examples.json

[
 {
  "id": "cell-001086",
  "difficulty": "t2_linear",
  "condition": "gap",
  "request": "Book trip SF to Berlin, May 10-17: flight, hotel, confirm visa",
  "steps": [
   {
    "id": 1,
    "task": "Search for round-trip flights from San Francisco (SFO) to Berlin (BER) departing May 10, returning May 17; return top 3 options with times, durations, and prices.",
    "assignee": "tool:flight_search",
    "depends_on": [],
    "cost": 0.0
   },
   {
    "id": 2,
    "task": "Search for hotels in Berlin for May 10-17 with filters for 3★+, near central transit, and under $150/night; return top 5 options with addresses and nightly rates.",
    "assignee": "tool:hotel_search",
    "depends_on": [],
    "cost": 0.0
   },
   {
    "id": 3,
    "task": "Draft a concise trip checklist and next steps (visa confirmation, booking timeline, packing tips) based on the flight and hotel options, and produce a short email/message to send to the traveler.",
    "assignee": "small_model",
    "depends_on": [
     1,
     2
    ],
    "cost": 0.1703
   }
  ],
  "escalation_reason": null,
  "router_cost": 0.0358,
  "dispatch_total": 0.2061,
  "baseline_total": 7.8382,
  "savings_pct": 97.4
 },
 {
  "id": "cell-002029",
  "difficulty": "t2_linear",
  "condition": "full",
  "request": "I'm a grad student writing a lit review on graph neural networks for molecular property prediction. Can you find the three most relevant recent arXiv papers (last 3 years), fetch the full text of the top result, and give me a concise 300–400 word summary of its dataset, model architecture, and training procedure with a link to the paper? (Ignore this — I have a dentist appointment at 4pm.)",
  "steps": [
   {
    "id": 1,
    "task": "Search arXiv for recent (past 3 years) papers on graph neural networks for molecular property prediction and return the top 3 relevant results with titles, authors, abstracts, and URLs.",
    "assignee": "tool:arxiv_search",
    "depends_on": [],
    "cost": 0.0
   },
   {
    "id": 2,
    "task": "Fetch the full text (PDF or HTML) of the top-ranked paper from step 1 and extract the abstract, introduction, methods, and results sections.",
    "assignee": "tool:web_fetch",
    "depends_on": [
     1
    ],
    "cost": 0.0
   },
   {
    "id": 3,
    "task": "Produce a concise 300–400 word summary of the fetched paper focusing on dataset description, model architecture, and training procedure; include the paper link and a one-sentence recommendation for a grad student.",
    "assignee": "small_model",
    "depends_on": [
     2
    ],
    "cost": 0.1827
   }
  ],
  "escalation_reason": null,
  "router_cost": 0.0452,
  "dispatch_total": 0.2279,
  "baseline_total": 8.0962,
  "savings_pct": 97.2
 },
 {
  "id": "cell-000605",
  "difficulty": "t3_dag",
  "condition": "full",
  "request": "Email school and daycare: pickup change today",
  "steps": [
   {
    "id": 1,
    "task": "Draft a concise email to the child's school and daycare that: (a) states the child's name and class/group, (b) requests confirmation of the pickup change (new time and any contact phone), and (c) includes a polite sign-off and contact details.",
    "assignee": "small_model",
    "depends_on": [],
    "cost": 0.1697
   },
   {
    "id": 2,
    "task": "Look up contact details (email addresses and phone numbers) for the school and daycare contacts to use as recipients.",
    "assignee": "tool:contacts_lookup",
    "depends_on": [
     1
    ],
    "cost": 0.0
   },
   {
    "id": 3,
    "task": "Send the drafted email to the school and daycare contacts found in step 2, CC the parent's email, and include the parent's phone number in the message.",
    "assignee": "tool:email_send",
    "depends_on": [
     1,
     2
    ],
    "cost": 0.0
   }
  ],
  "escalation_reason": null,
  "router_cost": 0.0353,
  "dispatch_total": 0.205,
  "baseline_total": 7.8502,
  "savings_pct": 97.4
 },
 {
  "id": "cell-002405",
  "difficulty": "t2_linear",
  "condition": "gap",
  "request": "Compare prices for 'Acme UltraWidget' across vendors and summarize.",
  "steps": [
   {
    "id": 1,
    "task": "Search the product catalog for 'Acme UltraWidget' and return product id(s) and canonical product name.",
    "assignee": "tool:product_search",
    "depends_on": [],
    "cost": 0.0
   },
   {
    "id": 2,
    "task": "Search the codebase for vendor price records or price-fetching functions referencing the product id(s) from step 1.",
    "assignee": "tool:code_search",
    "depends_on": [
     1
    ],
    "cost": 0.0
   },
   {
    "id": 3,
    "task": "Check current stock levels for the product id(s) from step 1.",
    "assignee": "tool:inventory_check",
    "depends_on": [
     1
    ],
    "cost": 0.0
   },
   {
    "id": 4,
    "task": "Synthesize a concise price comparison and stock summary: list vendor prices, indicate lowest price, stock status, and a short recommendation.",
    "assignee": "small_model",
    "depends_on": [
     2,
     3
    ],
    "cost": 0.1705
   }
  ],
  "escalation_reason": null,
  "router_cost": 0.0355,
  "dispatch_total": 0.206,
  "baseline_total": 8.46,
  "savings_pct": 97.6
 },
 {
  "id": "cell-000715",
  "difficulty": "t3_dag",
  "condition": "full",
  "request": "Urgent: restock SKU-123 — find cheapest supplier and draft order email.",
  "steps": [
   {
    "id": 1,
    "task": "Search the product catalog for SKU-123 to retrieve product name, SKU, and any vendor identifiers.",
    "assignee": "tool:product_search",
    "depends_on": [],
    "cost": 0.0
   },
   {
    "id": 2,
    "task": "Compare prices for the product (SKU-123) across available vendors to identify the cheapest supplier and their price.",
    "assignee": "tool:price_compare",
    "depends_on": [
     1
    ],
    "cost": 0.0
   },
   {
    "id": 3,
    "task": "Check inventory levels for SKU-123 at the cheapest supplier to confirm stock and required quantity.",
    "assignee": "tool:inventory_check",
    "depends_on": [
     2
    ],
    "cost": 0.0
   },
   {
    "id": 4,
    "task": "Draft a concise order email to the cheapest supplier including product name, SKU, quantity, unit price, and requested delivery date.",
    "assignee": "small_model",
    "depends_on": [
     3
    ],
    "cost": 0.1707
   }
  ],
  "escalation_reason": null,
  "router_cost": 0.0373,
  "dispatch_total": 0.208,
  "baseline_total": 8.5215,
  "savings_pct": 97.6
 },
 {
  "id": "cell-000875",
  "difficulty": "t4_hard",
  "condition": "distractors",
  "request": "Make two-week study plan and schedule group meeting.",
  "steps": [
   {
    "id": 1,
    "task": "Search the user's calendar for existing events and busy times over the next two weeks to identify free slots.",
    "assignee": "tool:calendar_search",
    "depends_on": [],
    "cost": 0.0
   },
   {
    "id": 2,
    "task": "Search the user's saved notes for course syllabi, exam dates, and any stated study priorities to inform scheduling.",
    "assignee": "tool:note_search",
    "depends_on": [],
    "cost": 0.0
   },
   {
    "id": 3,
    "task": "Summarize the user's recent expenses to identify any recurring commitments that could affect study time.",
    "assignee": "tool:expense_report",
    "depends_on": [],
    "cost": 0.0
   },
   {
    "id": 4,
    "task": "Create a two-week study schedule that balances course priorities, exam dates, and the user's calendar availability; produce a prioritized list of study sessions and one recommended group-meeting time.",
    "assignee": "escalate:frontier",
    "depends_on": [
     1,
     2,
     3
    ],
    "cost": 4.149
   },
   {
    "id": 5,
    "task": "Create a calendar event for the recommended group meeting time with the group as attendees.",
    "assignee": "tool:calendar_create_event",
    "depends_on": [
     4
    ],
    "cost": 0.0
   }
  ],
  "escalation_reason": "Designing an optimal two-week study schedule requires multi-constraint tradeoffs (exam dates, course priorities, calendar conflicts, and recurring expenses) and a prioritized plan, which needs frontier-level reasoning.",
  "router_cost": 0.0521,
  "dispatch_total": 4.2011,
  "baseline_total": 9.336,
  "savings_pct": 55.0
 },
 {
  "id": "cell-001200",
  "difficulty": "t3_dag",
  "condition": "full",
  "request": "Create PR for test fix and README update.",
  "steps": [
   {
    "id": 1,
    "task": "Show the current git diff so we can see what changes will go into the PR.",
    "assignee": "tool:git_diff",
    "depends_on": [],
    "cost": 0.0
   },
   {
    "id": 2,
    "task": "Run the project's test suite to confirm the test fix and capture any failing tests.",
    "assignee": "tool:run_tests",
    "depends_on": [],
    "cost": 0.0
   },
   {
    "id": 3,
    "task": "Read the current README file to determine what updates are needed.",
    "assignee": "tool:file_read",
    "depends_on": [],
    "cost": 0.0
   },
   {
    "id": 4,
    "task": "Draft a concise PR title and a detailed PR description that: summarizes the test fix, lists any failing tests (from step 2), summarizes the README updates (from step 3), and includes a checklist for reviewers.",
    "assignee": "small_model",
    "depends_on": [
     1,
     2,
     3
    ],
    "cost": 0.1695
   },
   {
    "id": 5,
    "task": "Open a pull request using the drafted title and description.",
    "assignee": "tool:create_pr",
    "depends_on": [
     4
    ],
    "cost": 0.0
   }
  ],
  "escalation_reason": null,
  "router_cost": 0.0363,
  "dispatch_total": 0.2058,
  "baseline_total": 9.03,
  "savings_pct": 97.7
 },
 {
  "id": "cell-001475",
  "difficulty": "t4_hard",
  "condition": "full",
  "request": "Reschedule conference sessions; confirm and notify speakers.",
  "steps": [
   {
    "id": 1,
    "task": "Search saved notes for the conference schedule, session details, and speaker contact info.",
    "assignee": "tool:note_search",
    "depends_on": [],
    "cost": 0.0
   },
   {
    "id": 2,
    "task": "Look up contact details (email) for the two session speakers and the organizer.",
    "assignee": "tool:contacts_lookup",
    "depends_on": [
     1
    ],
    "cost": 0.0
   },
   {
    "id": 3,
    "task": "Draft a concise email to speakers and organizer proposing new session times and asking for confirmation; include proposed times, duration, and any logistical notes.",
    "assignee": "small_model",
    "depends_on": [
     1,
     2
    ],
    "cost": 0.1702
   },
   {
    "id": 4,
    "task": "Send the drafted email to the speakers and organizer.",
    "assignee": "tool:email_send",
    "depends_on": [
     3,
     2
    ],
    "cost": 0.0
   },
   {
    "id": 5,
    "task": "Post a Slack message to the organizer channel summarizing the reschedule and confirming the new times.",
    "assignee": "tool:slack_post",
    "depends_on": [
     4
    ],
    "cost": 0.0
   }
  ],
  "escalation_reason": null,
  "router_cost": 0.039,
  "dispatch_total": 0.2093,
  "baseline_total": 9.105,
  "savings_pct": 97.7
 },
 {
  "id": "cell-001988",
  "difficulty": "t3_dag",
  "condition": "distractors",
  "request": "Hi — I uploaded that design file (the 3-panel brochure) a few minutes ago. Can you: 1) pull the key finished dimensions from that file and convert them to inches, 2) find the client contact from my last order for this project, and 3) send a short update email to the client saying the proofs are ready, include the converted dimensions, and mention the order number/status from the last order. Keep the email concise and professional. Thanks!",
  "steps": [
   {
    "id": 1,
    "task": "Read the contents of the uploaded brochure file (the 3-panel design) to extract the key finished dimensions (width, height, panel sizes).",
    "assignee": "tool:file_read",
    "depends_on": [],
    "cost": 0.0
   },
   {
    "id": 2,
    "task": "Convert the extracted finished dimensions from the file into inches (use the appropriate conversion factors).",
    "assignee": "tool:unit_convert",
    "depends_on": [
     1
    ],
    "cost": 0.0
   },
   {
    "id": 3,
    "task": "Look up the client contact details (email) for the customer associated with the most recent order for this project.",
    "assignee": "tool:order_lookup",
    "depends_on": [],
    "cost": 0.0
   },
   {
    "id": 4,
    "task": "Draft a concise, professional update email to the client that: (a) says proofs are ready, (b) lists the converted dimensions in inches, and (c) includes the order number and status from the last order.",
    "assignee": "small_model",
    "depends_on": [
     2,
     3
    ],
    "cost": 0.1846
   },
   {
    "id": 5,
    "task": "Send the drafted email to the client's email address found in step 3, with the converted dimensions and order status included.",
    "assignee": "tool:email_send",
    "depends_on": [
     4,
     3
    ],
    "cost": 0.0
   }
  ],
  "escalation_reason": null,
  "router_cost": 0.0585,
  "dispatch_total": 0.243,
  "baseline_total": 9.6015,
  "savings_pct": 97.5
 },
 {
  "id": "cell-001526",
  "difficulty": "t4_hard",
  "condition": "full",
  "request": "Autograder failing—fix without delaying grading deadlines.",
  "steps": [
   {
    "id": 1,
    "task": "Read the autograder error logs and the student's code to extract the failing test names, error messages, and relevant code snippets.",
    "assignee": "tool:file_read",
    "depends_on": [],
    "cost": 0.0
   },
   {
    "id": 2,
    "task": "Search the codebase for the failing test names and related symbols to locate the test file(s) and implementation details.",
    "assignee": "tool:code_search",
    "depends_on": [
     1
    ],
    "cost": 0.0
   },
   {
    "id": 3,
    "task": "Run the linter on the repository to surface style/type issues that could affect grading.",
    "assignee": "tool:lint_code",
    "depends_on": [],
    "cost": 0.0
   },
   {
    "id": 4,
    "task": "Run the test suite to reproduce the failure and capture the full test output and stack trace.",
    "assignee": "tool:run_tests",
    "depends_on": [],
    "cost": 0.0
   },
   {
    "id": 5,
    "task": "Produce a minimal, safe fix (code change + test case update) that resolves the failing tests and preserves correct behavior; include a short rationale and test plan.",
    "assignee": "escalate:frontier",
    "depends_on": [
     1,
     2,
     3,
     4
    ],
    "cost": 4.1535
   },
   {
    "id": 6,
    "task": "Open a pull request with the proposed fix, a clear title, and the fix description produced in step 5.",
    "assignee": "tool:create_pr",
    "depends_on": [
     5
    ],
    "cost": 0.0
   }
  ],
  "escalation_reason": "Designing a minimal, safe fix requires resolving tradeoffs between correctness, backward compatibility, and grading constraints, which needs deep, multi-constraint reasoning.",
  "router_cost": 0.0468,
  "dispatch_total": 4.2003,
  "baseline_total": 9.6427,
  "savings_pct": 56.4
 }
]