JZvLbHz | taco-6888__JZvLbHz | 1.0 | "Let me analyze the coding task carefully.\n\n## Problem Understanding\n\n1. Two boys compete in a t(...TRUNCATED) | "Running test 0...\nTest 0: PASSED\nRunning test 1...\nTest 1: PASSED\nRunning test 10...\nTest 10: (...TRUNCATED) |
End of preview. Expand in Data Studio
Agent trace dataset
OpenCode/Harbor rollout traces from the MarinSkyRL run
rl-tasktrove-dq-sweep-30b-qwen3-coder-30-20260726-235656-574ba8, exported with
make_and_upload_trace_dataset --episodes last (the last episode of each trial — the rollouts
the policy was trained on).
Coverage
Built from the complete trial set on durable object storage, not from a local evidence bundle.
| quantity | value |
|---|---|
| trial directories on object storage | 21711 |
trials with a result.json (scored) |
21619 |
| rows published | 21604 |
| coverage against scored trials | 99.9% |
Trials without a result.json were cut off mid-episode and never scored: they carry no reward and
no verifier output, so they are excluded.
Secret scan
Every shard was scanned before publication, over decoded string cells rather than the compressed
parquet bytes, for JWTs, AWS access-key ids, sk- / hf_ / gh*_ tokens, and PEM private-key
markers.
The Iris proxy token that authenticates rollouts to the inference endpoint appears in zero
shards. That is structural, not luck: the exporter reads only agent/trajectory.json, the task
instruction, the reward, and verifier output. It never reads config.json, exception.txt,
agent/opencode.txt, or the full result.json — the result column holds just the reward string.
The token is written into config.json and result.json on the source object store, and neither
file is an export input.
Four rows contain secret-shaped strings. Each was inspected individually; none is a credential, and all four are agent-transcript or task-prompt content that is kept deliberately — masking it would corrupt the research artifact to satisfy a pattern match.
| shard | what it is |
|---|---|
train-00068 |
HS256 JWT expired 2019-06-05, scraped web content inside a task |
train-00089 |
GitHub raw.githubusercontent.com presigned-asset JWT, expired 2026-07-27; a public CDN URL the agent fetched |
train-00012 |
The TACO task prompt itself, showing an example key structure with an elided body (MIIEowIBAAKCAQEA...) — no key material |
train-00100 |
A model-hallucinated key body in an edit to Debian's ssl-cert-snakeoil.key placeholder: not valid base64, ragged line lengths, fails load_der_private_key |
- Downloads last month
- 5