evaluation_id string | scenario_id string | shuffle_seed int64 | roles list | fragments list | distractor_id string | reference_need_sets unknown | role_assignment_prompt string | prompt_writing_prompt string |
|---|---|---|---|---|---|---|---|---|
pg_000__seed_1 | pg_000 | 1 | [
"coder",
"reviewer"
] | [
{
"id": "f1",
"text": "After finding a bug, submit it to a **fresh** reviewer for review. Do not review your own report.\n\n**After the review returns:** execute all action steps in the reviewer's output.",
"is_distractor": false
},
{
"id": "f2",
"text": "Claude responds well to clear, expli... | f2 | {
"coder": [
"f5",
"f6",
"f7",
"f1"
],
"reviewer": [
"f5",
"f3",
"f4"
]
} | I need you to set up a 2-agent pipeline for bug bounty.
The two roles are coder and reviewer: coder is responsible for finding bugs, reviewer is responsible for judging whether what the coder found counts as a bug.
After each bug the coder finds, they must spawn a **fresh** reviewer to evaluate it — this drives the loo... | I need you to set up a 2-agent pipeline for bug bounty.
The two roles are coder and reviewer: coder is responsible for finding bugs, reviewer is responsible for judging whether what the coder found counts as a bug.
After each bug the coder finds, they must spawn a **fresh** reviewer to evaluate it — this drives the loo... |
pg_000__seed_42 | pg_000 | 42 | [
"coder",
"reviewer"
] | [
{
"id": "f1",
"text": "This project has been running stably for many years — any obvious fatal bug would have killed it long ago. So:\n\n- Don't report problems that are obvious at a glance (e.g. \"some map is never erased\") — if it were that obvious, the project team would have fixed it themselves\n- Trul... | f5 | {
"coder": [
"f6",
"f3",
"f1",
"f2"
],
"reviewer": [
"f6",
"f7",
"f4"
]
} | I need you to set up a 2-agent pipeline for bug bounty.
The two roles are coder and reviewer: coder is responsible for finding bugs, reviewer is responsible for judging whether what the coder found counts as a bug.
After each bug the coder finds, they must spawn a **fresh** reviewer to evaluate it — this drives the loo... | I need you to set up a 2-agent pipeline for bug bounty.
The two roles are coder and reviewer: coder is responsible for finding bugs, reviewer is responsible for judging whether what the coder found counts as a bug.
After each bug the coder finds, they must spawn a **fresh** reviewer to evaluate it — this drives the loo... |
pg_001__seed_1 | pg_001 | 1 | [
"coder",
"reviewer"
] | [
{
"id": "f1",
"text": "After finding a bug, submit it to a **fresh** reviewer for review. Do not review your own report.\n\n**After the review returns:** execute all action steps in the reviewer's output. Then immediately go back to step 1: re-read ideas.xml and pick the next idea by UCB to keep working. Do... | f2 | {
"coder": [
"f5",
"f3",
"f7",
"f1"
],
"reviewer": [
"f5",
"f4",
"f6"
]
} | I need you to set up a 2-agent pipeline for bug bounty. The two roles are coder and reviewer: coder is responsible for finding bugs, reviewer is responsible for judging whether what the coder found counts as a bug. The flow is roughly:
```
WHILE TRUE:
1. coder picks an idea it likes and explores it deeply, or appe... | I need you to set up a 2-agent pipeline for bug bounty. The two roles are coder and reviewer: coder is responsible for finding bugs, reviewer is responsible for judging whether what the coder found counts as a bug. The flow is roughly:
```
WHILE TRUE:
1. coder picks an idea it likes and explores it deeply, or appe... |
pg_001__seed_42 | pg_001 | 42 | [
"coder",
"reviewer"
] | [
{
"id": "f1",
"text": "This project has been running stably for many years — any obvious fatal bug would have killed it long ago. So:\n\n- Don't report problems that are obvious at a glance (e.g. \"some map is never erased\") — if it were that obvious, the project team would have fixed it themselves\n- Trul... | f5 | {
"coder": [
"f6",
"f7",
"f1",
"f2"
],
"reviewer": [
"f6",
"f4",
"f3"
]
} | I need you to set up a 2-agent pipeline for bug bounty. The two roles are coder and reviewer: coder is responsible for finding bugs, reviewer is responsible for judging whether what the coder found counts as a bug. The flow is roughly:
```
WHILE TRUE:
1. coder picks an idea it likes and explores it deeply, or appe... | I need you to set up a 2-agent pipeline for bug bounty. The two roles are coder and reviewer: coder is responsible for finding bugs, reviewer is responsible for judging whether what the coder found counts as a bug. The flow is roughly:
```
WHILE TRUE:
1. coder picks an idea it likes and explores it deeply, or appe... |
pg_002__seed_1 | pg_002 | 1 | [
"dispatcher",
"coder",
"reviewer"
] | [
{
"id": "f1",
"text": "Your only job is to alternate calls to the coder and reviewer agents, driving the explore → write → review → improve loop.\n\nRead the remaining-round count from `counter.txt`. If the file doesn't exist or the count ≤ 0, print \"rounds exhausted\" and stop.\n\nLoop:\n\n```\nwhile coun... | f2 | {
"dispatcher": [
"f1"
],
"coder": [
"f5",
"f7",
"f3"
],
"reviewer": [
"f5",
"f4",
"f6",
"f3"
]
} | I need you to set up a 3-agent pipeline for bug bounty: dispatcher, coder, and reviewer.
- **dispatcher** orchestrates: calls coder, extracts the report file path from coder's reply, then calls reviewer on that path, decrements a counter, and loops. Does no analysis itself.
- **coder** explores the codebase, picks an ... | I need you to set up a 3-agent pipeline for bug bounty: dispatcher, coder, and reviewer.
- **dispatcher** orchestrates: calls coder, extracts the report file path from coder's reply, then calls reviewer on that path, decrements a counter, and loops. Does no analysis itself.
- **coder** explores the codebase, picks an ... |
pg_002__seed_42 | pg_002 | 42 | [
"dispatcher",
"coder",
"reviewer"
] | [
{
"id": "f1",
"text": "This project has been running stably for many years — any obvious fatal bug would have killed it long ago. So:\n\n- Don't report problems that are obvious at a glance (e.g. \"some map is never erased\") — if it were that obvious, the project team would have fixed it themselves\n- Trul... | f5 | {
"dispatcher": [
"f2"
],
"coder": [
"f6",
"f1",
"f7"
],
"reviewer": [
"f6",
"f4",
"f3",
"f7"
]
} | I need you to set up a 3-agent pipeline for bug bounty: dispatcher, coder, and reviewer.
- **dispatcher** orchestrates: calls coder, extracts the report file path from coder's reply, then calls reviewer on that path, decrements a counter, and loops. Does no analysis itself.
- **coder** explores the codebase, picks an ... | I need you to set up a 3-agent pipeline for bug bounty: dispatcher, coder, and reviewer.
- **dispatcher** orchestrates: calls coder, extracts the report file path from coder's reply, then calls reviewer on that path, decrements a counter, and loops. Does no analysis itself.
- **coder** explores the codebase, picks an ... |
pg_003__seed_1 | pg_003 | 1 | [
"dispatcher",
"coder",
"reviewer"
] | [
{
"id": "f1",
"text": "When you finish, briefly report to the dispatcher what you did.",
"is_distractor": false
},
{
"id": "f2",
"text": "Each round, before calling the coder, read ideas.xml and select which idea the coder should work on:\n\n1. Every idea carries an `attempts` count and a re... | f7 | {
"dispatcher": [
"f5",
"f2"
],
"coder": [
"f6",
"f8",
"f3",
"f1"
],
"reviewer": [
"f6",
"f9",
"f4",
"f1"
]
} | I need you to set up a 3-agent pipeline for bug bounty: dispatcher, coder, and reviewer.
- **dispatcher** orchestrates: each round it selects which idea to pursue next, calls the coder on that idea, extracts the report file path from the coder's reply, then calls the reviewer on that path, decrements a counter, and lo... | I need you to set up a 3-agent pipeline for bug bounty: dispatcher, coder, and reviewer.
- **dispatcher** orchestrates: each round it selects which idea to pursue next, calls the coder on that idea, extracts the report file path from the coder's reply, then calls the reviewer on that path, decrements a counter, and lo... |
pg_003__seed_42 | pg_003 | 42 | [
"dispatcher",
"coder",
"reviewer"
] | [
{
"id": "f1",
"text": "You drive the explore → write → review → improve loop: each round, select the next idea, then alternate calls to the coder and reviewer agents.\n\nRead the remaining-round count from `counter.txt`. If the file doesn't exist or the count ≤ 0, print \"rounds exhausted\" and stop.\n\nLoo... | f5 | {
"dispatcher": [
"f1",
"f2"
],
"coder": [
"f8",
"f9",
"f3",
"f7"
],
"reviewer": [
"f8",
"f6",
"f4",
"f7"
]
} | I need you to set up a 3-agent pipeline for bug bounty: dispatcher, coder, and reviewer.
- **dispatcher** orchestrates: each round it selects which idea to pursue next, calls the coder on that idea, extracts the report file path from the coder's reply, then calls the reviewer on that path, decrements a counter, and lo... | I need you to set up a 3-agent pipeline for bug bounty: dispatcher, coder, and reviewer.
- **dispatcher** orchestrates: each round it selects which idea to pursue next, calls the coder on that idea, extracts the report file path from the coder's reply, then calls the reviewer on that path, decrements a counter, and lo... |
pg_004__seed_1 | pg_004 | 1 | [
"dispatcher",
"plan-creator",
"plan-critic",
"coder",
"code-critic"
] | [
{
"id": "f1",
"text": "Keep `SOLUTION.md` lean. Its body — everything except the `<review>` block — must stay under 3000 words (~20KB), about a four-page note; record only the final state and move iteration-by-iteration history into a separate `CHANGELOG.md`. Bold text must not exceed 5% of the file: if eve... | f7 | {
"dispatcher": [
"f2"
],
"plan-creator": [
"f3",
"f10",
"f9",
"f5"
],
"plan-critic": [
"f3",
"f4",
"f8",
"f5"
],
"coder": [
"f3",
"f12",
"f6",
"f1"
],
"code-critic": [
"f3",
"f11",
"f8"
]
} | I need you to set up a 5-agent pipeline that solves mathematical optimization problems. A problem arrives as a natural-language statement in `problem.md`, and the pipeline turns it into a vetted plan and then working, verified solver code.
The five agents:
- **dispatcher** orchestrates the run: it drives the plan loo... | I need you to set up a 5-agent pipeline that solves mathematical optimization problems. A problem arrives as a natural-language statement in `problem.md`, and the pipeline turns it into a vetted plan and then working, verified solver code.
The five agents:
- **dispatcher** orchestrates the run: it drives the plan loo... |
pg_004__seed_42 | pg_004 | 42 | [
"dispatcher",
"plan-creator",
"plan-critic",
"coder",
"code-critic"
] | [
{
"id": "f1",
"text": "Keep `SOLUTION.md` lean. Its body — everything except the `<review>` block — must stay under 3000 words (~20KB), about a four-page note; record only the final state and move iteration-by-iteration history into a separate `CHANGELOG.md`. Bold text must not exceed 5% of the file: if eve... | f8 | {
"dispatcher": [
"f7"
],
"plan-creator": [
"f10",
"f11",
"f9",
"f2"
],
"plan-critic": [
"f10",
"f4",
"f12",
"f2"
],
"coder": [
"f10",
"f3",
"f6",
"f1"
],
"code-critic": [
"f10",
"f5",
"f12"
]
} | I need you to set up a 5-agent pipeline that solves mathematical optimization problems. A problem arrives as a natural-language statement in `problem.md`, and the pipeline turns it into a vetted plan and then working, verified solver code.
The five agents:
- **dispatcher** orchestrates the run: it drives the plan loo... | I need you to set up a 5-agent pipeline that solves mathematical optimization problems. A problem arrives as a natural-language statement in `problem.md`, and the pipeline turns it into a vetted plan and then working, verified solver code.
The five agents:
- **dispatcher** orchestrates the run: it drives the plan loo... |
pg_005__seed_1 | pg_005 | 1 | [
"dispatcher",
"plan-creator",
"plan-refiner",
"plan-critic",
"coder",
"code-critic"
] | [
{
"id": "f1",
"text": "Keep your assigned plan's `SOLUTION.md` lean. Its body — everything except the `<review>` block — must stay under 3000 words (~20KB), about a four-page note; record only the final state and move iteration-by-iteration history into the plan's `CHANGELOG.md`. Bold text must not exceed 5... | f7 | {
"dispatcher": [
"f2"
],
"plan-creator": [
"f3",
"f9",
"f5"
],
"plan-refiner": [
"f3",
"f5",
"f10"
],
"plan-critic": [
"f3",
"f5",
"f4",
"f8"
],
"coder": [
"f3",
"f6",
"f1",
"f12"
],
"code-critic": [
"f3",
"f11",
"f8"
]
} | I need you to set up a 6-agent pipeline that solves mathematical optimization problems. A problem arrives as a natural-language statement in `problem.md`, and the pipeline turns it into a vetted plan and then working, verified solver code.
The six agents:
- **dispatcher** orchestrates the run: it drives the plan phas... | I need you to set up a 6-agent pipeline that solves mathematical optimization problems. A problem arrives as a natural-language statement in `problem.md`, and the pipeline turns it into a vetted plan and then working, verified solver code.
The six agents:
- **dispatcher** orchestrates the run: it drives the plan phas... |
pg_005__seed_42 | pg_005 | 42 | [
"dispatcher",
"plan-creator",
"plan-refiner",
"plan-critic",
"coder",
"code-critic"
] | [
{
"id": "f1",
"text": "Keep your assigned plan's `SOLUTION.md` lean. Its body — everything except the `<review>` block — must stay under 3000 words (~20KB), about a four-page note; record only the final state and move iteration-by-iteration history into the plan's `CHANGELOG.md`. Bold text must not exceed 5... | f8 | {
"dispatcher": [
"f7"
],
"plan-creator": [
"f10",
"f9",
"f2"
],
"plan-refiner": [
"f10",
"f2",
"f11"
],
"plan-critic": [
"f10",
"f2",
"f4",
"f12"
],
"coder": [
"f10",
"f6",
"f1",
"f3"
],
"code-critic": [
"f10",
"f5",
"f12"
... | I need you to set up a 6-agent pipeline that solves mathematical optimization problems. A problem arrives as a natural-language statement in `problem.md`, and the pipeline turns it into a vetted plan and then working, verified solver code.
The six agents:
- **dispatcher** orchestrates the run: it drives the plan phas... | I need you to set up a 6-agent pipeline that solves mathematical optimization problems. A problem arrives as a natural-language statement in `problem.md`, and the pipeline turns it into a vetted plan and then working, verified solver code.
The six agents:
- **dispatcher** orchestrates the run: it drives the plan phas... |
pg_006__seed_1 | pg_006 | 1 | [
"dispatcher",
"scientist",
"coder",
"reviewer"
] | [
{
"id": "f1",
"text": "Review `HIGH.md` adversarially with the standards of a top venue.\nYou are not part of the team and not their internal QA.\nYour job is to reject drift, weak evidence, and claims larger than what the data shows.\n\nCover four layers.\n\n- Drift. Does the project still attack the origi... | f11 | {
"dispatcher": [
"f2"
],
"scientist": [
"f7",
"f6",
"f8",
"f10",
"f9"
],
"coder": [
"f7",
"f5",
"f3",
"f9"
],
"reviewer": [
"f7",
"f1",
"f4",
"f9"
]
} | I need you to set up a 4-agent research pipeline.
The 4 agents are a dispatcher and 3 research agents (scientist, coder, reviewer).
The scientist plans experiments, interprets results, and responds to reviews.
The scientist writes the high-level notebook `HIGH.md`.
The body of `HIGH.md` contains the problem framing, t... | I need you to set up a 4-agent research pipeline.
The 4 agents are a dispatcher and 3 research agents (scientist, coder, reviewer).
The scientist plans experiments, interprets results, and responds to reviews.
The scientist writes the high-level notebook `HIGH.md`.
The body of `HIGH.md` contains the problem framing, t... |
pg_006__seed_42 | pg_006 | 42 | [
"dispatcher",
"scientist",
"coder",
"reviewer"
] | [
{
"id": "f1",
"text": "Score the work from 1 to 10.\nRecord the verdict by overwriting the single `<review score=X>` block at the end of `HIGH.md`.\nAlways keep exactly one `<review>` block.\nReplace it, never append a second.\n\nScore strictly on real merit.\nDo not give credit for the work being honest ab... | f3 | {
"dispatcher": [
"f4"
],
"scientist": [
"f9",
"f2",
"f8",
"f7",
"f10"
],
"coder": [
"f9",
"f5",
"f11",
"f10"
],
"reviewer": [
"f9",
"f6",
"f1",
"f10"
]
} | I need you to set up a 4-agent research pipeline.
The 4 agents are a dispatcher and 3 research agents (scientist, coder, reviewer).
The scientist plans experiments, interprets results, and responds to reviews.
The scientist writes the high-level notebook `HIGH.md`.
The body of `HIGH.md` contains the problem framing, t... | I need you to set up a 4-agent research pipeline.
The 4 agents are a dispatcher and 3 research agents (scientist, coder, reviewer).
The scientist plans experiments, interprets results, and responds to reviews.
The scientist writes the high-level notebook `HIGH.md`.
The body of `HIGH.md` contains the problem framing, t... |
pg_007__seed_1 | pg_007 | 1 | [
"dispatcher",
"theorist",
"theory-reviewer",
"coder",
"code-reviewer"
] | [
{
"id": "f1",
"text": "Keep `SOLUTION.md` lean. Its body — everything except the `<review>` block — must stay under 3000 words (~20KB), about a four-page note; record only the final state and move iteration-by-iteration history into a separate `CHANGELOG.md`. Bold text must not exceed 5% of the file: if eve... | f10 | {
"dispatcher": [
"f2"
],
"theorist": [
"f3",
"f5",
"f7",
"f11",
"f6"
],
"theory-reviewer": [
"f3",
"f7",
"f12",
"f4",
"f6"
],
"coder": [
"f3",
"f8",
"f1",
"f13",
"f6"
],
"code-reviewer": [
"f3",
"f9",
"f4",
"f6"
]
} | I need you to set up a 5-agent pipeline that solves mathematical optimization problems by developing theoretical analysis and solver code in parallel, with each side informing the other. A problem arrives as a natural-language statement in `problem.md`. The pipeline alternates between revising the theory (bounds, invar... | I need you to set up a 5-agent pipeline that solves mathematical optimization problems by developing theoretical analysis and solver code in parallel, with each side informing the other. A problem arrives as a natural-language statement in `problem.md`. The pipeline alternates between revising the theory (bounds, invar... |
pg_007__seed_42 | pg_007 | 42 | [
"dispatcher",
"theorist",
"theory-reviewer",
"coder",
"code-reviewer"
] | [
{
"id": "f1",
"text": "Read `problem.md` and `THEORY.md`, then implement the solver:\n\n- Include every import. Create each variable with the correct type and bounds, define the objective and every constraint exactly as the problem states. Solve, check the solution status, and print both the solution and th... | f10 | {
"dispatcher": [
"f5"
],
"theorist": [
"f11",
"f6",
"f2",
"f12",
"f9"
],
"theory-reviewer": [
"f11",
"f2",
"f4",
"f7",
"f9"
],
"coder": [
"f11",
"f1",
"f8",
"f3",
"f9"
],
"code-reviewer": [
"f11",
"f13",
"f7",
"f9"
]
} | I need you to set up a 5-agent pipeline that solves mathematical optimization problems by developing theoretical analysis and solver code in parallel, with each side informing the other. A problem arrives as a natural-language statement in `problem.md`. The pipeline alternates between revising the theory (bounds, invar... | I need you to set up a 5-agent pipeline that solves mathematical optimization problems by developing theoretical analysis and solver code in parallel, with each side informing the other. A problem arrives as a natural-language statement in `problem.md`. The pipeline alternates between revising the theory (bounds, invar... |
pg_008__seed_1 | pg_008 | 1 | [
"dispatcher",
"theorist",
"theory-reviewer",
"coder",
"code-reviewer"
] | [
{
"id": "f1",
"text": "Keep `SOLUTION.md` lean. Its body — everything except the `<review>` block — must stay under 3000 words (~20KB), about a four-page note; record only the final state and move iteration-by-iteration history into a separate `CHANGELOG.md`. Bold text must not exceed 5% of the file: if eve... | f10 | {
"dispatcher": [
"f2"
],
"theorist": [
"f3",
"f5",
"f7",
"f11",
"f6"
],
"theory-reviewer": [
"f3",
"f7",
"f12",
"f4",
"f6"
],
"coder": [
"f3",
"f8",
"f1",
"f13",
"f6"
],
"code-reviewer": [
"f3",
"f9",
"f4",
"f6"
]
} | I need you to set up a 5-agent pipeline that solves mathematical optimization problems by developing theoretical analysis and solver code in parallel, with each side informing the other. A problem arrives as a natural-language statement in `problem.md`. The pipeline alternates between revising the theory (bounds, invar... | I need you to set up a 5-agent pipeline that solves mathematical optimization problems by developing theoretical analysis and solver code in parallel, with each side informing the other. A problem arrives as a natural-language statement in `problem.md`. The pipeline alternates between revising the theory (bounds, invar... |
pg_008__seed_42 | pg_008 | 42 | [
"dispatcher",
"theorist",
"theory-reviewer",
"coder",
"code-reviewer"
] | [
{
"id": "f1",
"text": "You operate on the plan the dispatcher has assigned you to this round. Read `problem.md`, your assigned plan's `PLAN.md`, and your assigned plan's `THEORY.md`, then implement the solver:\n\n- Include every import. Create each variable with the correct type and bounds, define the objec... | f10 | {
"dispatcher": [
"f5"
],
"theorist": [
"f11",
"f6",
"f2",
"f12",
"f9"
],
"theory-reviewer": [
"f11",
"f2",
"f4",
"f7",
"f9"
],
"coder": [
"f11",
"f1",
"f8",
"f3",
"f9"
],
"code-reviewer": [
"f11",
"f13",
"f7",
"f9"
]
} | I need you to set up a 5-agent pipeline that solves mathematical optimization problems by developing theoretical analysis and solver code in parallel, with each side informing the other. A problem arrives as a natural-language statement in `problem.md`. The pipeline alternates between revising the theory (bounds, invar... | I need you to set up a 5-agent pipeline that solves mathematical optimization problems by developing theoretical analysis and solver code in parallel, with each side informing the other. A problem arrives as a natural-language statement in `problem.md`. The pipeline alternates between revising the theory (bounds, invar... |
pg_009__seed_1 | pg_009 | 1 | [
"supervisor",
"phdstudent",
"librarian"
] | [
{
"id": "f1",
"text": "You are a geotechnical engineering literature fetcher.\nYour sole task is to take a literature request from the caller, search open-access academic sources, download and verify candidate PDFs, and register them.\n\nWorkflow.\n\n- Restate the request as you understand it in one sentenc... | f7 | {
"supervisor": [
"f6",
"f8",
"f9",
"f3"
],
"phdstudent": [
"f6",
"f5",
"f4",
"f3"
],
"librarian": [
"f1",
"f2"
]
} | I need you to set up a 3-agent geotechnical engineering simulation workflow.
The 3 agents are supervisor, phdstudent, and librarian.
The supervisor is a senior geotechnical engineering professor who dispatches phdstudents to do Abaqus tasks, reviews their replies, and gives next-step guidance.
The phdstudent is an ex... | I need you to set up a 3-agent geotechnical engineering simulation workflow.
The 3 agents are supervisor, phdstudent, and librarian.
The supervisor is a senior geotechnical engineering professor who dispatches phdstudents to do Abaqus tasks, reviews their replies, and gives next-step guidance.
The phdstudent is an ex... |
pg_009__seed_42 | pg_009 | 42 | [
"supervisor",
"phdstudent",
"librarian"
] | [
{
"id": "f1",
"text": "You are an experienced geotechnical engineering PhD student.\nYour job is to take the supervisor's letter, do the Abaqus simulation and pre/post-processing it asks for, then reply with a structured report.\n\nFor every Abaqus run, set a hard wall-clock limit of 600 seconds on yourself... | f5 | {
"supervisor": [
"f8",
"f9",
"f6",
"f3"
],
"phdstudent": [
"f8",
"f1",
"f4",
"f3"
],
"librarian": [
"f7",
"f2"
]
} | I need you to set up a 3-agent geotechnical engineering simulation workflow.
The 3 agents are supervisor, phdstudent, and librarian.
The supervisor is a senior geotechnical engineering professor who dispatches phdstudents to do Abaqus tasks, reviews their replies, and gives next-step guidance.
The phdstudent is an ex... | I need you to set up a 3-agent geotechnical engineering simulation workflow.
The 3 agents are supervisor, phdstudent, and librarian.
The supervisor is a senior geotechnical engineering professor who dispatches phdstudents to do Abaqus tasks, reviews their replies, and gives next-step guidance.
The phdstudent is an ex... |
pg_010__seed_1 | pg_010 | 1 | [
"patent_attorney",
"patent_examiner"
] | [
{
"id": "f1",
"text": "After drafting a patent claim, submit it to a **fresh** patent_examiner for review. Do not review your own draft.\n\n**After the review returns:** execute all action steps in the patent_examiner's output.",
"is_distractor": false
},
{
"id": "f2",
"text": "Claude respon... | f2 | {
"patent_attorney": [
"f5",
"f6",
"f7",
"f1"
],
"patent_examiner": [
"f5",
"f3",
"f4"
]
} | I need you to set up a 2-agent pipeline for a highly technical utility patent application.
The two roles are patent_attorney and patent_examiner: patent_attorney is responsible for drafting the broadest possible claims, patent_examiner is responsible for searching for prior art or obviousness to reject them.
After each... | I need you to set up a 2-agent pipeline for a highly technical utility patent application.
The two roles are patent_attorney and patent_examiner: patent_attorney is responsible for drafting the broadest possible claims, patent_examiner is responsible for searching for prior art or obviousness to reject them.
After each... |
pg_010__seed_42 | pg_010 | 42 | [
"patent_attorney",
"patent_examiner"
] | [
{
"id": "f1",
"text": "This technological field has been heavily researched for many years — any obvious, broad mechanism would have been patented or published long ago. So:\n\n- Don't draft claims that cover standard, obvious industry practices at a glance (e.g. \"a basic client-server architecture\") — if... | f5 | {
"patent_attorney": [
"f6",
"f3",
"f1",
"f2"
],
"patent_examiner": [
"f6",
"f7",
"f4"
]
} | I need you to set up a 2-agent pipeline for a highly technical utility patent application.
The two roles are patent_attorney and patent_examiner: patent_attorney is responsible for drafting the broadest possible claims, patent_examiner is responsible for searching for prior art or obviousness to reject them.
After each... | I need you to set up a 2-agent pipeline for a highly technical utility patent application.
The two roles are patent_attorney and patent_examiner: patent_attorney is responsible for drafting the broadest possible claims, patent_examiner is responsible for searching for prior art or obviousness to reject them.
After each... |
pg_011__seed_1 | pg_011 | 1 | [
"investigative_journalist",
"libel_lawyer"
] | [
{
"id": "f1",
"text": "After drafting an allegation, submit it to a **fresh** libel_lawyer for vetting. Do not vet your own draft.\n\n**After the vetting returns:** execute all action steps in the libel_lawyer's output.",
"is_distractor": false
},
{
"id": "f2",
"text": "Claude responds well ... | f2 | {
"investigative_journalist": [
"f5",
"f6",
"f7",
"f1"
],
"libel_lawyer": [
"f5",
"f3",
"f4"
]
} | I need you to set up a 2-agent pipeline for the pre-publication vetting of an explosive investigative journalism exposé.
The two roles are investigative_journalist and libel_lawyer: investigative_journalist is responsible for digging up shocking allegations and drafting the story, libel_lawyer is responsible for ruthle... | I need you to set up a 2-agent pipeline for the pre-publication vetting of an explosive investigative journalism exposé.
The two roles are investigative_journalist and libel_lawyer: investigative_journalist is responsible for digging up shocking allegations and drafting the story, libel_lawyer is responsible for ruthle... |
pg_011__seed_42 | pg_011 | 42 | [
"investigative_journalist",
"libel_lawyer"
] | [
{
"id": "f1",
"text": "This subject has survived intense public scrutiny for many years — any obvious scandal would have destroyed their career long ago. So:\n\n- Don't report allegations that are obvious at a glance (e.g. \"they sometimes yell at staff\") — if it were that obvious and provable, tabloid med... | f5 | {
"investigative_journalist": [
"f6",
"f3",
"f1",
"f2"
],
"libel_lawyer": [
"f6",
"f7",
"f4"
]
} | I need you to set up a 2-agent pipeline for the pre-publication vetting of an explosive investigative journalism exposé.
The two roles are investigative_journalist and libel_lawyer: investigative_journalist is responsible for digging up shocking allegations and drafting the story, libel_lawyer is responsible for ruthle... | I need you to set up a 2-agent pipeline for the pre-publication vetting of an explosive investigative journalism exposé.
The two roles are investigative_journalist and libel_lawyer: investigative_journalist is responsible for digging up shocking allegations and drafting the story, libel_lawyer is responsible for ruthle... |
pg_012__seed_1 | pg_012 | 1 | [
"quantitative_researcher",
"risk_manager"
] | [
{
"id": "f1",
"text": "After finding an alpha signal, submit it to a **fresh** risk_manager for review. Do not review your own report.\n\n**After the review returns:** execute all action steps in the risk_manager's output.",
"is_distractor": false
},
{
"id": "f2",
"text": "Claude responds we... | f2 | {
"quantitative_researcher": [
"f5",
"f6",
"f7",
"f1"
],
"risk_manager": [
"f5",
"f3",
"f4"
]
} | I need you to set up a 2-agent pipeline for algorithmic trading strategy validation.
The two roles are quantitative_researcher and risk_manager: quantitative_researcher is responsible for developing trading strategies and claiming high market alpha, risk_manager is responsible for poking holes in the strategy by findin... | I need you to set up a 2-agent pipeline for algorithmic trading strategy validation.
The two roles are quantitative_researcher and risk_manager: quantitative_researcher is responsible for developing trading strategies and claiming high market alpha, risk_manager is responsible for poking holes in the strategy by findin... |
pg_012__seed_42 | pg_012 | 42 | [
"quantitative_researcher",
"risk_manager"
] | [
{
"id": "f1",
"text": "These markets are highly efficient and have been mined by professionals for years — any obvious arbitrage opportunity would have been traded away long ago. So:\n\n- Don't report signals that are obvious at a glance (e.g. \"simple moving average crossover\") — if it were that obvious, ... | f5 | {
"quantitative_researcher": [
"f6",
"f3",
"f1",
"f2"
],
"risk_manager": [
"f6",
"f7",
"f4"
]
} | I need you to set up a 2-agent pipeline for algorithmic trading strategy validation.
The two roles are quantitative_researcher and risk_manager: quantitative_researcher is responsible for developing trading strategies and claiming high market alpha, risk_manager is responsible for poking holes in the strategy by findin... | I need you to set up a 2-agent pipeline for algorithmic trading strategy validation.
The two roles are quantitative_researcher and risk_manager: quantitative_researcher is responsible for developing trading strategies and claiming high market alpha, risk_manager is responsible for poking holes in the strategy by findin... |
pg_013__seed_1 | pg_013 | 1 | [
"clinical_trial_sponsor",
"regulatory_assessor"
] | [
{
"id": "f1",
"text": "After compiling the clinical data package, submit it to a **fresh** regulatory_assessor for review. Do not review your own dossier.\n\n**After the review returns:** execute all action steps in the regulatory_assessor's output.",
"is_distractor": false
},
{
"id": "f2",
... | f2 | {
"clinical_trial_sponsor": [
"f5",
"f6",
"f7",
"f1"
],
"regulatory_assessor": [
"f5",
"f3",
"f4"
]
} | I need you to set up a 2-agent regulatory approval pipeline for a new pharmaceutical drug.
The two roles are clinical_trial_sponsor and regulatory_assessor: clinical_trial_sponsor is responsible for submitting clinical data proving efficacy, regulatory_assessor is responsible for rigorously auditing the methodology for... | I need you to set up a 2-agent regulatory approval pipeline for a new pharmaceutical drug.
The two roles are clinical_trial_sponsor and regulatory_assessor: clinical_trial_sponsor is responsible for submitting clinical data proving efficacy, regulatory_assessor is responsible for rigorously auditing the methodology for... |
pg_013__seed_42 | pg_013 | 42 | [
"clinical_trial_sponsor",
"regulatory_assessor"
] | [
{
"id": "f1",
"text": "This standard of care has been established for many years — any obvious broad-spectrum efficacy would have been discovered long ago. So:\n\n- Don't focus on proving outcomes that are obvious at a glance (e.g., \"the drug slightly lowers temperature\") — if it were that simple, generic... | f5 | {
"clinical_trial_sponsor": [
"f6",
"f3",
"f1",
"f2"
],
"regulatory_assessor": [
"f6",
"f7",
"f4"
]
} | I need you to set up a 2-agent regulatory approval pipeline for a new pharmaceutical drug.
The two roles are clinical_trial_sponsor and regulatory_assessor: clinical_trial_sponsor is responsible for submitting clinical data proving efficacy, regulatory_assessor is responsible for rigorously auditing the methodology for... | I need you to set up a 2-agent regulatory approval pipeline for a new pharmaceutical drug.
The two roles are clinical_trial_sponsor and regulatory_assessor: clinical_trial_sponsor is responsible for submitting clinical data proving efficacy, regulatory_assessor is responsible for rigorously auditing the methodology for... |
pg_014__seed_1 | pg_014 | 1 | [
"field_operative",
"counterintelligence_analyst"
] | [
{
"id": "f1",
"text": "After drafting an intelligence report, submit it to a **fresh** counterintelligence_analyst for review. Do not review your own report.\n\n**After the review returns:** execute all action steps in the counterintelligence_analyst's output.",
"is_distractor": false
},
{
"id":... | f2 | {
"field_operative": [
"f5",
"f6",
"f7",
"f1"
],
"counterintelligence_analyst": [
"f5",
"f3",
"f4"
]
} | I need you to set up a 2-agent pipeline for national security intelligence evaluation.
The two roles are field_operative and counterintelligence_analyst: field_operative is responsible for identifying imminent threats, counterintelligence_analyst is responsible for judging whether what the field_operative found counts ... | I need you to set up a 2-agent pipeline for national security intelligence evaluation.
The two roles are field_operative and counterintelligence_analyst: field_operative is responsible for identifying imminent threats, counterintelligence_analyst is responsible for judging whether what the field_operative found counts ... |
pg_014__seed_42 | pg_014 | 42 | [
"field_operative",
"counterintelligence_analyst"
] | [
{
"id": "f1",
"text": "The adversary's counterintelligence operations have been sophisticated and disciplined for many years — any obvious vulnerability or leaked plot would have been shut down or used as bait long ago. So:\n\n- Don't report intelligence that is obvious at a glance (e.g., \"enemy troop move... | f5 | {
"field_operative": [
"f6",
"f3",
"f1",
"f2"
],
"counterintelligence_analyst": [
"f6",
"f7",
"f4"
]
} | I need you to set up a 2-agent pipeline for national security intelligence evaluation.
The two roles are field_operative and counterintelligence_analyst: field_operative is responsible for identifying imminent threats, counterintelligence_analyst is responsible for judging whether what the field_operative found counts ... | I need you to set up a 2-agent pipeline for national security intelligence evaluation.
The two roles are field_operative and counterintelligence_analyst: field_operative is responsible for identifying imminent threats, counterintelligence_analyst is responsible for judging whether what the field_operative found counts ... |
pg_015__seed_1 | pg_015 | 1 | [
"medical_writer",
"regulatory_reviewer"
] | [
{
"id": "f1",
"text": "After drafting a claim, submit it to a **fresh** regulatory_reviewer for review. Do not review your own submission.\n\n**After the review returns:** execute all action steps in the regulatory_reviewer’s output.",
"is_distractor": false
},
{
"id": "f2",
"text": "Claude ... | f2 | {
"medical_writer": [
"f5",
"f6",
"f7",
"f1"
],
"regulatory_reviewer": [
"f5",
"f3",
"f4"
]
} | I need you to set up a 2-agent pipeline for clinical-trial evidence assessment.
The two roles are medical_writer and regulatory_reviewer: medical_writer is responsible for identifying a safety or efficacy claim supported by trial materials, regulatory_reviewer is responsible for judging whether the claim is adequately ... | I need you to set up a 2-agent pipeline for clinical-trial evidence assessment.
The two roles are medical_writer and regulatory_reviewer: medical_writer is responsible for identifying a safety or efficacy claim supported by trial materials, regulatory_reviewer is responsible for judging whether the claim is adequately ... |
pg_015__seed_42 | pg_015 | 42 | [
"medical_writer",
"regulatory_reviewer"
] | [
{
"id": "f1",
"text": "This clinical program has already been reviewed internally and the trial record has multiple layers of quality control — any obvious unsupported claim would likely have been caught during medical, statistical, or clinical review. So:\n\n- Don't submit claims that are obvious at a glan... | f5 | {
"medical_writer": [
"f6",
"f3",
"f1",
"f2"
],
"regulatory_reviewer": [
"f6",
"f7",
"f4"
]
} | I need you to set up a 2-agent pipeline for clinical-trial evidence assessment.
The two roles are medical_writer and regulatory_reviewer: medical_writer is responsible for identifying a safety or efficacy claim supported by trial materials, regulatory_reviewer is responsible for judging whether the claim is adequately ... | I need you to set up a 2-agent pipeline for clinical-trial evidence assessment.
The two roles are medical_writer and regulatory_reviewer: medical_writer is responsible for identifying a safety or efficacy claim supported by trial materials, regulatory_reviewer is responsible for judging whether the claim is adequately ... |
pg_016__seed_1 | pg_016 | 1 | [
"provenance_researcher",
"authenticity_reviewer"
] | [
{
"id": "f1",
"text": "After developing a finding, submit it to a **fresh** authenticity_reviewer for review. Do not review your own report.\n\n**After the review returns:** execute all action steps in the authenticity_reviewer’s output.",
"is_distractor": false
},
{
"id": "f2",
"text": "Cla... | f2 | {
"provenance_researcher": [
"f5",
"f6",
"f7",
"f1"
],
"authenticity_reviewer": [
"f5",
"f3",
"f4"
]
} | I need you to set up a 2-agent pipeline for art-market provenance and authenticity due diligence.
The two roles are provenance_researcher and authenticity_reviewer: provenance_researcher is responsible for developing a documented argument that an artwork is genuine, misattributed, or otherwise materially different from... | I need you to set up a 2-agent pipeline for art-market provenance and authenticity due diligence.
The two roles are provenance_researcher and authenticity_reviewer: provenance_researcher is responsible for developing a documented argument that an artwork is genuine, misattributed, or otherwise materially different from... |
pg_016__seed_42 | pg_016 | 42 | [
"provenance_researcher",
"authenticity_reviewer"
] | [
{
"id": "f1",
"text": "This object has likely passed through dealers, cataloguers, collectors, and specialists before — any obvious attribution problem would probably have been noticed already. So:\n\n- Don't report issues that are obvious at a glance (e.g. \"there is a missing year in the ownership chain\"... | f5 | {
"provenance_researcher": [
"f6",
"f3",
"f1",
"f2"
],
"authenticity_reviewer": [
"f6",
"f7",
"f4"
]
} | I need you to set up a 2-agent pipeline for art-market provenance and authenticity due diligence.
The two roles are provenance_researcher and authenticity_reviewer: provenance_researcher is responsible for developing a documented argument that an artwork is genuine, misattributed, or otherwise materially different from... | I need you to set up a 2-agent pipeline for art-market provenance and authenticity due diligence.
The two roles are provenance_researcher and authenticity_reviewer: provenance_researcher is responsible for developing a documented argument that an artwork is genuine, misattributed, or otherwise materially different from... |
pg_017__seed_1 | pg_017 | 1 | [
"tax_preparer",
"irs_examiner"
] | [
{
"id": "f1",
"text": "After developing a tax position, submit it to a **fresh** irs_examiner for review. Do not review your own position.\n\n**After the review returns:** execute all action steps in the irs_examiner's output.",
"is_distractor": false
},
{
"id": "f2",
"text": "Claude respond... | f2 | {
"tax_preparer": [
"f5",
"f6",
"f7",
"f1"
],
"irs_examiner": [
"f5",
"f3",
"f4"
]
} | I need you to set up a 2-agent pipeline for tax deduction and credit substantiation.
The two roles are tax_preparer and irs_examiner: tax_preparer is responsible for developing a claimed deduction or credit position, irs_examiner is responsible for judging whether the position the tax_preparer developed is legally vali... | I need you to set up a 2-agent pipeline for tax deduction and credit substantiation.
The two roles are tax_preparer and irs_examiner: tax_preparer is responsible for developing a claimed deduction or credit position, irs_examiner is responsible for judging whether the position the tax_preparer developed is legally vali... |
pg_017__seed_42 | pg_017 | 42 | [
"tax_preparer",
"irs_examiner"
] | [
{
"id": "f1",
"text": "The tax law and IRS audit process have been tested for many years — any obvious unsupported deduction or plainly ineligible credit will be disallowed quickly. So:\n\n- Don't propose positions that are facially defective, such as claiming a personal expense as a business deduction with... | f5 | {
"tax_preparer": [
"f6",
"f3",
"f1",
"f2"
],
"irs_examiner": [
"f6",
"f7",
"f4"
]
} | I need you to set up a 2-agent pipeline for tax deduction and credit substantiation.
The two roles are tax_preparer and irs_examiner: tax_preparer is responsible for developing a claimed deduction or credit position, irs_examiner is responsible for judging whether the position the tax_preparer developed is legally vali... | I need you to set up a 2-agent pipeline for tax deduction and credit substantiation.
The two roles are tax_preparer and irs_examiner: tax_preparer is responsible for developing a claimed deduction or credit position, irs_examiner is responsible for judging whether the position the tax_preparer developed is legally vali... |
pg_018__seed_1 | pg_018 | 1 | [
"forensic_engineer",
"plan_check_engineer"
] | [
{
"id": "f1",
"text": "After finding a defect, submit it to a **fresh** plan_check_engineer for review. Do not review your own report.\n\n**After the review returns:** execute all action steps in the plan_check_engineer's output.",
"is_distractor": false
},
{
"id": "f2",
"text": "Claude resp... | f2 | {
"forensic_engineer": [
"f5",
"f6",
"f7",
"f1"
],
"plan_check_engineer": [
"f5",
"f3",
"f4"
]
} | I need you to set up a 2-agent pipeline for structural building-defect claims.
The two roles are forensic_engineer and plan_check_engineer: forensic_engineer is responsible for identifying a structural defect or code-noncompliance issue, plan_check_engineer is responsible for judging whether what the forensic_engineer ... | I need you to set up a 2-agent pipeline for structural building-defect claims.
The two roles are forensic_engineer and plan_check_engineer: forensic_engineer is responsible for identifying a structural defect or code-noncompliance issue, plan_check_engineer is responsible for judging whether what the forensic_engineer ... |
pg_018__seed_42 | pg_018 | 42 | [
"forensic_engineer",
"plan_check_engineer"
] | [
{
"id": "f1",
"text": "This building has been occupied and maintained for many years — any obvious life-safety defect would likely have produced clear symptoms, prior enforcement action, or emergency repairs long ago. So:\n\n- Don't report conditions that are obvious at a glance without demonstrating struct... | f5 | {
"forensic_engineer": [
"f6",
"f3",
"f1",
"f2"
],
"plan_check_engineer": [
"f6",
"f7",
"f4"
]
} | I need you to set up a 2-agent pipeline for structural building-defect claims.
The two roles are forensic_engineer and plan_check_engineer: forensic_engineer is responsible for identifying a structural defect or code-noncompliance issue, plan_check_engineer is responsible for judging whether what the forensic_engineer ... | I need you to set up a 2-agent pipeline for structural building-defect claims.
The two roles are forensic_engineer and plan_check_engineer: forensic_engineer is responsible for identifying a structural defect or code-noncompliance issue, plan_check_engineer is responsible for judging whether what the forensic_engineer ... |
pg_019__seed_1 | pg_019 | 1 | [
"field_archaeologist",
"acquisitions_curator"
] | [
{
"id": "f1",
"text": "After proposing a provenience interpretation, submit it to a **fresh** acquisitions_curator for review. Do not review your own interpretation.\n\n**After the review returns:** execute all action steps in the acquisitions_curator's output.",
"is_distractor": false
},
{
"id"... | f2 | {
"field_archaeologist": [
"f5",
"f6",
"f7",
"f1"
],
"acquisitions_curator": [
"f5",
"f3",
"f4"
]
} | I need you to set up a 2-agent pipeline for archaeological artifact provenience assessment.
The two roles are field_archaeologist and acquisitions_curator: field_archaeologist is responsible for proposing an artifact's origin and archaeological context, acquisitions_curator is responsible for judging whether the propos... | I need you to set up a 2-agent pipeline for archaeological artifact provenience assessment.
The two roles are field_archaeologist and acquisitions_curator: field_archaeologist is responsible for proposing an artifact's origin and archaeological context, acquisitions_curator is responsible for judging whether the propos... |
pg_019__seed_42 | pg_019 | 42 | [
"field_archaeologist",
"acquisitions_curator"
] | [
{
"id": "f1",
"text": "This collection area has been studied for many years — any obvious attribution based only on a familiar motif, clay color, or object type is unlikely to be sufficient. So:\n\n- Don't propose origins that are obvious at a glance (e.g. \"red-slipped bowl, therefore from Site X\") — if i... | f5 | {
"field_archaeologist": [
"f6",
"f3",
"f1",
"f2"
],
"acquisitions_curator": [
"f6",
"f7",
"f4"
]
} | I need you to set up a 2-agent pipeline for archaeological artifact provenience assessment.
The two roles are field_archaeologist and acquisitions_curator: field_archaeologist is responsible for proposing an artifact's origin and archaeological context, acquisitions_curator is responsible for judging whether the propos... | I need you to set up a 2-agent pipeline for archaeological artifact provenience assessment.
The two roles are field_archaeologist and acquisitions_curator: field_archaeologist is responsible for proposing an artifact's origin and archaeological context, acquisitions_curator is responsible for judging whether the propos... |
pg_020__seed_1 | pg_020 | 1 | [
"patent_attorney",
"patent_examiner"
] | [
{
"id": "f1",
"text": "After drafting an application, submit it to a **fresh** patent_examiner for examination. Do not examine your own application.\n\n**After the examination returns:** execute all action steps in the patent_examiner's output. Then immediately go back to step 1: re-read inventions.xml and ... | f2 | {
"patent_attorney": [
"f5",
"f3",
"f7",
"f1"
],
"patent_examiner": [
"f5",
"f4",
"f6"
]
} | I need you to set up a 2-agent pipeline for patent application drafting. The two roles are patent_attorney and patent_examiner: patent_attorney is responsible for drafting patent applications from engineering concepts, patent_examiner is responsible for judging whether what the patent_attorney drafted counts as a valid... | I need you to set up a 2-agent pipeline for patent application drafting. The two roles are patent_attorney and patent_examiner: patent_attorney is responsible for drafting patent applications from engineering concepts, patent_examiner is responsible for judging whether what the patent_attorney drafted counts as a valid... |
pg_020__seed_42 | pg_020 | 42 | [
"patent_attorney",
"patent_examiner"
] | [
{
"id": "f1",
"text": "This technical field has been heavily researched for many years — any obvious engineering mechanism has long been patented. So:\n\n- Don't draft claims around obvious combinations of known elements (e.g. \"attaching a standard motor to a wheel\") — if it were that obvious, prior art w... | f5 | {
"patent_attorney": [
"f6",
"f7",
"f1",
"f2"
],
"patent_examiner": [
"f6",
"f4",
"f3"
]
} | I need you to set up a 2-agent pipeline for patent application drafting. The two roles are patent_attorney and patent_examiner: patent_attorney is responsible for drafting patent applications from engineering concepts, patent_examiner is responsible for judging whether what the patent_attorney drafted counts as a valid... | I need you to set up a 2-agent pipeline for patent application drafting. The two roles are patent_attorney and patent_examiner: patent_attorney is responsible for drafting patent applications from engineering concepts, patent_examiner is responsible for judging whether what the patent_attorney drafted counts as a valid... |
pg_021__seed_1 | pg_021 | 1 | [
"clinical_researcher",
"regulatory_reviewer"
] | [
{
"id": "f1",
"text": "After compiling a submission report, submit it to a **fresh** regulatory_reviewer for review. Do not review your own report.\n\n**After the review returns:** execute all action steps in the regulatory_reviewer's output. Then immediately go back to step 1: re-read hypotheses.xml and pi... | f2 | {
"clinical_researcher": [
"f5",
"f3",
"f7",
"f1"
],
"regulatory_reviewer": [
"f5",
"f4",
"f6"
]
} | I need you to set up a 2-agent pipeline for pharmaceutical clinical trial data submission. The two roles are clinical_researcher and regulatory_reviewer: clinical_researcher is responsible for analyzing trial data to establish drug efficacy and identify safety signals, while regulatory_reviewer is responsible for stric... | I need you to set up a 2-agent pipeline for pharmaceutical clinical trial data submission. The two roles are clinical_researcher and regulatory_reviewer: clinical_researcher is responsible for analyzing trial data to establish drug efficacy and identify safety signals, while regulatory_reviewer is responsible for stric... |
pg_021__seed_42 | pg_021 | 42 | [
"clinical_researcher",
"regulatory_reviewer"
] | [
{
"id": "f1",
"text": "This clinical trial has been analyzed extensively for years — any obvious major efficacy signal without safety trade-offs would have been proven long ago. So:\n\n- Don't report findings that are superficial or ignore key covariates (e.g., \"the raw mean is higher in the treatment grou... | f5 | {
"clinical_researcher": [
"f6",
"f7",
"f1",
"f2"
],
"regulatory_reviewer": [
"f6",
"f4",
"f3"
]
} | I need you to set up a 2-agent pipeline for pharmaceutical clinical trial data submission. The two roles are clinical_researcher and regulatory_reviewer: clinical_researcher is responsible for analyzing trial data to establish drug efficacy and identify safety signals, while regulatory_reviewer is responsible for stric... | I need you to set up a 2-agent pipeline for pharmaceutical clinical trial data submission. The two roles are clinical_researcher and regulatory_reviewer: clinical_researcher is responsible for analyzing trial data to establish drug efficacy and identify safety signals, while regulatory_reviewer is responsible for stric... |
pg_022__seed_1 | pg_022 | 1 | [
"investigative_journalist",
"fact_checker"
] | [
{
"id": "f1",
"text": "After drafting a story, submit it to a **fresh** fact_checker for aggressive legal review. Do not review your own draft.\n\n**After the review returns:** execute all action steps in the fact_checker's output. Then immediately go back to step 1: re-read ideas.xml and pick the next lead... | f2 | {
"investigative_journalist": [
"f5",
"f3",
"f7",
"f1"
],
"fact_checker": [
"f5",
"f4",
"f6"
]
} | I need you to set up a 2-agent pipeline for investigative journalism exposing high-level corporate corruption. The two roles are investigative_journalist and fact_checker: investigative_journalist is responsible for uncovering corrupt practices and drafting the exposé, while fact_checker acts as an aggressive internal ... | I need you to set up a 2-agent pipeline for investigative journalism exposing high-level corporate corruption. The two roles are investigative_journalist and fact_checker: investigative_journalist is responsible for uncovering corrupt practices and drafting the exposé, while fact_checker acts as an aggressive internal ... |
pg_022__seed_42 | pg_022 | 42 | [
"investigative_journalist",
"fact_checker"
] | [
{
"id": "f1",
"text": "This corporation has been operating its schemes securely for many years — any obvious legal violation would have triggered regulatory action long ago. So:\n\n- Don't report anomalies that are obvious at a glance (e.g., \"a corporate officer's salary seems high\") — if it were that obv... | f5 | {
"investigative_journalist": [
"f6",
"f7",
"f1",
"f2"
],
"fact_checker": [
"f6",
"f4",
"f3"
]
} | I need you to set up a 2-agent pipeline for investigative journalism exposing high-level corporate corruption. The two roles are investigative_journalist and fact_checker: investigative_journalist is responsible for uncovering corrupt practices and drafting the exposé, while fact_checker acts as an aggressive internal ... | I need you to set up a 2-agent pipeline for investigative journalism exposing high-level corporate corruption. The two roles are investigative_journalist and fact_checker: investigative_journalist is responsible for uncovering corrupt practices and drafting the exposé, while fact_checker acts as an aggressive internal ... |
pg_023__seed_1 | pg_023 | 1 | [
"mathematician",
"referee"
] | [
{
"id": "f1",
"text": "After formulating a proof, submit it to a **fresh** referee for review. Do not review your own manuscript.\n\n**After the review returns:** execute all action steps in the referee's output. Then immediately go back to step 1: re-read ideas.xml and pick the next idea by UCB to keep wor... | f2 | {
"mathematician": [
"f5",
"f3",
"f7",
"f1"
],
"referee": [
"f5",
"f4",
"f6"
]
} | I need you to set up a 2-agent pipeline for formulating complex formal proofs for long-standing mathematical conjectures. The two roles are mathematician and referee: mathematician is responsible for formulating the proofs, referee is responsible for adversarial academic peer review specifically looking for hidden assu... | I need you to set up a 2-agent pipeline for formulating complex formal proofs for long-standing mathematical conjectures. The two roles are mathematician and referee: mathematician is responsible for formulating the proofs, referee is responsible for adversarial academic peer review specifically looking for hidden assu... |
pg_023__seed_42 | pg_023 | 42 | [
"mathematician",
"referee"
] | [
{
"id": "f1",
"text": "These mathematical fields have been intensely studied for many years — any obvious elementary proof would have been discovered long ago. So:\n\n- Don't submit derivations that are obvious at a glance (e.g. \"applying standard induction\") — if it were that obvious, the mathematical co... | f5 | {
"mathematician": [
"f6",
"f7",
"f1",
"f2"
],
"referee": [
"f6",
"f4",
"f3"
]
} | I need you to set up a 2-agent pipeline for formulating complex formal proofs for long-standing mathematical conjectures. The two roles are mathematician and referee: mathematician is responsible for formulating the proofs, referee is responsible for adversarial academic peer review specifically looking for hidden assu... | I need you to set up a 2-agent pipeline for formulating complex formal proofs for long-standing mathematical conjectures. The two roles are mathematician and referee: mathematician is responsible for formulating the proofs, referee is responsible for adversarial academic peer review specifically looking for hidden assu... |
pg_024__seed_1 | pg_024 | 1 | [
"tax_strategist",
"tax_auditor"
] | [
{
"id": "f1",
"text": "After structuring a scheme, submit it to a **fresh** tax_auditor for audit. Do not audit your own proposal.\n\n**After the audit returns:** execute all action steps in the tax_auditor's output. Then immediately go back to step 1: re-read ideas.xml and pick the next idea by UCB to keep... | f2 | {
"tax_strategist": [
"f5",
"f3",
"f7",
"f1"
],
"tax_auditor": [
"f5",
"f4",
"f6"
]
} | I need you to set up a 2-agent pipeline for structuring multinational corporate tax avoidance schemes. The two roles are tax_strategist and tax_auditor: tax_strategist is responsible for finding tax loopholes and structuring aggressive avoidance schemes, tax_auditor is responsible for exhaustive auditing to disallow de... | I need you to set up a 2-agent pipeline for structuring multinational corporate tax avoidance schemes. The two roles are tax_strategist and tax_auditor: tax_strategist is responsible for finding tax loopholes and structuring aggressive avoidance schemes, tax_auditor is responsible for exhaustive auditing to disallow de... |
pg_024__seed_42 | pg_024 | 42 | [
"tax_strategist",
"tax_auditor"
] | [
{
"id": "f1",
"text": "The corporate structure and baseline tax filings have been stable for many years — any obvious deduction would have been claimed long ago. So:\n\n- Don't propose schemes that are obvious at a glance (e.g. \"simple offshore shell company\") — if it were that obvious, the corporate acco... | f5 | {
"tax_strategist": [
"f6",
"f7",
"f1",
"f2"
],
"tax_auditor": [
"f6",
"f4",
"f3"
]
} | I need you to set up a 2-agent pipeline for structuring multinational corporate tax avoidance schemes. The two roles are tax_strategist and tax_auditor: tax_strategist is responsible for finding tax loopholes and structuring aggressive avoidance schemes, tax_auditor is responsible for exhaustive auditing to disallow de... | I need you to set up a 2-agent pipeline for structuring multinational corporate tax avoidance schemes. The two roles are tax_strategist and tax_auditor: tax_strategist is responsible for finding tax loopholes and structuring aggressive avoidance schemes, tax_auditor is responsible for exhaustive auditing to disallow de... |
pg_025__seed_1 | pg_025 | 1 | [
"signal analyst",
"medical safety reviewer"
] | [
{
"id": "f1",
"text": "After documenting a safety signal, submit it to a **fresh** medical safety reviewer for review. Do not review your own report.\n\n**After the review returns:** execute all action steps in the medical safety reviewer's output. Then immediately go back to step 1: re-read ideas.xml and p... | f2 | {
"signal analyst": [
"f5",
"f3",
"f7",
"f1"
],
"medical safety reviewer": [
"f5",
"f4",
"f6"
]
} | I need you to set up a 2-agent pipeline for a pharmacovigilance signal-detection challenge. The two roles are signal analyst and medical safety reviewer: signal analyst is responsible for finding and documenting plausible drug safety signals from public evidence, medical safety reviewer is responsible for judging wheth... | I need you to set up a 2-agent pipeline for a pharmacovigilance signal-detection challenge. The two roles are signal analyst and medical safety reviewer: signal analyst is responsible for finding and documenting plausible drug safety signals from public evidence, medical safety reviewer is responsible for judging wheth... |
pg_025__seed_42 | pg_025 | 42 | [
"signal analyst",
"medical safety reviewer"
] | [
{
"id": "f1",
"text": "Most marketed drugs already have extensive labeling, trial data, and post-marketing surveillance history. Obvious known adverse reactions are rarely valuable as new signals. So:\n\n- Don't report issues that are plainly already established in the current prescribing information, risk ... | f5 | {
"signal analyst": [
"f6",
"f7",
"f1",
"f2"
],
"medical safety reviewer": [
"f6",
"f4",
"f3"
]
} | I need you to set up a 2-agent pipeline for a pharmacovigilance signal-detection challenge. The two roles are signal analyst and medical safety reviewer: signal analyst is responsible for finding and documenting plausible drug safety signals from public evidence, medical safety reviewer is responsible for judging wheth... | I need you to set up a 2-agent pipeline for a pharmacovigilance signal-detection challenge. The two roles are signal analyst and medical safety reviewer: signal analyst is responsible for finding and documenting plausible drug safety signals from public evidence, medical safety reviewer is responsible for judging wheth... |
pg_026__seed_1 | pg_026 | 1 | [
"forensic accountant",
"audit reviewer"
] | [
{
"id": "f1",
"text": "After finding a potential accounting irregularity, submit it to a **fresh** audit reviewer for review. Do not review your own report.\n\n**After the review returns:** execute all action steps in the audit reviewer's output. Then immediately go back to step 1: re-read allegations.xml a... | f2 | {
"forensic accountant": [
"f5",
"f3",
"f7",
"f1"
],
"audit reviewer": [
"f5",
"f4",
"f6"
]
} | I need you to set up a 2-agent pipeline for a forensic accounting challenge. The two roles are forensic accountant and audit reviewer: forensic accountant is responsible for finding potentially material accounting irregularities in public financial disclosures, audit reviewer is responsible for judging whether what the... | I need you to set up a 2-agent pipeline for a forensic accounting challenge. The two roles are forensic accountant and audit reviewer: forensic accountant is responsible for finding potentially material accounting irregularities in public financial disclosures, audit reviewer is responsible for judging whether what the... |
pg_026__seed_42 | pg_026 | 42 | [
"forensic accountant",
"audit reviewer"
] | [
{
"id": "f1",
"text": "Most public companies have been audited, reviewed by regulators, and scrutinized by investors for years — any obvious arithmetic typo or plainly inconsistent subtotal is likely to have been noticed already. So:\n\n- Don't report issues that are obvious at a glance, such as a single fo... | f5 | {
"forensic accountant": [
"f6",
"f7",
"f1",
"f2"
],
"audit reviewer": [
"f6",
"f4",
"f3"
]
} | I need you to set up a 2-agent pipeline for a forensic accounting challenge. The two roles are forensic accountant and audit reviewer: forensic accountant is responsible for finding potentially material accounting irregularities in public financial disclosures, audit reviewer is responsible for judging whether what the... | I need you to set up a 2-agent pipeline for a forensic accounting challenge. The two roles are forensic accountant and audit reviewer: forensic accountant is responsible for finding potentially material accounting irregularities in public financial disclosures, audit reviewer is responsible for judging whether what the... |
pg_027__seed_1 | pg_027 | 1 | [
"provenance researcher",
"authentication curator"
] | [
{
"id": "f1",
"text": "After developing a claim, submit it to a **fresh** authentication curator for review. Do not review your own dossier.\n\n**After the review returns:** execute all action steps in the authentication curator's output. Then immediately go back to step 1: re-read ideas.xml and pick the ne... | f2 | {
"provenance researcher": [
"f5",
"f3",
"f7",
"f1"
],
"authentication curator": [
"f5",
"f4",
"f6"
]
} | I need you to set up a 2-agent pipeline for a museum provenance and authenticity challenge. The two roles are provenance researcher and authentication curator: provenance researcher is responsible for developing claims that an artifact is forged, misattributed, or illicitly sourced; authentication curator is responsibl... | I need you to set up a 2-agent pipeline for a museum provenance and authenticity challenge. The two roles are provenance researcher and authentication curator: provenance researcher is responsible for developing claims that an artifact is forged, misattributed, or illicitly sourced; authentication curator is responsibl... |
pg_027__seed_42 | pg_027 | 42 | [
"provenance researcher",
"authentication curator"
] | [
{
"id": "f1",
"text": "This collection has been catalogued, exhibited, and cited for many years — any obvious problem would likely have been noticed by registrars, curators, or outside scholars. So:\n\n- Don't report issues that are obvious at a glance (e.g. \"the acquisition date is missing from the public... | f5 | {
"provenance researcher": [
"f6",
"f7",
"f1",
"f2"
],
"authentication curator": [
"f6",
"f4",
"f3"
]
} | I need you to set up a 2-agent pipeline for a museum provenance and authenticity challenge. The two roles are provenance researcher and authentication curator: provenance researcher is responsible for developing claims that an artifact is forged, misattributed, or illicitly sourced; authentication curator is responsibl... | I need you to set up a 2-agent pipeline for a museum provenance and authenticity challenge. The two roles are provenance researcher and authentication curator: provenance researcher is responsible for developing claims that an artifact is forged, misattributed, or illicitly sourced; authentication curator is responsibl... |
pg_028__seed_1 | pg_028 | 1 | [
"environmental compliance analyst",
"permit reviewer"
] | [
{
"id": "f1",
"text": "After finding a potential permitting flaw, submit it to a **fresh** permit reviewer for review. Do not review your own report.\n\n**After the review returns:** execute all action steps in the permit reviewer's output. Then immediately go back to step 1: re-read ideas.xml and pick the ... | f2 | {
"environmental compliance analyst": [
"f5",
"f3",
"f7",
"f1"
],
"permit reviewer": [
"f5",
"f4",
"f6"
]
} | I need you to set up a 2-agent pipeline for an environmental permitting challenge. The two roles are environmental compliance analyst and permit reviewer: environmental compliance analyst is responsible for identifying potential flaws in the proposed infrastructure project's environmental review, permit reviewer is res... | I need you to set up a 2-agent pipeline for an environmental permitting challenge. The two roles are environmental compliance analyst and permit reviewer: environmental compliance analyst is responsible for identifying potential flaws in the proposed infrastructure project's environmental review, permit reviewer is res... |
pg_028__seed_42 | pg_028 | 42 | [
"environmental compliance analyst",
"permit reviewer"
] | [
{
"id": "f1",
"text": "This proposed project has already been through consultant drafting, agency screening, and at least one round of public-facing review — any obvious permitting defect would likely have been corrected already. So:\n\n- Don't report problems that are obvious at a glance (e.g. \"the docume... | f5 | {
"environmental compliance analyst": [
"f6",
"f7",
"f1",
"f2"
],
"permit reviewer": [
"f6",
"f4",
"f3"
]
} | I need you to set up a 2-agent pipeline for an environmental permitting challenge. The two roles are environmental compliance analyst and permit reviewer: environmental compliance analyst is responsible for identifying potential flaws in the proposed infrastructure project's environmental review, permit reviewer is res... | I need you to set up a 2-agent pipeline for an environmental permitting challenge. The two roles are environmental compliance analyst and permit reviewer: environmental compliance analyst is responsible for identifying potential flaws in the proposed infrastructure project's environmental review, permit reviewer is res... |
pg_029__seed_1 | pg_029 | 1 | [
"prover",
"referee"
] | [
{
"id": "f1",
"text": "After finding a proof or counterexample, submit it to a **fresh** referee for review. Do not review your own submission.\n\n**After the review returns:** execute all action steps in the referee's output. Then immediately go back to step 1: re-read ideas.xml and pick the next idea by U... | f2 | {
"prover": [
"f5",
"f3",
"f7",
"f1"
],
"referee": [
"f5",
"f4",
"f6"
]
} | I need you to set up a 2-agent pipeline for a research mathematics proof challenge. The two roles are prover and referee: prover is responsible for developing proofs or counterexamples for nontrivial mathematical claims, referee is responsible for judging whether what the prover submitted is valid, complete, and actual... | I need you to set up a 2-agent pipeline for a research mathematics proof challenge. The two roles are prover and referee: prover is responsible for developing proofs or counterexamples for nontrivial mathematical claims, referee is responsible for judging whether what the prover submitted is valid, complete, and actual... |
pg_029__seed_42 | pg_029 | 42 | [
"prover",
"referee"
] | [
{
"id": "f1",
"text": "The claims in this challenge are nontrivial — any short, routine argument would likely already be known or would fail on a hidden hypothesis. So:\n\n- Don't submit arguments that are only plausible at a glance (e.g. \"apply compactness\" without verifying the space is compact) — if it... | f5 | {
"prover": [
"f6",
"f7",
"f1",
"f2"
],
"referee": [
"f6",
"f4",
"f3"
]
} | I need you to set up a 2-agent pipeline for a research mathematics proof challenge. The two roles are prover and referee: prover is responsible for developing proofs or counterexamples for nontrivial mathematical claims, referee is responsible for judging whether what the prover submitted is valid, complete, and actual... | I need you to set up a 2-agent pipeline for a research mathematics proof challenge. The two roles are prover and referee: prover is responsible for developing proofs or counterexamples for nontrivial mathematical claims, referee is responsible for judging whether what the prover submitted is valid, complete, and actual... |
pg_030__seed_1 | pg_030 | 1 | [
"docket_clerk",
"defense_attorney",
"mock_prosecutor"
] | [
{
"id": "f1",
"text": "Your only job is to alternate calls to the defense_attorney and mock_prosecutor agents, driving the research → draft → review → improve loop.\n\nRead the remaining-round count from `counter.txt`. If the file doesn't exist or the count ≤ 0, print \"rounds exhausted\" and stop.\n\nLoop:... | f2 | {
"docket_clerk": [
"f1"
],
"defense_attorney": [
"f5",
"f7",
"f3"
],
"mock_prosecutor": [
"f5",
"f4",
"f6",
"f3"
]
} | I need you to set up a 3-agent pipeline for complex litigation defense drafting: docket_clerk, defense_attorney, and mock_prosecutor.
- **docket_clerk** orchestrates: calls defense_attorney, extracts the motion draft file path from defense_attorney's reply, then calls mock_prosecutor on that path, decrements a counter... | I need you to set up a 3-agent pipeline for complex litigation defense drafting: docket_clerk, defense_attorney, and mock_prosecutor.
- **docket_clerk** orchestrates: calls defense_attorney, extracts the motion draft file path from defense_attorney's reply, then calls mock_prosecutor on that path, decrements a counter... |
pg_030__seed_42 | pg_030 | 42 | [
"docket_clerk",
"defense_attorney",
"mock_prosecutor"
] | [
{
"id": "f1",
"text": "The prosecution's case has been thoroughly vetted by their team—any obvious procedural errors or basic statutory violations would have been squashed long ago. So:\n\n- Don't file motions on basic, obvious objections (e.g., \"a minor typo in a subpoena\")—if it were that obvious, the p... | f5 | {
"docket_clerk": [
"f2"
],
"defense_attorney": [
"f6",
"f1",
"f7"
],
"mock_prosecutor": [
"f6",
"f4",
"f3",
"f7"
]
} | I need you to set up a 3-agent pipeline for complex litigation defense drafting: docket_clerk, defense_attorney, and mock_prosecutor.
- **docket_clerk** orchestrates: calls defense_attorney, extracts the motion draft file path from defense_attorney's reply, then calls mock_prosecutor on that path, decrements a counter... | I need you to set up a 3-agent pipeline for complex litigation defense drafting: docket_clerk, defense_attorney, and mock_prosecutor.
- **docket_clerk** orchestrates: calls defense_attorney, extracts the motion draft file path from defense_attorney's reply, then calls mock_prosecutor on that path, decrements a counter... |
pg_031__seed_1 | pg_031 | 1 | [
"editorial_assistant",
"investigative_journalist",
"libel_lawyer"
] | [
{
"id": "f1",
"text": "Your only job is to alternate calls to the investigative_journalist and libel_lawyer agents, driving the explore → draft → review → revise loop.\n\nRead the remaining-round count from `counter.txt`. If the file doesn't exist or the count ≤ 0, print \"deadlines exhausted\" and stop.\n\... | f2 | {
"editorial_assistant": [
"f1"
],
"investigative_journalist": [
"f5",
"f7",
"f3"
],
"libel_lawyer": [
"f5",
"f4",
"f6",
"f3"
]
} | I need you to set up a 3-agent pipeline for developing high-stakes investigative journalism exposés: editorial_assistant, investigative_journalist, and libel_lawyer.
- **editorial_assistant** orchestrates: calls investigative_journalist, extracts the draft file path from the journalist's reply, then calls libel_lawyer... | I need you to set up a 3-agent pipeline for developing high-stakes investigative journalism exposés: editorial_assistant, investigative_journalist, and libel_lawyer.
- **editorial_assistant** orchestrates: calls investigative_journalist, extracts the draft file path from the journalist's reply, then calls libel_lawyer... |
pg_031__seed_42 | pg_031 | 42 | [
"editorial_assistant",
"investigative_journalist",
"libel_lawyer"
] | [
{
"id": "f1",
"text": "This corporation has been operating for years — any blatant, surface-level fraud would have been caught by auditors or the SEC long ago. So:\n\n- Don't report allegations that are obvious at a glance (e.g. \"a random expense report is missing\") — if it were that obvious, regulators w... | f5 | {
"editorial_assistant": [
"f2"
],
"investigative_journalist": [
"f6",
"f1",
"f7"
],
"libel_lawyer": [
"f6",
"f4",
"f3",
"f7"
]
} | I need you to set up a 3-agent pipeline for developing high-stakes investigative journalism exposés: editorial_assistant, investigative_journalist, and libel_lawyer.
- **editorial_assistant** orchestrates: calls investigative_journalist, extracts the draft file path from the journalist's reply, then calls libel_lawyer... | I need you to set up a 3-agent pipeline for developing high-stakes investigative journalism exposés: editorial_assistant, investigative_journalist, and libel_lawyer.
- **editorial_assistant** orchestrates: calls investigative_journalist, extracts the draft file path from the journalist's reply, then calls libel_lawyer... |
pg_032__seed_1 | pg_032 | 1 | [
"trial_coordinator",
"lead_researcher",
"regulatory_auditor"
] | [
{
"id": "f1",
"text": "Your only job is to alternate calls to the lead_researcher and regulatory_auditor agents, driving the explore → write → audit → improve loop.\n\nRead the remaining-round count from `counter.txt`. If the file doesn't exist or the count ≤ 0, print \"rounds exhausted\" and stop.\n\nLoop:... | f2 | {
"trial_coordinator": [
"f1"
],
"lead_researcher": [
"f5",
"f7",
"f3"
],
"regulatory_auditor": [
"f5",
"f4",
"f6",
"f3"
]
} | I need you to set up a 3-agent pipeline for experimental pharmacology clinical trials: trial_coordinator, lead_researcher, and regulatory_auditor.
- **trial_coordinator** orchestrates: calls lead_researcher, extracts the clinical trial protocol file path from lead_researcher's reply, then calls regulatory_auditor on t... | I need you to set up a 3-agent pipeline for experimental pharmacology clinical trials: trial_coordinator, lead_researcher, and regulatory_auditor.
- **trial_coordinator** orchestrates: calls lead_researcher, extracts the clinical trial protocol file path from lead_researcher's reply, then calls regulatory_auditor on t... |
pg_032__seed_42 | pg_032 | 42 | [
"trial_coordinator",
"lead_researcher",
"regulatory_auditor"
] | [
{
"id": "f1",
"text": "This field is heavily saturated with safe, incremental studies. If you want a breakthrough:\n\n- Don't propose conservative, standard dosing regimens — if it were that obvious, competitor pharma companies would have already secured the IP.\n- Truly groundbreaking efficacy requires agg... | f5 | {
"trial_coordinator": [
"f2"
],
"lead_researcher": [
"f6",
"f1",
"f7"
],
"regulatory_auditor": [
"f6",
"f4",
"f3",
"f7"
]
} | I need you to set up a 3-agent pipeline for experimental pharmacology clinical trials: trial_coordinator, lead_researcher, and regulatory_auditor.
- **trial_coordinator** orchestrates: calls lead_researcher, extracts the clinical trial protocol file path from lead_researcher's reply, then calls regulatory_auditor on t... | I need you to set up a 3-agent pipeline for experimental pharmacology clinical trials: trial_coordinator, lead_researcher, and regulatory_auditor.
- **trial_coordinator** orchestrates: calls lead_researcher, extracts the clinical trial protocol file path from lead_researcher's reply, then calls regulatory_auditor on t... |
pg_033__seed_1 | pg_033 | 1 | [
"operations_officer",
"intelligence_analyst",
"red_team_challenger"
] | [
{
"id": "f1",
"text": "Your only job is to alternate calls to the intelligence_analyst and red_team_challenger agents, driving the explore → write → challenge → improve loop.\n\nRead the remaining-round count from `counter.txt`. If the file doesn't exist or the count ≤ 0, print \"rounds exhausted\" and stop... | f2 | {
"operations_officer": [
"f1"
],
"intelligence_analyst": [
"f5",
"f7",
"f3"
],
"red_team_challenger": [
"f5",
"f4",
"f6",
"f3"
]
} | I need you to set up a 3-agent pipeline for formulating tactical intelligence threat assessments: operations_officer, intelligence_analyst, and red_team_challenger.
- **operations_officer** orchestrates: calls intelligence_analyst, extracts the assessment file path from intelligence_analyst's reply, then calls red_tea... | I need you to set up a 3-agent pipeline for formulating tactical intelligence threat assessments: operations_officer, intelligence_analyst, and red_team_challenger.
- **operations_officer** orchestrates: calls intelligence_analyst, extracts the assessment file path from intelligence_analyst's reply, then calls red_tea... |
pg_033__seed_42 | pg_033 | 42 | [
"operations_officer",
"intelligence_analyst",
"red_team_challenger"
] | [
{
"id": "f1",
"text": "This theater of operations has been actively monitored for years — any obvious, massive troop movements or surface-level threats would have been flagged by automated SIGINT already. So:\n\n- Don't report anomalies that are obvious at a glance (e.g. \"comm chatter increased\") — if it ... | f5 | {
"operations_officer": [
"f2"
],
"intelligence_analyst": [
"f6",
"f1",
"f7"
],
"red_team_challenger": [
"f6",
"f4",
"f3",
"f7"
]
} | I need you to set up a 3-agent pipeline for formulating tactical intelligence threat assessments: operations_officer, intelligence_analyst, and red_team_challenger.
- **operations_officer** orchestrates: calls intelligence_analyst, extracts the assessment file path from intelligence_analyst's reply, then calls red_tea... | I need you to set up a 3-agent pipeline for formulating tactical intelligence threat assessments: operations_officer, intelligence_analyst, and red_team_challenger.
- **operations_officer** orchestrates: calls intelligence_analyst, extracts the assessment file path from intelligence_analyst's reply, then calls red_tea... |
pg_034__seed_1 | pg_034 | 1 | [
"desk_manager",
"portfolio_manager",
"risk_auditor"
] | [
{
"id": "f1",
"text": "Your only job is to alternate calls to the portfolio_manager and risk_auditor agents, driving the explore → write → audit → improve loop.\n\nRead the remaining-round count from `counter.txt`. If the file doesn't exist or the count ≤ 0, print \"rounds exhausted\" and stop.\n\nLoop:\n\n... | f2 | {
"desk_manager": [
"f1"
],
"portfolio_manager": [
"f5",
"f7",
"f3"
],
"risk_auditor": [
"f5",
"f4",
"f6",
"f3"
]
} | I need you to set up a 3-agent pipeline for a hedge fund desk: desk_manager, portfolio_manager, and risk_auditor.
- **desk_manager** orchestrates: calls portfolio_manager, extracts the pitch document path from portfolio_manager's reply, then calls risk_auditor on that path, decrements a counter, and loops. Does no ana... | I need you to set up a 3-agent pipeline for a hedge fund desk: desk_manager, portfolio_manager, and risk_auditor.
- **desk_manager** orchestrates: calls portfolio_manager, extracts the pitch document path from portfolio_manager's reply, then calls risk_auditor on that path, decrements a counter, and loops. Does no ana... |
pg_034__seed_42 | pg_034 | 42 | [
"desk_manager",
"portfolio_manager",
"risk_auditor"
] | [
{
"id": "f1",
"text": "This fund has been trading stably for many years — any obvious arbitrage opportunity would have been priced out by the market long ago. So:\n\n- Don't pitch trades that are obvious at a glance (e.g., \"the spot price and futures price diverge slightly\") — if it were that obvious, hig... | f5 | {
"desk_manager": [
"f2"
],
"portfolio_manager": [
"f6",
"f1",
"f7"
],
"risk_auditor": [
"f6",
"f4",
"f3",
"f7"
]
} | I need you to set up a 3-agent pipeline for a hedge fund desk: desk_manager, portfolio_manager, and risk_auditor.
- **desk_manager** orchestrates: calls portfolio_manager, extracts the pitch document path from portfolio_manager's reply, then calls risk_auditor on that path, decrements a counter, and loops. Does no ana... | I need you to set up a 3-agent pipeline for a hedge fund desk: desk_manager, portfolio_manager, and risk_auditor.
- **desk_manager** orchestrates: calls portfolio_manager, extracts the pitch document path from portfolio_manager's reply, then calls risk_auditor on that path, decrements a counter, and loops. Does no ana... |
pg_035__seed_1 | pg_035 | 1 | [
"case coordinator",
"safety signal analyst",
"medical safety reviewer"
] | [
{
"id": "f1",
"text": "Your only job is to alternate calls to the safety signal analyst and medical safety reviewer agents, driving the investigate → write → review → improve loop.\n\nRead the remaining-round count from `counter.txt`. If the file doesn't exist or the count ≤ 0, print \"rounds exhausted\" an... | f2 | {
"case coordinator": [
"f1"
],
"safety signal analyst": [
"f5",
"f7",
"f3"
],
"medical safety reviewer": [
"f5",
"f4",
"f6",
"f3"
]
} | I need you to set up a 3-agent pipeline for pharmacovigilance signal assessment: case coordinator, safety signal analyst, and medical safety reviewer.
- **case coordinator** orchestrates: calls safety signal analyst, extracts the dossier file path from safety signal analyst's reply, then calls medical safety reviewer ... | I need you to set up a 3-agent pipeline for pharmacovigilance signal assessment: case coordinator, safety signal analyst, and medical safety reviewer.
- **case coordinator** orchestrates: calls safety signal analyst, extracts the dossier file path from safety signal analyst's reply, then calls medical safety reviewer ... |
pg_035__seed_42 | pg_035 | 42 | [
"case coordinator",
"safety signal analyst",
"medical safety reviewer"
] | [
{
"id": "f1",
"text": "This product portfolio has been monitored for many years — any obvious labeled adverse reaction or well-known class effect should already be documented. So:\n\n- Don't escalate observations that are obvious at a glance (e.g. \"nausea is reported with the drug\") — if it were that obvi... | f5 | {
"case coordinator": [
"f2"
],
"safety signal analyst": [
"f6",
"f1",
"f7"
],
"medical safety reviewer": [
"f6",
"f4",
"f3",
"f7"
]
} | I need you to set up a 3-agent pipeline for pharmacovigilance signal assessment: case coordinator, safety signal analyst, and medical safety reviewer.
- **case coordinator** orchestrates: calls safety signal analyst, extracts the dossier file path from safety signal analyst's reply, then calls medical safety reviewer ... | I need you to set up a 3-agent pipeline for pharmacovigilance signal assessment: case coordinator, safety signal analyst, and medical safety reviewer.
- **case coordinator** orchestrates: calls safety signal analyst, extracts the dossier file path from safety signal analyst's reply, then calls medical safety reviewer ... |
pg_036__seed_1 | pg_036 | 1 | [
"audit docket clerk",
"tax credit preparer",
"IRS examining agent"
] | [
{
"id": "f1",
"text": "Your only job is to alternate calls to the tax credit preparer and IRS examining agent agents, driving the review records → write memo → examine → improve loop.\n\nRead the remaining-round count from `counter.txt`. If the file doesn't exist or the count ≤ 0, print \"rounds exhausted\"... | f2 | {
"audit docket clerk": [
"f1"
],
"tax credit preparer": [
"f5",
"f7",
"f3"
],
"IRS examining agent": [
"f5",
"f4",
"f6",
"f3"
]
} | I need you to set up a 3-agent pipeline for a corporate R&D tax credit audit: audit docket clerk, tax credit preparer, and IRS examining agent.
- **audit docket clerk** orchestrates: calls tax credit preparer, extracts the substantiation memo file path from the preparer's reply, then calls IRS examining agent on that ... | I need you to set up a 3-agent pipeline for a corporate R&D tax credit audit: audit docket clerk, tax credit preparer, and IRS examining agent.
- **audit docket clerk** orchestrates: calls tax credit preparer, extracts the substantiation memo file path from the preparer's reply, then calls IRS examining agent on that ... |
pg_036__seed_42 | pg_036 | 42 | [
"audit docket clerk",
"tax credit preparer",
"IRS examining agent"
] | [
{
"id": "f1",
"text": "This company has claimed R&D credits for many years — any obvious unsupported expense would likely have been challenged already. So:\n\n- Don't rely on problems or defenses that are obvious at a glance (e.g. \"this employee has an engineering title, therefore all wages qualify\") — if... | f5 | {
"audit docket clerk": [
"f2"
],
"tax credit preparer": [
"f6",
"f1",
"f7"
],
"IRS examining agent": [
"f6",
"f4",
"f3",
"f7"
]
} | I need you to set up a 3-agent pipeline for a corporate R&D tax credit audit: audit docket clerk, tax credit preparer, and IRS examining agent.
- **audit docket clerk** orchestrates: calls tax credit preparer, extracts the substantiation memo file path from the preparer's reply, then calls IRS examining agent on that ... | I need you to set up a 3-agent pipeline for a corporate R&D tax credit audit: audit docket clerk, tax credit preparer, and IRS examining agent.
- **audit docket clerk** orchestrates: calls tax credit preparer, extracts the substantiation memo file path from the preparer's reply, then calls IRS examining agent on that ... |
pg_037__seed_1 | pg_037 | 1 | [
"registrar",
"provenance researcher",
"authentication examiner"
] | [
{
"id": "f1",
"text": "Your only job is to alternate calls to the provenance researcher and authentication examiner agents, driving the investigate → write → examine → improve loop.\n\nRead the remaining-round count from `counter.txt`. If the file doesn't exist or the count ≤ 0, print \"rounds exhausted\" a... | f2 | {
"registrar": [
"f1"
],
"provenance researcher": [
"f5",
"f7",
"f3"
],
"authentication examiner": [
"f5",
"f4",
"f6",
"f3"
]
} | I need you to set up a 3-agent pipeline for fine-art provenance and authenticity vetting: registrar, provenance researcher, and authentication examiner.
- **registrar** orchestrates: calls provenance researcher, extracts the dossier file path from provenance researcher's reply, then calls authentication examiner on th... | I need you to set up a 3-agent pipeline for fine-art provenance and authenticity vetting: registrar, provenance researcher, and authentication examiner.
- **registrar** orchestrates: calls provenance researcher, extracts the dossier file path from provenance researcher's reply, then calls authentication examiner on th... |
pg_037__seed_42 | pg_037 | 42 | [
"registrar",
"provenance researcher",
"authentication examiner"
] | [
{
"id": "f1",
"text": "The collection has already undergone years of basic cataloguing and intake checks — any obvious mismatch would likely have been caught long ago. So:\n\n- Don't report issues that are obvious at a glance (e.g. \"the label is missing a date\") — if it were that obvious, the registrar or... | f5 | {
"registrar": [
"f2"
],
"provenance researcher": [
"f6",
"f1",
"f7"
],
"authentication examiner": [
"f6",
"f4",
"f3",
"f7"
]
} | I need you to set up a 3-agent pipeline for fine-art provenance and authenticity vetting: registrar, provenance researcher, and authentication examiner.
- **registrar** orchestrates: calls provenance researcher, extracts the dossier file path from provenance researcher's reply, then calls authentication examiner on th... | I need you to set up a 3-agent pipeline for fine-art provenance and authenticity vetting: registrar, provenance researcher, and authentication examiner.
- **registrar** orchestrates: calls provenance researcher, extracts the dossier file path from provenance researcher's reply, then calls authentication examiner on th... |
pg_038__seed_1 | pg_038 | 1 | [
"inspection scheduler",
"bridge inspection engineer",
"structural peer reviewer"
] | [
{
"id": "f1",
"text": "Your only job is to alternate calls to the bridge inspection engineer and structural peer reviewer agents, driving the investigate → write → review → improve loop.\n\nRead the remaining-round count from `counter.txt`. If the file doesn't exist or the count ≤ 0, print \"rounds exhauste... | f2 | {
"inspection scheduler": [
"f1"
],
"bridge inspection engineer": [
"f5",
"f7",
"f3"
],
"structural peer reviewer": [
"f5",
"f4",
"f6",
"f3"
]
} | I need you to set up a 3-agent pipeline for bridge inspection escalation: inspection scheduler, bridge inspection engineer, and structural peer reviewer.
- **inspection scheduler** orchestrates: calls bridge inspection engineer, extracts the defect report file path from bridge inspection engineer's reply, then calls s... | I need you to set up a 3-agent pipeline for bridge inspection escalation: inspection scheduler, bridge inspection engineer, and structural peer reviewer.
- **inspection scheduler** orchestrates: calls bridge inspection engineer, extracts the defect report file path from bridge inspection engineer's reply, then calls s... |
pg_038__seed_42 | pg_038 | 42 | [
"inspection scheduler",
"bridge inspection engineer",
"structural peer reviewer"
] | [
{
"id": "f1",
"text": "Most bridges in this program have been inspected on regular cycles for many years — any obvious severe defect would likely already be posted, repaired, or captured in prior inspection notes. So:\n\n- Don't escalate conditions that are obvious at a glance and already routine maintenanc... | f5 | {
"inspection scheduler": [
"f2"
],
"bridge inspection engineer": [
"f6",
"f1",
"f7"
],
"structural peer reviewer": [
"f6",
"f4",
"f3",
"f7"
]
} | I need you to set up a 3-agent pipeline for bridge inspection escalation: inspection scheduler, bridge inspection engineer, and structural peer reviewer.
- **inspection scheduler** orchestrates: calls bridge inspection engineer, extracts the defect report file path from bridge inspection engineer's reply, then calls s... | I need you to set up a 3-agent pipeline for bridge inspection escalation: inspection scheduler, bridge inspection engineer, and structural peer reviewer.
- **inspection scheduler** orchestrates: calls bridge inspection engineer, extracts the defect report file path from bridge inspection engineer's reply, then calls s... |
pg_039__seed_1 | pg_039 | 1 | [
"survey coordinator",
"detection analyst",
"false-positive referee"
] | [
{
"id": "f1",
"text": "Your only job is to alternate calls to the detection analyst and false-positive referee agents, driving the explore → write → review → improve loop.\n\nRead the remaining-round count from `counter.txt`. If the file doesn't exist or the count ≤ 0, print \"rounds exhausted\" and stop.\n... | f2 | {
"survey coordinator": [
"f1"
],
"detection analyst": [
"f5",
"f7",
"f3"
],
"false-positive referee": [
"f5",
"f4",
"f6",
"f3"
]
} | I need you to set up a 3-agent pipeline for astronomical survey validation: survey coordinator, detection analyst, and false-positive referee.
- **survey coordinator** orchestrates: calls detection analyst, extracts the dossier file path from detection analyst's reply, then calls false-positive referee on that path, d... | I need you to set up a 3-agent pipeline for astronomical survey validation: survey coordinator, detection analyst, and false-positive referee.
- **survey coordinator** orchestrates: calls detection analyst, extracts the dossier file path from detection analyst's reply, then calls false-positive referee on that path, d... |
pg_039__seed_42 | pg_039 | 42 | [
"survey coordinator",
"detection analyst",
"false-positive referee"
] | [
{
"id": "f1",
"text": "This survey has already passed through mature automated vetting, human triage, and several catalog-level sanity checks. Any obvious false alarm would likely have been filtered out already. So:\n\n- Don't prepare dossiers for problems that are obvious at a glance, such as a single bad ... | f5 | {
"survey coordinator": [
"f2"
],
"detection analyst": [
"f6",
"f1",
"f7"
],
"false-positive referee": [
"f6",
"f4",
"f3",
"f7"
]
} | I need you to set up a 3-agent pipeline for astronomical survey validation: survey coordinator, detection analyst, and false-positive referee.
- **survey coordinator** orchestrates: calls detection analyst, extracts the dossier file path from detection analyst's reply, then calls false-positive referee on that path, d... | I need you to set up a 3-agent pipeline for astronomical survey validation: survey coordinator, detection analyst, and false-positive referee.
- **survey coordinator** orchestrates: calls detection analyst, extracts the dossier file path from detection analyst's reply, then calls false-positive referee on that path, d... |
pg_040__seed_1 | pg_040 | 1 | [
"docket_manager",
"litigation_associate",
"mock_judge"
] | [
{
"id": "f1",
"text": "When you finish, briefly report to the docket_manager what you did.",
"is_distractor": false
},
{
"id": "f2",
"text": "Each round, before calling the litigation_associate, read arguments.xml and select which legal argument idea the litigation_associate should work on:\... | f7 | {
"docket_manager": [
"f5",
"f2"
],
"litigation_associate": [
"f6",
"f8",
"f3",
"f1"
],
"mock_judge": [
"f6",
"f9",
"f4",
"f1"
]
} | I need you to set up a 3-agent pipeline for high-stakes corporate tax defense litigation: docket_manager, litigation_associate, and mock_judge.
- **docket_manager** orchestrates: each round it selects which legal argument idea to pursue next, calls the litigation_associate on that idea, extracts the legal brief file p... | I need you to set up a 3-agent pipeline for high-stakes corporate tax defense litigation: docket_manager, litigation_associate, and mock_judge.
- **docket_manager** orchestrates: each round it selects which legal argument idea to pursue next, calls the litigation_associate on that idea, extracts the legal brief file p... |
pg_040__seed_42 | pg_040 | 42 | [
"docket_manager",
"litigation_associate",
"mock_judge"
] | [
{
"id": "f1",
"text": "You drive the research → draft → judge → refine loop: each round, select the next legal argument idea, then alternate calls to the litigation_associate and mock_judge agents.\n\nRead the remaining-round count from `counter.txt`. If the file doesn't exist or the count ≤ 0, print \"roun... | f5 | {
"docket_manager": [
"f1",
"f2"
],
"litigation_associate": [
"f8",
"f9",
"f3",
"f7"
],
"mock_judge": [
"f8",
"f6",
"f4",
"f7"
]
} | I need you to set up a 3-agent pipeline for high-stakes corporate tax defense litigation: docket_manager, litigation_associate, and mock_judge.
- **docket_manager** orchestrates: each round it selects which legal argument idea to pursue next, calls the litigation_associate on that idea, extracts the legal brief file p... | I need you to set up a 3-agent pipeline for high-stakes corporate tax defense litigation: docket_manager, litigation_associate, and mock_judge.
- **docket_manager** orchestrates: each round it selects which legal argument idea to pursue next, calls the litigation_associate on that idea, extracts the legal brief file p... |
pg_041__seed_1 | pg_041 | 1 | [
"strategy_allocator",
"quant_researcher",
"risk_manager"
] | [
{
"id": "f1",
"text": "When you finish, briefly report to the strategy_allocator what you did.",
"is_distractor": false
},
{
"id": "f2",
"text": "Each round, before calling the quant_researcher, read ideas.xml and select which idea the quant_researcher should work on:\n\n1. Every idea carrie... | f7 | {
"strategy_allocator": [
"f5",
"f2"
],
"quant_researcher": [
"f6",
"f8",
"f3",
"f1"
],
"risk_manager": [
"f6",
"f9",
"f4",
"f1"
]
} | I need you to set up a 3-agent pipeline for quantitative trading alpha discovery: strategy_allocator, quant_researcher, and risk_manager.
- **strategy_allocator** orchestrates: each round it selects which idea to pursue next, calls the quant_researcher on that idea, extracts the report file path from the quant_researc... | I need you to set up a 3-agent pipeline for quantitative trading alpha discovery: strategy_allocator, quant_researcher, and risk_manager.
- **strategy_allocator** orchestrates: each round it selects which idea to pursue next, calls the quant_researcher on that idea, extracts the report file path from the quant_researc... |
pg_041__seed_42 | pg_041 | 42 | [
"strategy_allocator",
"quant_researcher",
"risk_manager"
] | [
{
"id": "f1",
"text": "You drive the explore → write → review → improve loop: each round, select the next idea, then alternate calls to the quant_researcher and risk_manager agents.\n\nRead the remaining-round count from `counter.txt`. If the file doesn't exist or the count ≤ 0, print \"rounds exhausted\" a... | f5 | {
"strategy_allocator": [
"f1",
"f2"
],
"quant_researcher": [
"f8",
"f9",
"f3",
"f7"
],
"risk_manager": [
"f8",
"f6",
"f4",
"f7"
]
} | I need you to set up a 3-agent pipeline for quantitative trading alpha discovery: strategy_allocator, quant_researcher, and risk_manager.
- **strategy_allocator** orchestrates: each round it selects which idea to pursue next, calls the quant_researcher on that idea, extracts the report file path from the quant_researc... | I need you to set up a 3-agent pipeline for quantitative trading alpha discovery: strategy_allocator, quant_researcher, and risk_manager.
- **strategy_allocator** orchestrates: each round it selects which idea to pursue next, calls the quant_researcher on that idea, extracts the report file path from the quant_researc... |
pg_042__seed_1 | pg_042 | 1 | [
"operations_director",
"intelligence_analyst",
"red_teamer"
] | [
{
"id": "f1",
"text": "When you finish, briefly report to the operations_director what you did.",
"is_distractor": false
},
{
"id": "f2",
"text": "Each round, before calling the intelligence_analyst, read hypotheses.xml and select which hypothesis the intelligence_analyst should work on:\n\n... | f7 | {
"operations_director": [
"f5",
"f2"
],
"intelligence_analyst": [
"f6",
"f8",
"f3",
"f1"
],
"red_teamer": [
"f6",
"f9",
"f4",
"f1"
]
} | I need you to set up a 3-agent pipeline for covert supply chain intelligence: operations_director, intelligence_analyst, and red_teamer.
- **operations_director** orchestrates: each round it selects which hypothesis to pursue next, calls the intelligence_analyst on that hypothesis, extracts the dossier file path from ... | I need you to set up a 3-agent pipeline for covert supply chain intelligence: operations_director, intelligence_analyst, and red_teamer.
- **operations_director** orchestrates: each round it selects which hypothesis to pursue next, calls the intelligence_analyst on that hypothesis, extracts the dossier file path from ... |
pg_042__seed_42 | pg_042 | 42 | [
"operations_director",
"intelligence_analyst",
"red_teamer"
] | [
{
"id": "f1",
"text": "You drive the explore → write → debunk → refine loop: each round, select the next hypothesis, then alternate calls to the intelligence_analyst and red_teamer agents.\n\nRead the remaining-round count from `counter.txt`. If the file doesn't exist or the count ≤ 0, print \"rounds exhaus... | f5 | {
"operations_director": [
"f1",
"f2"
],
"intelligence_analyst": [
"f8",
"f9",
"f3",
"f7"
],
"red_teamer": [
"f8",
"f6",
"f4",
"f7"
]
} | I need you to set up a 3-agent pipeline for covert supply chain intelligence: operations_director, intelligence_analyst, and red_teamer.
- **operations_director** orchestrates: each round it selects which hypothesis to pursue next, calls the intelligence_analyst on that hypothesis, extracts the dossier file path from ... | I need you to set up a 3-agent pipeline for covert supply chain intelligence: operations_director, intelligence_analyst, and red_teamer.
- **operations_director** orchestrates: each round it selects which hypothesis to pursue next, calls the intelligence_analyst on that hypothesis, extracts the dossier file path from ... |
pg_043__seed_1 | pg_043 | 1 | [
"pipeline_controller",
"discovery_scientist",
"clinical_auditor"
] | [
{
"id": "f1",
"text": "When you finish, briefly report to the pipeline_controller what you did.",
"is_distractor": false
},
{
"id": "f2",
"text": "Each round, before calling the discovery_scientist, read hypotheses.xml and select which hypothesis the discovery_scientist should work on:\n\n1.... | f7 | {
"pipeline_controller": [
"f5",
"f2"
],
"discovery_scientist": [
"f6",
"f8",
"f3",
"f1"
],
"clinical_auditor": [
"f6",
"f9",
"f4",
"f1"
]
} | I need you to set up a 3-agent pipeline for identifying and validating novel off-label therapeutic applications for existing pharmacological compounds: pipeline_controller, discovery_scientist, and clinical_auditor.
- **pipeline_controller** orchestrates: each round it selects which hypothesis to pursue next, calls th... | I need you to set up a 3-agent pipeline for identifying and validating novel off-label therapeutic applications for existing pharmacological compounds: pipeline_controller, discovery_scientist, and clinical_auditor.
- **pipeline_controller** orchestrates: each round it selects which hypothesis to pursue next, calls th... |
pg_043__seed_42 | pg_043 | 42 | [
"pipeline_controller",
"discovery_scientist",
"clinical_auditor"
] | [
{
"id": "f1",
"text": "You drive the explore → write → audit → improve loop: each round, select the next hypothesis, then alternate calls to the discovery_scientist and clinical_auditor agents.\n\nRead the remaining-round count from `counter.txt`. If the file doesn't exist or the count ≤ 0, print \"rounds e... | f5 | {
"pipeline_controller": [
"f1",
"f2"
],
"discovery_scientist": [
"f8",
"f9",
"f3",
"f7"
],
"clinical_auditor": [
"f8",
"f6",
"f4",
"f7"
]
} | I need you to set up a 3-agent pipeline for identifying and validating novel off-label therapeutic applications for existing pharmacological compounds: pipeline_controller, discovery_scientist, and clinical_auditor.
- **pipeline_controller** orchestrates: each round it selects which hypothesis to pursue next, calls th... | I need you to set up a 3-agent pipeline for identifying and validating novel off-label therapeutic applications for existing pharmacological compounds: pipeline_controller, discovery_scientist, and clinical_auditor.
- **pipeline_controller** orchestrates: each round it selects which hypothesis to pursue next, calls th... |
pg_044__seed_1 | pg_044 | 1 | [
"assignment_desk_chief",
"investigative_reporter",
"defamation_counsel"
] | [
{
"id": "f1",
"text": "When you finish, briefly report to the assignment_desk_chief what you did.",
"is_distractor": false
},
{
"id": "f2",
"text": "Each round, before calling the investigative_reporter, read ideas.xml and select which idea the investigative_reporter should work on:\n\n1. Ev... | f7 | {
"assignment_desk_chief": [
"f5",
"f2"
],
"investigative_reporter": [
"f6",
"f8",
"f3",
"f1"
],
"defamation_counsel": [
"f6",
"f9",
"f4",
"f1"
]
} | I need you to set up a 3-agent pipeline for an investigative journalism project: assignment_desk_chief, investigative_reporter, and defamation_counsel.
- **assignment_desk_chief** orchestrates: each round it selects which story idea to pursue next, calls the investigative_reporter on that idea, extracts the exposé dra... | I need you to set up a 3-agent pipeline for an investigative journalism project: assignment_desk_chief, investigative_reporter, and defamation_counsel.
- **assignment_desk_chief** orchestrates: each round it selects which story idea to pursue next, calls the investigative_reporter on that idea, extracts the exposé dra... |
pg_044__seed_42 | pg_044 | 42 | [
"assignment_desk_chief",
"investigative_reporter",
"defamation_counsel"
] | [
{
"id": "f1",
"text": "You drive the investigate → draft → review → improve loop: each round, select the next idea, then alternate calls to the investigative_reporter and defamation_counsel agents.\n\nRead the remaining-round count from `counter.txt`. If the file doesn't exist or the count ≤ 0, print \"roun... | f5 | {
"assignment_desk_chief": [
"f1",
"f2"
],
"investigative_reporter": [
"f8",
"f9",
"f3",
"f7"
],
"defamation_counsel": [
"f8",
"f6",
"f4",
"f7"
]
} | I need you to set up a 3-agent pipeline for an investigative journalism project: assignment_desk_chief, investigative_reporter, and defamation_counsel.
- **assignment_desk_chief** orchestrates: each round it selects which story idea to pursue next, calls the investigative_reporter on that idea, extracts the exposé dra... | I need you to set up a 3-agent pipeline for an investigative journalism project: assignment_desk_chief, investigative_reporter, and defamation_counsel.
- **assignment_desk_chief** orchestrates: each round it selects which story idea to pursue next, calls the investigative_reporter on that idea, extracts the exposé dra... |
pg_045__seed_1 | pg_045 | 1 | [
"signal coordinator",
"pharmacovigilance analyst",
"regulatory safety reviewer"
] | [
{
"id": "f1",
"text": "When you finish, briefly report to the signal coordinator what you did.",
"is_distractor": false
},
{
"id": "f2",
"text": "Each round, before calling the pharmacovigilance analyst, read signals.xml and select which candidate signal the pharmacovigilance analyst should ... | f7 | {
"signal coordinator": [
"f5",
"f2"
],
"pharmacovigilance analyst": [
"f6",
"f8",
"f3",
"f1"
],
"regulatory safety reviewer": [
"f6",
"f9",
"f4",
"f1"
]
} | I need you to set up a 3-agent pipeline for pharmaceutical clinical-trial safety signal assessment: signal coordinator, pharmacovigilance analyst, and regulatory safety reviewer.
- **signal coordinator** orchestrates: each round it selects which candidate adverse-event signal to assess next, calls the pharmacovigilanc... | I need you to set up a 3-agent pipeline for pharmaceutical clinical-trial safety signal assessment: signal coordinator, pharmacovigilance analyst, and regulatory safety reviewer.
- **signal coordinator** orchestrates: each round it selects which candidate adverse-event signal to assess next, calls the pharmacovigilanc... |
pg_045__seed_42 | pg_045 | 42 | [
"signal coordinator",
"pharmacovigilance analyst",
"regulatory safety reviewer"
] | [
{
"id": "f1",
"text": "You drive the select → assess → review → improve loop: each round, select the next candidate signal, then alternate calls to the pharmacovigilance analyst and regulatory safety reviewer agents.\n\nRead the remaining-round count from `counter.txt`. If the file doesn't exist or the coun... | f5 | {
"signal coordinator": [
"f1",
"f2"
],
"pharmacovigilance analyst": [
"f8",
"f9",
"f3",
"f7"
],
"regulatory safety reviewer": [
"f8",
"f6",
"f4",
"f7"
]
} | I need you to set up a 3-agent pipeline for pharmaceutical clinical-trial safety signal assessment: signal coordinator, pharmacovigilance analyst, and regulatory safety reviewer.
- **signal coordinator** orchestrates: each round it selects which candidate adverse-event signal to assess next, calls the pharmacovigilanc... | I need you to set up a 3-agent pipeline for pharmaceutical clinical-trial safety signal assessment: signal coordinator, pharmacovigilance analyst, and regulatory safety reviewer.
- **signal coordinator** orchestrates: each round it selects which candidate adverse-event signal to assess next, calls the pharmacovigilanc... |
pg_046__seed_1 | pg_046 | 1 | [
"case manager",
"provenance researcher",
"restitution examiner"
] | [
{
"id": "f1",
"text": "When you finish, briefly report to the case manager what you did.",
"is_distractor": false
},
{
"id": "f2",
"text": "Each round, before calling the provenance researcher, read leads.xml and select which lead the provenance researcher should work on:\n\n1. Every lead ca... | f7 | {
"case manager": [
"f5",
"f2"
],
"provenance researcher": [
"f6",
"f8",
"f3",
"f1"
],
"restitution examiner": [
"f6",
"f9",
"f4",
"f1"
]
} | I need you to set up a 3-agent pipeline for museum antiquities provenance investigation: case manager, provenance researcher, and restitution examiner.
- **case manager** orchestrates: each round it selects which contested artifact or lead to pursue next, calls the provenance researcher on that item, extracts the doss... | I need you to set up a 3-agent pipeline for museum antiquities provenance investigation: case manager, provenance researcher, and restitution examiner.
- **case manager** orchestrates: each round it selects which contested artifact or lead to pursue next, calls the provenance researcher on that item, extracts the doss... |
pg_046__seed_42 | pg_046 | 42 | [
"case manager",
"provenance researcher",
"restitution examiner"
] | [
{
"id": "f1",
"text": "You drive the investigate → document → examine → improve loop: each round, select the next lead, then alternate calls to the provenance researcher and restitution examiner agents.\n\nRead the remaining-round count from `counter.txt`. If the file doesn't exist or the count ≤ 0, print \... | f5 | {
"case manager": [
"f1",
"f2"
],
"provenance researcher": [
"f8",
"f9",
"f3",
"f7"
],
"restitution examiner": [
"f8",
"f6",
"f4",
"f7"
]
} | I need you to set up a 3-agent pipeline for museum antiquities provenance investigation: case manager, provenance researcher, and restitution examiner.
- **case manager** orchestrates: each round it selects which contested artifact or lead to pursue next, calls the provenance researcher on that item, extracts the doss... | I need you to set up a 3-agent pipeline for museum antiquities provenance investigation: case manager, provenance researcher, and restitution examiner.
- **case manager** orchestrates: each round it selects which contested artifact or lead to pursue next, calls the provenance researcher on that item, extracts the doss... |
pg_047__seed_1 | pg_047 | 1 | [
"permitting coordinator",
"impact modeler",
"environmental auditor"
] | [
{
"id": "f1",
"text": "When you finish, briefly report to the permitting coordinator what you did.",
"is_distractor": false
},
{
"id": "f2",
"text": "Each round, before calling the impact modeler, read ideas.xml and select which risk hypothesis the impact modeler should work on:\n\n1. Every ... | f7 | {
"permitting coordinator": [
"f5",
"f2"
],
"impact modeler": [
"f6",
"f8",
"f3",
"f1"
],
"environmental auditor": [
"f6",
"f9",
"f4",
"f1"
]
} | I need you to set up a 3-agent pipeline for environmental impact permitting on a major infrastructure project: permitting coordinator, impact modeler, and environmental auditor.
- **permitting coordinator** orchestrates: each round it selects which risk hypothesis to pursue next, calls the impact modeler on that hypot... | I need you to set up a 3-agent pipeline for environmental impact permitting on a major infrastructure project: permitting coordinator, impact modeler, and environmental auditor.
- **permitting coordinator** orchestrates: each round it selects which risk hypothesis to pursue next, calls the impact modeler on that hypot... |
pg_047__seed_42 | pg_047 | 42 | [
"permitting coordinator",
"impact modeler",
"environmental auditor"
] | [
{
"id": "f1",
"text": "You drive the explore → write → audit → improve loop: each round, select the next risk hypothesis, then alternate calls to the impact modeler and environmental auditor agents.\n\nRead the remaining-round count from `counter.txt`. If the file doesn't exist or the count ≤ 0, print \"rou... | f5 | {
"permitting coordinator": [
"f1",
"f2"
],
"impact modeler": [
"f8",
"f9",
"f3",
"f7"
],
"environmental auditor": [
"f8",
"f6",
"f4",
"f7"
]
} | I need you to set up a 3-agent pipeline for environmental impact permitting on a major infrastructure project: permitting coordinator, impact modeler, and environmental auditor.
- **permitting coordinator** orchestrates: each round it selects which risk hypothesis to pursue next, calls the impact modeler on that hypot... | I need you to set up a 3-agent pipeline for environmental impact permitting on a major infrastructure project: permitting coordinator, impact modeler, and environmental auditor.
- **permitting coordinator** orchestrates: each round it selects which risk hypothesis to pursue next, calls the impact modeler on that hypot... |
pg_048__seed_1 | pg_048 | 1 | [
"audit docket coordinator",
"tax position analyst",
"revenue examiner"
] | [
{
"id": "f1",
"text": "When you finish, briefly report to the audit docket coordinator what you did.",
"is_distractor": false
},
{
"id": "f2",
"text": "Each round, before calling the tax position analyst, read positions.xml and select which position the tax position analyst should work on:\n... | f7 | {
"audit docket coordinator": [
"f5",
"f2"
],
"tax position analyst": [
"f6",
"f8",
"f3",
"f1"
],
"revenue examiner": [
"f6",
"f9",
"f4",
"f1"
]
} | I need you to set up a 3-agent pipeline for corporate tax audit defense: audit docket coordinator, tax position analyst, and revenue examiner.
- **audit docket coordinator** orchestrates: each round it selects which questionable tax position to pursue next, calls the tax position analyst on that position, extracts the... | I need you to set up a 3-agent pipeline for corporate tax audit defense: audit docket coordinator, tax position analyst, and revenue examiner.
- **audit docket coordinator** orchestrates: each round it selects which questionable tax position to pursue next, calls the tax position analyst on that position, extracts the... |
pg_048__seed_42 | pg_048 | 42 | [
"audit docket coordinator",
"tax position analyst",
"revenue examiner"
] | [
{
"id": "f1",
"text": "You drive the select → analyze → examine → improve loop: each round, select the next questionable tax position, then alternate calls to the tax position analyst and revenue examiner agents.\n\nRead the remaining-round count from `counter.txt`. If the file doesn't exist or the count ≤ ... | f5 | {
"audit docket coordinator": [
"f1",
"f2"
],
"tax position analyst": [
"f8",
"f9",
"f3",
"f7"
],
"revenue examiner": [
"f8",
"f6",
"f4",
"f7"
]
} | I need you to set up a 3-agent pipeline for corporate tax audit defense: audit docket coordinator, tax position analyst, and revenue examiner.
- **audit docket coordinator** orchestrates: each round it selects which questionable tax position to pursue next, calls the tax position analyst on that position, extracts the... | I need you to set up a 3-agent pipeline for corporate tax audit defense: audit docket coordinator, tax position analyst, and revenue examiner.
- **audit docket coordinator** orchestrates: each round it selects which questionable tax position to pursue next, calls the tax position analyst on that position, extracts the... |
pg_049__seed_1 | pg_049 | 1 | [
"matter coordinator",
"patent search specialist",
"invalidity reviewer"
] | [
{
"id": "f1",
"text": "When you finish, briefly report to the matter coordinator what you did.",
"is_distractor": false
},
{
"id": "f2",
"text": "Each round, before calling the patent search specialist, read theories.xml and select which theory the patent search specialist should work on:\n\... | f7 | {
"matter coordinator": [
"f5",
"f2"
],
"patent search specialist": [
"f6",
"f8",
"f3",
"f1"
],
"invalidity reviewer": [
"f6",
"f9",
"f4",
"f1"
]
} | I need you to set up a 3-agent pipeline for patent validity challenge preparation: matter coordinator, patent search specialist, and invalidity reviewer.
- **matter coordinator** orchestrates: each round it selects which prior-art theory to pursue next, calls the patent search specialist on that theory, extracts the c... | I need you to set up a 3-agent pipeline for patent validity challenge preparation: matter coordinator, patent search specialist, and invalidity reviewer.
- **matter coordinator** orchestrates: each round it selects which prior-art theory to pursue next, calls the patent search specialist on that theory, extracts the c... |
pg_049__seed_42 | pg_049 | 42 | [
"matter coordinator",
"patent search specialist",
"invalidity reviewer"
] | [
{
"id": "f1",
"text": "You drive the select → search → review → improve loop: each round, select the next prior-art theory, then alternate calls to the patent search specialist and invalidity reviewer agents.\n\nRead the remaining-round count from `counter.txt`. If the file doesn't exist or the count ≤ 0, p... | f5 | {
"matter coordinator": [
"f1",
"f2"
],
"patent search specialist": [
"f8",
"f9",
"f3",
"f7"
],
"invalidity reviewer": [
"f8",
"f6",
"f4",
"f7"
]
} | I need you to set up a 3-agent pipeline for patent validity challenge preparation: matter coordinator, patent search specialist, and invalidity reviewer.
- **matter coordinator** orchestrates: each round it selects which prior-art theory to pursue next, calls the patent search specialist on that theory, extracts the c... | I need you to set up a 3-agent pipeline for patent validity challenge preparation: matter coordinator, patent search specialist, and invalidity reviewer.
- **matter coordinator** orchestrates: each round it selects which prior-art theory to pursue next, calls the patent search specialist on that theory, extracts the c... |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.