# s1mb-generalization-diverse-choice

English evaluation cases authored using GPT-6-Astra. This review contains answers; do not send it as model input.

## synthetic-choice-001

### Instruction

```json
"Select the communication the user wants to stop."
```

### State

```json
{
  "message": "The account still works. Stop the weekly promotional email, but keep security alerts and invoices coming."
}
```

### Criteria

- `option_1` (value: None): "All communications"
- `option_2` (value: None): "Security alerts"
- `option_3` (value: None): "Invoices"
- `option_4` (value: None): "Promotional email"
- `option_5` (value: None): "Account access"
- `option_6` (value: None): "Shipment updates"
- `option_7` (value: None): "Password resets"
- `option_8` (value: None): "Direct messages"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 1.0, 'option_5': 0.0, 'option_6': 0.0, 'option_7': 0.0, 'option_8': 0.0}

The user explicitly asks to stop weekly promotional email.

## synthetic-choice-002

### Instruction

```json
"Which control would carry out this exact intention?"
```

### State

```json
{
  "available_actions": "The actions have exactly their ordinary meanings in this editor.",
  "intention": "Keep the document, but stop others seeing it"
}
```

### Criteria

- `option_1` (value: None): "Add comment"
- `option_2` (value: None): "Unpublish document"
- `option_3` (value: None): "Rename document"
- `option_4` (value: None): "Export document"
- `option_5` (value: None): "Duplicate document"
- `option_6` (value: None): "Translate document"
- `option_7` (value: None): "Print document"
- `option_8` (value: None): "Delete document"

Target: {'option_1': 0.0, 'option_2': 1.0, 'option_3': 0.0, 'option_4': 0.0, 'option_5': 0.0, 'option_6': 0.0, 'option_7': 0.0, 'option_8': 0.0}

Unpublishing removes public visibility while preserving the document.

## synthetic-choice-003

### Instruction

```json
"Choose the intended sense of 'pitch'."
```

### State

```json
"The singer's pitch was slightly higher than the note played on the piano."
```

### Criteria

- `option_1` (value: None): "Sports field"
- `option_2` (value: None): "Sales presentation"
- `option_3` (value: None): "Tent installation"
- `option_4` (value: None): "Dark resin"
- `option_5` (value: None): "Aircraft nose angle"
- `option_6` (value: None): "Roof slope"
- `option_7` (value: None): "Musical frequency level"
- `option_8` (value: None): "Throwing a ball"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 0.0, 'option_5': 0.0, 'option_6': 0.0, 'option_7': 1.0, 'option_8': 0.0}

The comparison is between musical notes.

## synthetic-choice-004

### Instruction

```json
"Select the exact pair of communications the user explicitly wants to retain."
```

### State

```json
{
  "message": "The account still works. Stop the weekly promotional email, but keep security alerts and invoices coming."
}
```

### Criteria

- `option_1` (value: None): "Security alerts and invoices"
- `option_2` (value: None): "Promotions and security alerts"
- `option_3` (value: None): "No communications"
- `option_4` (value: None): "Invoices and shipment updates"
- `option_5` (value: None): "Promotions and invoices"
- `option_6` (value: None): "Security alerts and direct messages"
- `option_7` (value: None): "Direct messages and promotions"
- `option_8` (value: None): "Password resets and shipment updates"

Target: {'option_1': 1.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 0.0, 'option_5': 0.0, 'option_6': 0.0, 'option_7': 0.0, 'option_8': 0.0}

Security alerts and invoices are explicitly retained.

## synthetic-choice-005

### Instruction

```json
"What's the main communicative purpose?"
```

### State

```json
"I'm sorry I interrupted you. Please finish what you were saying."
```

### Criteria

- `option_1` (value: None): "Boast"
- `option_2` (value: None): "Ask for directions"
- `option_3` (value: None): "Threaten"
- `option_4` (value: None): "Reject an invitation"
- `option_5` (value: None): "Apologize and return the floor"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 0.0, 'option_5': 1.0}

The speaker apologizes and invites the other person to continue.

## synthetic-choice-006

### Instruction

```json
"Choose the status supported by the final log line."
```

### State

```json
[
  "queued",
  "started",
  "validation passed",
  "write failed: disk full"
]
```

### Criteria

- `option_1` (value: None): "Validation rejected input"
- `option_2` (value: None): "Running normally"
- `option_3` (value: None): "Queued only"
- `option_4` (value: None): "Completed successfully"
- `option_5` (value: None): "Cancelled by user"
- `option_6` (value: None): "Failed while writing output"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 0.0, 'option_5': 0.0, 'option_6': 1.0}

The final event is a disk-full write failure.

## synthetic-choice-007

### Instruction

```json
"Pick the thing that meets both needs."
```

### State

```json
{
  "drinks": "Hot caffeine-free herbal tea; iced caffeine-free herbal tea; hot caffeinated coffee",
  "needs": "hot and caffeine-free"
}
```

### Criteria

- `option_1` (value: None): "Iced herbal tea"
- `option_2` (value: None): "Hot herbal tea"
- `option_3` (value: None): "Hot coffee"

Target: {'option_1': 0.0, 'option_2': 1.0, 'option_3': 0.0}

Only hot herbal tea meets both conditions as described.

## synthetic-choice-008

### Instruction

```json
"Classify the relationship to the claim. Do not fill in missing facts."
```

### State

```json
{
  "claim": "The center opens on Sunday",
  "passage": "The center is closed every Sunday."
}
```

### Criteria

- `option_1` (value: None): "Contradicts"
- `option_2` (value: None): "Supports"
- `option_3` (value: None): "Does not resolve"

Target: {'option_1': 1.0, 'option_2': 0.0, 'option_3': 0.0}

Sunday closure directly contradicts Sunday opening.

## synthetic-choice-009

### Instruction

```json
"Extract the final owner after all revisions."
```

### State

```json
[
  {
    "event": "assigned",
    "owner": "Ada"
  },
  {
    "event": "reassigned",
    "owner": "Ben"
  },
  {
    "event": "confirmed",
    "owner": "Ben"
  }
]
```

### Criteria

- `option_1` (value: None): "Ben"
- `option_2` (value: None): "No owner"
- `option_3` (value: None): "Ada"
- `option_4` (value: None): "Both Ada and Ben"

Target: {'option_1': 1.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 0.0}

The last assignment and confirmation name Ben.

## synthetic-choice-010

### Instruction

```json
"Identify the team explicitly responsible for the next action."
```

### State

```json
"Engineering has finished the fix. Support informed customers. Legal must now approve the revised wording before Marketing publishes it."
```

### Criteria

- `option_1` (value: None): "Legal"
- `option_2` (value: None): "Support"
- `option_3` (value: None): "Human Resources"
- `option_4` (value: None): "Marketing"
- `option_5` (value: None): "Engineering"
- `option_6` (value: None): "Finance"
- `option_7` (value: None): "Facilities"

Target: {'option_1': 1.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 0.0, 'option_5': 0.0, 'option_6': 0.0, 'option_7': 0.0}

Legal approval is the next required action before publication.

## synthetic-choice-011

### Instruction

```json
"Choose the one object that survives both filters."
```

### State

```json
{
  "filters": [
    "red",
    "square"
  ],
  "objects": {
    "A": "red circle",
    "B": "blue square",
    "C": "green triangle",
    "D": "red square",
    "E": "blue circle",
    "F": "green square"
  }
}
```

### Criteria

- `option_1` (value: None): "E"
- `option_2` (value: None): "D"
- `option_3` (value: None): "C"
- `option_4` (value: None): "F"
- `option_5` (value: None): "A"
- `option_6` (value: None): "B"

Target: {'option_1': 0.0, 'option_2': 1.0, 'option_3': 0.0, 'option_4': 0.0, 'option_5': 0.0, 'option_6': 0.0}

D is the only red square.

## synthetic-choice-012

### Instruction

```json
"Select the field containing the contact address, not the display name."
```

### State

```json
{
  "record": {
    "city": "Northport",
    "email": "riley@example.invalid",
    "id": "u17",
    "name": "Riley",
    "role": "editor"
  }
}
```

### Criteria

- `option_1` (value: None): "role"
- `option_2` (value: None): "city"
- `option_3` (value: None): "id"
- `option_4` (value: None): "name"
- `option_5` (value: None): "email"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 0.0, 'option_5': 1.0}

Email stores the contact address.

## synthetic-choice-013

### Instruction

```json
"Classify the field by its declared meaning, not by its appearance."
```

### State

```json
{
  "meaning": "customer identifier; leading zeros are significant",
  "value": "00127"
}
```

### Criteria

- `option_1` (value: None): "Identifier"
- `option_2` (value: None): "Quantity"

Target: {'option_1': 1.0, 'option_2': 0.0}

The value identifies a customer and must preserve leading zeros.

## synthetic-choice-014

### Instruction

```json
"Which item is covered by the stated exemption?"
```

### State

```json
{
  "items": [
    "sandwich",
    "open juice cup",
    "sealed water bottle",
    "apple",
    "coffee mug"
  ],
  "rule": "No food in the studio except sealed water bottles"
}
```

### Criteria

- `option_1` (value: None): "Sealed water bottle"
- `option_2` (value: None): "Coffee mug"
- `option_3` (value: None): "Open juice cup"
- `option_4` (value: None): "Sandwich"
- `option_5` (value: None): "Apple"

Target: {'option_1': 1.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 0.0, 'option_5': 0.0}

Only the sealed water bottle matches the exemption.

## synthetic-choice-015

### Instruction

```json
"Where is this package now? Read the whole event chain."
```

### State

```json
[
  {
    "event": "warehouse received"
  },
  {
    "event": "loaded onto van"
  },
  {
    "event": "delivery attempted, recipient absent"
  },
  {
    "event": "returned to local depot"
  }
]
```

### Criteria

- `option_1` (value: None): "Delivered to recipient"
- `option_2` (value: None): "At warehouse before dispatch"
- `option_3` (value: None): "At origin factory"
- `option_4` (value: None): "Currently on delivery van"
- `option_5` (value: None): "Lost with no location"
- `option_6` (value: None): "At local depot after failed delivery"
- `option_7` (value: None): "Returned to sender"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 0.0, 'option_5': 0.0, 'option_6': 1.0, 'option_7': 0.0}

The final event places it at the local depot, not with the sender.

## synthetic-choice-016

### Instruction

```json
"Which entry is malformed under the given format?"
```

### State

```json
{
  "entries": [
    "AB12",
    "CD34",
    "EF56",
    "G789",
    "HI90",
    "JK01"
  ],
  "format": "Exactly two ASCII letters followed by exactly two digits"
}
```

### Criteria

- `option_1` (value: None): "G789"
- `option_2` (value: None): "HI90"
- `option_3` (value: None): "JK01"
- `option_4` (value: None): "AB12"
- `option_5` (value: None): "CD34"
- `option_6` (value: None): "EF56"

Target: {'option_1': 1.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 0.0, 'option_5': 0.0, 'option_6': 0.0}

G789 has one letter and three digits.

## synthetic-choice-017

### Instruction

```json
"Which task must happen immediately before sealing under this procedure?"
```

### State

```json
[
  "print label",
  "pack item",
  "insert receipt",
  "seal box"
]
```

### Criteria

- `option_1` (value: None): "Insert receipt"
- `option_2` (value: None): "Print label"
- `option_3` (value: None): "Pack item"
- `option_4` (value: None): "Deliver parcel"

Target: {'option_1': 1.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 0.0}

Inserting the receipt directly precedes sealing.

## synthetic-choice-018

### Instruction

```json
{
  "credit": "Completed work only",
  "task": "Select the appropriate status"
}
```

### State

```json
{
  "record": "The report is written and reviewed, but not yet signed. Distribution follows signing."
}
```

### Criteria

- `option_1` (value: None): "Awaiting signature"
- `option_2` (value: None): "Distributed"
- `option_3` (value: None): "Awaiting review"
- `option_4` (value: None): "Not started"
- `option_5` (value: None): "Drafting"

Target: {'option_1': 1.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 0.0, 'option_5': 0.0}

Writing and review are complete; signature is pending.

## synthetic-choice-019

### Instruction

```json
"Choose the most precise accessible label for this icon's action."
```

### State

```json
{
  "action": "Search within the current document only",
  "icon": "magnifying glass"
}
```

### Criteria

- `option_1` (value: None): "Search the web"
- `option_2` (value: None): "Search this document"
- `option_3` (value: None): "Zoom in"
- `option_4` (value: None): "Button"
- `option_5` (value: None): "Magnifier"

Target: {'option_1': 0.0, 'option_2': 1.0, 'option_3': 0.0, 'option_4': 0.0, 'option_5': 0.0}

The label states the actual search scope and action.

## synthetic-choice-020

### Instruction

```json
"Choose the locker matching all three clues."
```

### State

```json
{
  "lockers": {
    "A": "upper red unlocked",
    "B": "lower blue unlocked",
    "C": "upper blue locked",
    "D": "lower red locked",
    "E": "upper blue unlocked",
    "F": "upper green unlocked",
    "G": "lower green unlocked",
    "H": "lower blue locked"
  },
  "need": "upper row, blue, unlocked"
}
```

### Criteria

- `option_1` (value: None): "Locker G"
- `option_2` (value: None): "Locker E"
- `option_3` (value: None): "Locker D"
- `option_4` (value: None): "Locker A"
- `option_5` (value: None): "Locker H"
- `option_6` (value: None): "Locker F"
- `option_7` (value: None): "Locker B"
- `option_8` (value: None): "Locker C"

Target: {'option_1': 0.0, 'option_2': 1.0, 'option_3': 0.0, 'option_4': 0.0, 'option_5': 0.0, 'option_6': 0.0, 'option_7': 0.0, 'option_8': 0.0}

Only E is upper, blue and unlocked.

## synthetic-choice-021

### Instruction

```json
"What's the most accurate description of the reply?"
```

### State

```json
{
  "reply": "Only if my train gets in before six.",
  "request": "Will you join on Friday?"
}
```

### Criteria

- `option_1` (value: None): "Unconditional acceptance"
- `option_2` (value: None): "Conditional acceptance"
- `option_3` (value: None): "Unconditional refusal"
- `option_4` (value: None): "Request for a different venue"

Target: {'option_1': 0.0, 'option_2': 1.0, 'option_3': 0.0, 'option_4': 0.0}

Joining depends on train arrival.

## synthetic-choice-022

### Instruction

```json
"Pick the document that actually answers both requested parts."
```

### State

```json
{
  "docs": {
    "A": "Room 3",
    "B": "Starts at 10:00",
    "C": "Room 3, starts at 10:00",
    "D": "Registration closes Friday",
    "E": "Lunch at noon",
    "F": "Bring a notebook",
    "G": "Workshop cancelled last year"
  },
  "query": "Where is the workshop and what time does it start?"
}
```

### Criteria

- `option_1` (value: None): "Document B"
- `option_2` (value: None): "Document G"
- `option_3` (value: None): "Document D"
- `option_4` (value: None): "Document F"
- `option_5` (value: None): "Document C"
- `option_6` (value: None): "Document A"
- `option_7` (value: None): "Document E"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 0.0, 'option_5': 1.0, 'option_6': 0.0, 'option_7': 0.0}

C supplies both location and start time.

## synthetic-choice-023

### Instruction

```json
"Which intent matches the query?"
```

### State

```json
"How can I end my membership? I'm not trying to change my password."
```

### Criteria

- `option_1` (value: None): "Reset password"
- `option_2` (value: None): "Cancel membership"

Target: {'option_1': 0.0, 'option_2': 1.0}

Cancellation is requested and password reset explicitly excluded.

## synthetic-choice-024

### Instruction

```json
"Choose the lost item whose description fits every clue."
```

### State

```json
{
  "clues": "A small black umbrella with a wooden handle",
  "items": {
    "A": "large black umbrella, wooden handle",
    "B": "small blue umbrella, wooden handle",
    "C": "small black umbrella, metal handle",
    "D": "small black umbrella, wooden handle",
    "E": "small black bag, wooden handle",
    "F": "large blue umbrella, metal handle"
  }
}
```

### Criteria

- `option_1` (value: None): "Item F"
- `option_2` (value: None): "Item C"
- `option_3` (value: None): "Item E"
- `option_4` (value: None): "Item A"
- `option_5` (value: None): "Item D"
- `option_6` (value: None): "Item B"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 0.0, 'option_5': 1.0, 'option_6': 0.0}

Only D matches size, color, item type and handle material.

## synthetic-choice-025

### Instruction

```json
"Are the intervals disjoint, touching, or overlapping? Endpoints alone do not count as overlap."
```

### State

```json
{
  "first": [
    1,
    3
  ],
  "second": [
    3,
    5
  ]
}
```

### Criteria

- `option_1` (value: None): "Touching at an endpoint"
- `option_2` (value: None): "Overlapping over a positive length"
- `option_3` (value: None): "Disjoint with a gap"

Target: {'option_1': 1.0, 'option_2': 0.0, 'option_3': 0.0}

The intervals share only endpoint 3.

## synthetic-choice-026

### Instruction

```json
"Select the operation that matches the requested transformation."
```

### State

```json
{
  "after": [
    "b",
    "a"
  ],
  "before": [
    "b",
    "a",
    "b"
  ],
  "requirement": "Keep first occurrences in original order"
}
```

### Criteria

- `option_1` (value: None): "Stable deduplication"
- `option_2` (value: None): "Sort descending"
- `option_3` (value: None): "Count elements"
- `option_4` (value: None): "Concatenate into a string"
- `option_5` (value: None): "Reverse"
- `option_6` (value: None): "Sort ascending"
- `option_7` (value: None): "Replace letters with digits"

Target: {'option_1': 1.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 0.0, 'option_5': 0.0, 'option_6': 0.0, 'option_7': 0.0}

The transformation removes repeated b while preserving first-occurrence order.

## synthetic-choice-027

### Instruction

```json
"Name the operation represented by the code."
```

### State

```json
"unique_items = list(dict.fromkeys(items))"
```

### Criteria

- `option_1` (value: None): "Count missing values"
- `option_2` (value: None): "Remove duplicates while preserving first occurrence order"
- `option_3` (value: None): "Sort descending"
- `option_4` (value: None): "Reverse the list"
- `option_5` (value: None): "Multiply numeric values"

Target: {'option_1': 0.0, 'option_2': 1.0, 'option_3': 0.0, 'option_4': 0.0, 'option_5': 0.0}

Dictionary keys retain first insertion order and eliminate duplicate keys.

## synthetic-choice-028

### Instruction

```json
"Choose the heading that describes what this paragraph actually does."
```

### State

```json
"First unplug the lamp. Next remove the cover. Finally replace the bulb and refit the cover."
```

### Criteria

- `option_1` (value: None): "Procedure"
- `option_2` (value: None): "Historical background"

Target: {'option_1': 1.0, 'option_2': 0.0}

The paragraph gives ordered actions.

## synthetic-choice-029

### Instruction

```json
"Who knows the hiding place according to the story? Select the exact knowledge set."
```

### State

```json
"Ava hid the key while Bo watched. Cy and Dee were outside and nobody told them. No one else was present."
```

### Criteria

- `option_1` (value: None): "Ava and Cy"
- `option_2` (value: None): "Bo and Dee"
- `option_3` (value: None): "Ava and Bo"
- `option_4` (value: None): "Ava only"
- `option_5` (value: None): "Bo only"
- `option_6` (value: None): "Nobody"
- `option_7` (value: None): "All four"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 1.0, 'option_4': 0.0, 'option_5': 0.0, 'option_6': 0.0, 'option_7': 0.0}

Ava hid it and Bo witnessed it; the others received no information.

## synthetic-choice-030

### Instruction

```json
"What's the speaker's stance toward the plan?"
```

### State

```json
"I'm not for it or against it yet. I need to see the costs first."
```

### Criteria

- `option_1` (value: None): "Opposed"
- `option_2` (value: None): "Undecided"
- `option_3` (value: None): "Supportive"

Target: {'option_1': 0.0, 'option_2': 1.0, 'option_3': 0.0}

The speaker explicitly withholds a position.

## synthetic-choice-031

### Instruction

```json
"Pick the customer's primary requested action."
```

### State

```json
"The parcel arrived fine, but the shoes are too small. Could I get the next size up instead?"
```

### Criteria

- `option_1` (value: None): "Replacement in another size"
- `option_2` (value: None): "Delivery tracking"
- `option_3` (value: None): "Payment receipt"

Target: {'option_1': 1.0, 'option_2': 0.0, 'option_3': 0.0}

The customer requests a larger size.

## synthetic-choice-032

### Instruction

```json
"Which departure is the earliest one I can catch? Walking finishes at 09:12; boarding is allowed until departure."
```

### State

```json
{
  "departures": [
    "09:00",
    "09:10",
    "09:15",
    "09:30",
    "10:00"
  ]
}
```

### Criteria

- `option_1` (value: None): "09:00"
- `option_2` (value: None): "09:30"
- `option_3` (value: None): "09:10"
- `option_4` (value: None): "09:15"
- `option_5` (value: None): "10:00"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 1.0, 'option_5': 0.0}

09:15 is the first departure after walking finishes.

## synthetic-choice-033

### Instruction

```json
"Which statement describes an observation rather than a proposed explanation?"
```

### State

```json
"A team recorded plant heights and discussed why they differed."
```

### Criteria

- `option_1` (value: None): "The soil could explain the difference"
- `option_2` (value: None): "Plant A measured 18 cm on day ten"
- `option_3` (value: None): "Extra light might cause faster growth"
- `option_4` (value: None): "Perhaps roots were damaged"
- `option_5` (value: None): "A temperature effect seems possible"

Target: {'option_1': 0.0, 'option_2': 1.0, 'option_3': 0.0, 'option_4': 0.0, 'option_5': 0.0}

A measured height is an observation, while the others propose causes.

## synthetic-choice-034

### Instruction

```json
"What kind of revision did the editor make?"
```

### State

```json
{
  "after": "The results suggest the method worked in this small sample.",
  "before": "The results prove the method always works."
}
```

### Criteria

- `option_1` (value: None): "Changed only punctuation"
- `option_2` (value: None): "Increased certainty and scope"
- `option_3` (value: None): "Removed all reference to results"
- `option_4` (value: None): "Reduced certainty and scope"
- `option_5` (value: None): "Added an unrelated topic"
- `option_6` (value: None): "Changed a past event into a future plan"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 1.0, 'option_5': 0.0, 'option_6': 0.0}

Prove/always becomes suggest/in this small sample.

## synthetic-choice-035

### Instruction

```json
"Select the applicable route under this fictional rule."
```

### State

```json
{
  "rule": "Members use the east entrance; nonmembers use the west entrance",
  "visitor": "nonmember"
}
```

### Criteria

- `option_1` (value: None): "West entrance"
- `option_2` (value: None): "East entrance"

Target: {'option_1': 1.0, 'option_2': 0.0}

The visitor is a nonmember.

## synthetic-choice-036

### Instruction

```json
"Which conclusion follows from the given hierarchy? Do not reverse the rules."
```

### State

```json
{
  "fact": "Pip is a lark",
  "rules": [
    "All larks are birds",
    "All birds are animals"
  ]
}
```

### Criteria

- `option_1` (value: None): "No lark is an animal"
- `option_2` (value: None): "All animals are larks"
- `option_3` (value: None): "Pip is not a bird"
- `option_4` (value: None): "Every animal is Pip"
- `option_5` (value: None): "Only Pip is a bird"
- `option_6` (value: None): "Pip is an animal"
- `option_7` (value: None): "Pip is a fish"
- `option_8` (value: None): "All birds are larks"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 0.0, 'option_5': 0.0, 'option_6': 1.0, 'option_7': 0.0, 'option_8': 0.0}

Lark implies bird, which implies animal.

## synthetic-choice-037

### Instruction

```json
"Which missing fact prevents answering the question?"
```

### State

```json
{
  "given": "Revenue was $900. Profit equals revenue minus expenses. Expenses were not recorded.",
  "question": "What was the profit?"
}
```

### Criteria

- `option_1` (value: None): "Company name"
- `option_2` (value: None): "Revenue"
- `option_3` (value: None): "Expenses"
- `option_4` (value: None): "Report font"
- `option_5` (value: None): "Number of pages"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 1.0, 'option_4': 0.0, 'option_5': 0.0}

Expenses are required to compute profit from the given formula.

## synthetic-choice-038

### Instruction

```json
"Which clause wins under the explicit precedence rule?"
```

### State

```json
{
  "general": "Guests may use all rooms",
  "precedence": "Specific restrictions override general permissions",
  "specific": "Guests may not enter the archive",
  "target": "archive"
}
```

### Criteria

- `option_1` (value: None): "Specific restriction"
- `option_2` (value: None): "No applicable clause"
- `option_3` (value: None): "Whichever was read first"
- `option_4` (value: None): "General permission"

Target: {'option_1': 1.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 0.0}

The specific archive restriction overrides the general permission.

## synthetic-choice-039

### Instruction

```json
"Where is the token after these moves?"
```

### State

```json
{
  "moves": [
    "drawer to table",
    "table to pocket",
    "pocket to shelf"
  ],
  "start": "drawer"
}
```

### Criteria

- `option_1` (value: None): "Pocket"
- `option_2` (value: None): "Shelf"
- `option_3` (value: None): "Table"
- `option_4` (value: None): "Drawer"

Target: {'option_1': 0.0, 'option_2': 1.0, 'option_3': 0.0, 'option_4': 0.0}

The last move places it on the shelf.

## synthetic-choice-040

### Instruction

```json
"Select the strongest warranted conclusion."
```

### State

```json
"Three sampled apples were sweet. No other apples were tasted."
```

### Criteria

- `option_1` (value: None): "The three sampled apples were sweet"
- `option_2` (value: None): "Exactly three apples in existence are sweet"
- `option_3` (value: None): "All apples are sweet"
- `option_4` (value: None): "No apples are sour"

Target: {'option_1': 1.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 0.0}

Only the sampled apples' sweetness is established.

## synthetic-choice-041

### Instruction

```json
"Which citation can support the exact claim?"
```

### State

```json
{
  "claim": "The east bridge opened in 2001",
  "sources": {
    "A": "West bridge opened in 2001",
    "B": "East bridge renovated in 2001",
    "C": "East bridge opened in 2001",
    "D": "East bridge designed in 1998",
    "E": "Bridge opening date unknown",
    "F": "East bridge closed in 2020"
  }
}
```

### Criteria

- `option_1` (value: None): "Source A"
- `option_2` (value: None): "Source B"
- `option_3` (value: None): "Source F"
- `option_4` (value: None): "Source D"
- `option_5` (value: None): "Source C"
- `option_6` (value: None): "Source E"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 0.0, 'option_5': 1.0, 'option_6': 0.0}

Only C matches the bridge, event and year.

## synthetic-choice-042

### Instruction

```json
"Find the failure category shown by this log."
```

### State

```json
"Connection established. Login accepted. Requested file /report.csv does not exist."
```

### Criteria

- `option_1` (value: None): "Network connection failure"
- `option_2` (value: None): "Authentication failure"
- `option_3` (value: None): "Successful download"
- `option_4` (value: None): "Missing resource"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 1.0}

Connection and authentication succeeded; the file is absent.

## synthetic-choice-043

### Instruction

```json
"Which sense of 'light' is being used?"
```

### State

```json
"This suitcase is light enough to carry upstairs with one hand."
```

### Criteria

- `option_1` (value: None): "Illumination"
- `option_2` (value: None): "Low weight"
- `option_3` (value: None): "Pale color"

Target: {'option_1': 0.0, 'option_2': 1.0, 'option_3': 0.0}

The sentence concerns carrying weight.

## synthetic-choice-044

### Instruction

```json
"What would directly answer the person's request without inventing a new task?"
```

### State

```json
"Could you tell me whether the meeting is online or in person? I already know the time."
```

### Criteria

- `option_1` (value: None): "It starts at ten"
- `option_2` (value: None): "The meeting is online"
- `option_3` (value: None): "The team has seven people"
- `option_4` (value: None): "Please reset your password"
- `option_5` (value: None): "The agenda has three items"
- `option_6` (value: None): "I like the venue"
- `option_7` (value: None): "Lunch is included"

Target: {'option_1': 0.0, 'option_2': 1.0, 'option_3': 0.0, 'option_4': 0.0, 'option_5': 0.0, 'option_6': 0.0, 'option_7': 0.0}

Online answers the format question; the time is already known.

## synthetic-choice-045

### Instruction

```json
"Select the message that accurately communicates this outcome and next step."
```

### State

```json
{
  "next_step": "Try another payment method",
  "outcome": "Payment failed; no money charged"
}
```

### Criteria

- `option_1` (value: None): "Refund complete"
- `option_2` (value: None): "Payment complete"
- `option_3` (value: None): "Your account was charged twice"
- `option_4` (value: None): "Your password expired"
- `option_5` (value: None): "Payment failed; nothing was charged. Try another payment method."
- `option_6` (value: None): "Order shipped"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 0.0, 'option_5': 1.0, 'option_6': 0.0}

Only one message matches both the failure, no-charge outcome and next action.

## synthetic-choice-046

### Instruction

```json
"Identify the antecedent specified by the sentence, not by proximity alone."
```

### State

```json
"The mechanic told the driver, 'Your car is ready.' The person who repaired the car then handed over the keys. In this story, only the mechanic repaired it."
```

### Criteria

- `option_1` (value: None): "Unmentioned passenger"
- `option_2` (value: None): "Driver"
- `option_3` (value: None): "Mechanic"
- `option_4` (value: None): "Car"
- `option_5` (value: None): "Keys"
- `option_6` (value: None): "No unique referent"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 1.0, 'option_4': 0.0, 'option_5': 0.0, 'option_6': 0.0}

The explicit repair fact identifies the mechanic.

## synthetic-choice-047

### Instruction

```json
"Which question would resolve the only stated blocker?"
```

### State

```json
{
  "status": "Scope, owner, budget, and location are confirmed. Start date is unknown."
}
```

### Criteria

- `option_1` (value: None): "Where will it happen?"
- `option_2` (value: None): "Who owns it?"
- `option_3` (value: None): "Is it already cancelled?"
- `option_4` (value: None): "What is the budget?"
- `option_5` (value: None): "What is the scope?"
- `option_6` (value: None): "When should it start?"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 0.0, 'option_5': 0.0, 'option_6': 1.0}

Only the start date remains unknown.

## synthetic-choice-048

### Instruction

```json
"Route this, please. Use the actual problem, not isolated keywords."
```

### State

```json
"I know my password. The reset email isn't the issue. My account says it's suspended after a policy review."
```

### Criteria

- `option_1` (value: None): "Change display name"
- `option_2` (value: None): "Email delivery"
- `option_3` (value: None): "Account suspension"
- `option_4` (value: None): "Password reset"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 1.0, 'option_4': 0.0}

The user explicitly identifies suspension and excludes reset issues.

## synthetic-choice-049

### Instruction

```json
"Label the recorded event relative to the reference time."
```

### State

```json
{
  "date": "same day",
  "event": "13:45",
  "reference": "14:00"
}
```

### Criteria

- `option_1` (value: None): "Later"
- `option_2` (value: None): "Earlier"

Target: {'option_1': 0.0, 'option_2': 1.0}

13:45 precedes 14:00.

## synthetic-choice-050

### Instruction

```json
"Choose the rewrite that retains the original uncertainty."
```

### State

```json
{
  "original": "The repair may reduce the noise."
}
```

### Criteria

- `option_1` (value: None): "The repair will eliminate the noise."
- `option_2` (value: None): "The repair already eliminated the noise."
- `option_3` (value: None): "The repair cannot affect the noise."
- `option_4` (value: None): "The repair might make the noise lower."
- `option_5` (value: None): "The repair must increase the noise."

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 1.0, 'option_5': 0.0}

Might reduce preserves the original possibility without certainty.

## synthetic-choice-051

### Instruction

```json
"What does this operation do to the input list?"
```

### State

```json
"result = sorted(items)"
```

### Criteria

- `option_1` (value: None): "Mutates it in place"
- `option_2` (value: None): "Leaves it unchanged"

Target: {'option_1': 0.0, 'option_2': 1.0}

sorted creates a new list.

## synthetic-choice-052

### Instruction

```json
"Choose the event that is a remembered past event, not part of the current action."
```

### State

```json
"Now, Erin waits at the gate, checks the clock, and hears a train. She remembers losing her ticket yesterday. Then she opens her bag."
```

### Criteria

- `option_1` (value: None): "Checking the clock"
- `option_2` (value: None): "Opening the bag"
- `option_3` (value: None): "Losing the ticket yesterday"
- `option_4` (value: None): "Waiting at the gate"
- `option_5` (value: None): "Hearing the train"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 1.0, 'option_4': 0.0, 'option_5': 0.0}

Losing the ticket is explicitly a memory from yesterday.

## synthetic-choice-053

### Instruction

```json
"Choose the best next clarification given the one missing required field."
```

### State

```json
{
  "message": "Order R14, blue mug, handle arrived broken. I haven't said whether I want a refund or replacement.",
  "required": [
    "order ID",
    "item",
    "problem",
    "desired resolution"
  ]
}
```

### Criteria

- `option_1` (value: None): "Which marketing emails do you read?"
- `option_2` (value: None): "What is your password?"
- `option_3` (value: None): "What is broken?"
- `option_4` (value: None): "When did you join the newsletter?"
- `option_5` (value: None): "Would you prefer a refund or replacement?"
- `option_6` (value: None): "Which item?"
- `option_7` (value: None): "What is your order ID?"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 0.0, 'option_5': 1.0, 'option_6': 0.0, 'option_7': 0.0}

Only the desired resolution is missing.

## synthetic-choice-054

### Instruction

```json
"Which statement is directly warranted?"
```

### State

```json
"A sensor recorded a temperature above 30 degrees at noon. It did not record other times, humidity, or rainfall."
```

### Criteria

- `option_1` (value: None): "It was above 30 degrees at noon"
- `option_2` (value: None): "It rained at noon"
- `option_3` (value: None): "It was above 30 all day"
- `option_4` (value: None): "Yesterday was cooler"
- `option_5` (value: None): "The sensor is always accurate"
- `option_6` (value: None): "Humidity was high"
- `option_7` (value: None): "It will be hot tomorrow"
- `option_8` (value: None): "Every nearby town was hot"

Target: {'option_1': 1.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 0.0, 'option_5': 0.0, 'option_6': 0.0, 'option_7': 0.0, 'option_8': 0.0}

Only the noon temperature is recorded; the other claims add unsupported information.

## synthetic-choice-055

### Instruction

```json
"Select the applicable retention period from the complete policy."
```

### State

```json
{
  "document_type": "log",
  "policy": {
    "contract": "90 days",
    "draft": "7 days",
    "log": "14 days",
    "newsletter": "1 day",
    "receipt": "30 days",
    "template": "indefinite"
  }
}
```

### Criteria

- `option_1` (value: None): "30 days"
- `option_2` (value: None): "14 days"
- `option_3` (value: None): "90 days"
- `option_4` (value: None): "Indefinite"
- `option_5` (value: None): "7 days"
- `option_6` (value: None): "1 day"

Target: {'option_1': 0.0, 'option_2': 1.0, 'option_3': 0.0, 'option_4': 0.0, 'option_5': 0.0, 'option_6': 0.0}

The policy assigns logs fourteen days.

## synthetic-choice-056

### Instruction

```json
"Choose the appropriate ticket category from this catalog."
```

### State

```json
"I can log in and pay, but your site is unreadable with my screen reader because the controls have no labels."
```

### Criteria

- `option_1` (value: None): "Tax receipt"
- `option_2` (value: None): "Marketing preferences"
- `option_3` (value: None): "Login"
- `option_4` (value: None): "Cancellation"
- `option_5` (value: None): "Accessibility"
- `option_6` (value: None): "Product availability"
- `option_7` (value: None): "Delivery"
- `option_8` (value: None): "Billing"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 0.0, 'option_5': 1.0, 'option_6': 0.0, 'option_7': 0.0, 'option_8': 0.0}

Unlabeled screen-reader controls are an accessibility issue.

## synthetic-choice-057

### Instruction

```json
"Find the requested account operation."
```

### State

```json
"Keep my account and subscription active, but stop sending promotional messages."
```

### Criteria

- `option_1` (value: None): "Cancel subscription"
- `option_2` (value: None): "Delete account"
- `option_3` (value: None): "Opt out of marketing"
- `option_4` (value: None): "Change billing address"
- `option_5` (value: None): "Reset password"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 1.0, 'option_4': 0.0, 'option_5': 0.0}

Only promotional messaging should stop.

## synthetic-choice-058

### Instruction

```json
{
  "avoid": "Visual appearance alone",
  "must_include": "Both the action and its target",
  "task": "Choose an icon description"
}
```

### State

```json
{
  "action": "Open settings for notification sounds"
}
```

### Criteria

- `option_1` (value: None): "Open account billing"
- `option_2` (value: None): "Mute every device"
- `option_3` (value: None): "Delete notifications"
- `option_4` (value: None): "Button"
- `option_5` (value: None): "Play music"
- `option_6` (value: None): "Bell"
- `option_7` (value: None): "Open notification-sound settings"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 0.0, 'option_5': 0.0, 'option_6': 0.0, 'option_7': 1.0}

Only that description matches the exact action and target.

## synthetic-choice-059

### Instruction

```json
"For the minutes, classify the decision status."
```

### State

```json
"The proposal was discussed, but the chair postponed the vote until next week."
```

### Criteria

- `option_1` (value: None): "Approved"
- `option_2` (value: None): "Rejected"
- `option_3` (value: None): "Deferred"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 1.0}

The vote and decision were postponed.

## synthetic-choice-060

### Instruction

```json
"What operation is this user asking for?"
```

### State

```json
"Take every timestamp in the log and order the entries from oldest to newest. Keep repeated entries."
```

### Criteria

- `option_1` (value: None): "Descending chronological sort"
- `option_2` (value: None): "Deduplication"
- `option_3` (value: None): "Grouping by severity only"
- `option_4` (value: None): "Random sampling"
- `option_5` (value: None): "Encryption"
- `option_6` (value: None): "Ascending chronological sort"
- `option_7` (value: None): "Counting words"
- `option_8` (value: None): "Redaction"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 0.0, 'option_5': 0.0, 'option_6': 1.0, 'option_7': 0.0, 'option_8': 0.0}

Oldest to newest means ascending sort; repeats must remain.

## synthetic-choice-061

### Instruction

```json
"Where should this ticket go?"
```

### State

```json
"I was billed twice for one order. Please reverse the extra charge."
```

### Criteria

- `option_1` (value: None): "Delivery"
- `option_2` (value: None): "Billing"

Target: {'option_1': 0.0, 'option_2': 1.0}

The problem is a duplicate charge, not shipping.

## synthetic-choice-062

### Instruction

```json
"Which record has the exact composite key?"
```

### State

```json
{
  "records": {
    "A": [
      "north",
      2024,
      "actual"
    ],
    "B": [
      "south",
      2025,
      "actual"
    ],
    "C": [
      "north",
      2025,
      "forecast"
    ],
    "D": [
      "north",
      2025,
      "actual"
    ],
    "E": [
      "south",
      2024,
      "forecast"
    ],
    "F": [
      "east",
      2025,
      "actual"
    ],
    "G": [
      "north",
      2026,
      "actual"
    ],
    "H": [
      "west",
      2025,
      "actual"
    ]
  },
  "target": {
    "kind": "actual",
    "region": "north",
    "year": 2025
  }
}
```

### Criteria

- `option_1` (value: None): "B"
- `option_2` (value: None): "F"
- `option_3` (value: None): "A"
- `option_4` (value: None): "H"
- `option_5` (value: None): "E"
- `option_6` (value: None): "G"
- `option_7` (value: None): "D"
- `option_8` (value: None): "C"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 0.0, 'option_5': 0.0, 'option_6': 0.0, 'option_7': 1.0, 'option_8': 0.0}

D matches region, year and kind.

## synthetic-choice-063

### Instruction

```json
"What should the editor flag here?"
```

### State

```json
"The event is on Tuesday, June 4. Later in the same note: the event is on Thursday, June 6. Both refer to the same event; neither is marked as a correction."
```

### Criteria

- `option_1` (value: None): "Missing event date"
- `option_2` (value: None): "Explicit cancellation"
- `option_3` (value: None): "Conflicting event dates"
- `option_4` (value: None): "Consistent repetition"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 1.0, 'option_4': 0.0}

Two uncorrected dates are given for the same event.

## synthetic-choice-064

### Instruction

```json
"Which value survives this last-write-wins update stream?"
```

### State

```json
{
  "initial": "red",
  "updates": [
    "blue",
    "green",
    "blue"
  ]
}
```

### Criteria

- `option_1` (value: None): "All three simultaneously"
- `option_2` (value: None): "Blue"
- `option_3` (value: None): "Red"
- `option_4` (value: None): "Green"

Target: {'option_1': 0.0, 'option_2': 1.0, 'option_3': 0.0, 'option_4': 0.0}

The final update is blue.

## synthetic-choice-065

### Instruction

```json
"Which reading fits the whole message?"
```

### State

```json
"Oh great, the printer jammed again. Just what I needed before the deadline."
```

### Criteria

- `option_1` (value: None): "Sincere praise"
- `option_2` (value: None): "Sarcastic complaint"

Target: {'option_1': 0.0, 'option_2': 1.0}

Repeated failure before a deadline makes the apparent praise sarcastic.

## synthetic-choice-066

### Instruction

```json
"Which meaning of 'draft' fits this context?"
```

### State

```json
"Please review the draft before I send the final contract."
```

### Criteria

- `option_1` (value: None): "Bank payment instrument"
- `option_2` (value: None): "Selection of sports players"
- `option_3` (value: None): "Serving beer from a tap"
- `option_4` (value: None): "Preliminary version"
- `option_5` (value: None): "Depth of a vessel"
- `option_6` (value: None): "Air current"
- `option_7` (value: None): "Military conscription"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 1.0, 'option_5': 0.0, 'option_6': 0.0, 'option_7': 0.0}

Review before a final contract concerns a preliminary version.

## synthetic-choice-067

### Instruction

```json
"Which day is the earliest available under all constraints? Use this fictional six-day calendar."
```

### State

```json
{
  "days_in_order": [
    "Day A",
    "Day B",
    "Day C",
    "Day D",
    "Day E",
    "Day F"
  ],
  "room_free": [
    "Day B",
    "Day D",
    "Day F"
  ],
  "speaker_free": [
    "Day C",
    "Day D",
    "Day F"
  ]
}
```

### Criteria

- `option_1` (value: None): "Day C"
- `option_2` (value: None): "Day B"
- `option_3` (value: None): "Day F"
- `option_4` (value: None): "Day D"
- `option_5` (value: None): "Day A"
- `option_6` (value: None): "Day E"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 1.0, 'option_5': 0.0, 'option_6': 0.0}

Day D is the earliest intersection of the availability sets.

## synthetic-choice-068

### Instruction

```json
"Pick the candidate that matches the same intent rather than just a shared word."
```

### State

```json
{
  "query": "How do I freeze my lost payment card?"
}
```

### Criteria

- `option_1` (value: None): "How to change a card's color"
- `option_2` (value: None): "Where to buy greeting cards"
- `option_3` (value: None): "How to thaw frozen food"
- `option_4` (value: None): "How to stop a lost card from being used"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 1.0}

Freezing a payment card means preventing its use.

## synthetic-choice-069

### Instruction

```json
"Which recipe meets the full request as described? Don't infer unspecified substitutions."
```

### State

```json
{
  "need": "Cold, savory, and no dairy",
  "recipes": {
    "A": "hot savory dairy-free stew",
    "B": "cold sweet dairy-free fruit cup",
    "C": "cold savory yogurt soup",
    "D": "cold savory dairy-free bean salad",
    "E": "hot sweet rice pudding with milk",
    "F": "hot savory cheese toast",
    "G": "cold sweet yogurt bowl"
  }
}
```

### Criteria

- `option_1` (value: None): "Recipe C"
- `option_2` (value: None): "Recipe A"
- `option_3` (value: None): "Recipe B"
- `option_4` (value: None): "Recipe E"
- `option_5` (value: None): "Recipe D"
- `option_6` (value: None): "Recipe F"
- `option_7` (value: None): "Recipe G"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 0.0, 'option_5': 1.0, 'option_6': 0.0, 'option_7': 0.0}

D alone satisfies temperature, flavor and dairy exclusion.

## synthetic-choice-070

### Instruction

```json
"Who is narrating this sentence?"
```

### State

```json
"I put my notebook away and followed the guide."
```

### Criteria

- `option_1` (value: None): "A participant using first person"
- `option_2` (value: None): "An outside narrator using third person"

Target: {'option_1': 1.0, 'option_2': 0.0}

I and my mark participant first-person narration.

## synthetic-choice-071

### Instruction

```json
"Choose the action permitted by this complete miniature policy."
```

### State

```json
{
  "actor": "guest",
  "policy": "Guests may read public posts. Only members may comment, upload, edit profiles, send messages, create groups, or invite others."
}
```

### Criteria

- `option_1` (value: None): "Invite others"
- `option_2` (value: None): "Comment"
- `option_3` (value: None): "Send messages"
- `option_4` (value: None): "Read public posts"
- `option_5` (value: None): "Edit profile"
- `option_6` (value: None): "Upload"
- `option_7` (value: None): "Create groups"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 1.0, 'option_5': 0.0, 'option_6': 0.0, 'option_7': 0.0}

Reading public posts is the only listed guest permission.

## synthetic-choice-072

### Instruction

```json
"Route by the requested action, please."
```

### State

```json
"The address on my order is wrong. It hasn't shipped; please send it to my new address instead."
```

### Criteria

- `option_1` (value: None): "Track shipment"
- `option_2` (value: None): "Report duplicate charge"
- `option_3` (value: None): "Change delivery address"
- `option_4` (value: None): "Download invoice"
- `option_5` (value: None): "Reset login"
- `option_6` (value: None): "Cancel membership"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 1.0, 'option_4': 0.0, 'option_5': 0.0, 'option_6': 0.0}

The user asks to change the shipping destination before dispatch.

## synthetic-choice-073

### Instruction

```json
"Choose the result for this exact input."
```

### State

```json
{
  "expression": "5 > 5",
  "language": "Python"
}
```

### Criteria

- `option_1` (value: None): "True"
- `option_2` (value: None): "False"
- `option_3` (value: None): "Raises an error"

Target: {'option_1': 0.0, 'option_2': 1.0, 'option_3': 0.0}

Strict greater-than excludes equality.

## synthetic-choice-074

### Instruction

```json
"Select the department named as the approval owner, not the requester."
```

### State

```json
{
  "approval_owner": "Procurement",
  "observers": [
    "Finance",
    "Legal"
  ],
  "requester": "Engineering"
}
```

### Criteria

- `option_1` (value: None): "Engineering"
- `option_2` (value: None): "Support"
- `option_3` (value: None): "Marketing"
- `option_4` (value: None): "Research"
- `option_5` (value: None): "Procurement"
- `option_6` (value: None): "Facilities"
- `option_7` (value: None): "Legal"
- `option_8` (value: None): "Finance"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 0.0, 'option_5': 1.0, 'option_6': 0.0, 'option_7': 0.0, 'option_8': 0.0}

Procurement is explicitly the approval owner.

## synthetic-choice-075

### Instruction

```json
"Identify the deliberately changed variable."
```

### State

```json
{
  "experiment": "Use identical pots, soil, and watering. Give each group a different number of hours of light, then measure growth."
}
```

### Criteria

- `option_1` (value: None): "Pot size"
- `option_2` (value: None): "Measured growth"
- `option_3` (value: None): "Hours of light"
- `option_4` (value: None): "Water quantity"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 1.0, 'option_4': 0.0}

Light duration is manipulated; growth is the measured outcome.

## synthetic-choice-076

### Instruction

```json
"Select the quantity that is the outcome, not an input or identifier."
```

### State

```json
{
  "study": "Across labeled pots, vary fertilizer amount while recording plant growth after 14 days. Keep soil, light and water fixed."
}
```

### Criteria

- `option_1` (value: None): "Water amount"
- `option_2` (value: None): "Pot label"
- `option_3` (value: None): "Plant growth after 14 days"
- `option_4` (value: None): "Soil type"
- `option_5` (value: None): "Light level"
- `option_6` (value: None): "Fertilizer amount"
- `option_7` (value: None): "Researcher's name"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 1.0, 'option_4': 0.0, 'option_5': 0.0, 'option_6': 0.0, 'option_7': 0.0}

Growth is the measured response to fertilizer amount.

## synthetic-choice-077

### Instruction

```json
"Find the latest appointment ending before 16:00. All last 30 minutes; ending exactly at 16:00 is NOT before it."
```

### State

```json
{
  "starts": [
    "12:00",
    "12:30",
    "13:00",
    "13:30",
    "14:00",
    "14:30",
    "15:00",
    "15:30"
  ]
}
```

### Criteria

- `option_1` (value: None): "14:30"
- `option_2` (value: None): "13:00"
- `option_3` (value: None): "15:00"
- `option_4` (value: None): "12:30"
- `option_5` (value: None): "14:00"
- `option_6` (value: None): "13:30"
- `option_7` (value: None): "15:30"
- `option_8` (value: None): "12:00"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 1.0, 'option_4': 0.0, 'option_5': 0.0, 'option_6': 0.0, 'option_7': 0.0, 'option_8': 0.0}

15:00 ends at 15:30; 15:30 ends exactly at the excluded boundary.

## synthetic-choice-078

### Instruction

```json
"What should the meeting organizer do next under the stated workflow?"
```

### State

```json
{
  "completed": [
    "choose date",
    "book room"
  ],
  "workflow": "Choose date, book room, send invitations, collect responses, finalize attendance"
}
```

### Criteria

- `option_1` (value: None): "Send invitations"
- `option_2` (value: None): "Collect responses"
- `option_3` (value: None): "Finalize attendance"
- `option_4` (value: None): "Book room"
- `option_5` (value: None): "Choose date"

Target: {'option_1': 1.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 0.0, 'option_5': 0.0}

Invitations are the next incomplete step.

## synthetic-choice-079

### Instruction

```json
"Which passage answers the requested fact most directly?"
```

### State

```json
{
  "passages": {
    "fare": "Tickets cost $6.",
    "history": "The service began in 1982.",
    "location": "Board at Pier 2.",
    "schedule": "Departure is 08:30."
  },
  "question": "What time does the ferry leave?"
}
```

### Criteria

- `option_1` (value: None): "Fare passage"
- `option_2` (value: None): "History passage"
- `option_3` (value: None): "Location passage"
- `option_4` (value: None): "Schedule passage"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 1.0}

Only the schedule gives a departure time.

## synthetic-choice-080

### Instruction

```json
"Which result should this parser produce under the stated priority rule?"
```

### State

```json
{
  "rules_in_order": [
    "If text starts with !, classify as command",
    "Else if it ends with ?, classify as question",
    "Otherwise classify as statement"
  ],
  "text": "!status?"
}
```

### Criteria

- `option_1` (value: None): "Empty input"
- `option_2` (value: None): "Question"
- `option_3` (value: None): "Statement"
- `option_4` (value: None): "Invalid encoding"
- `option_5` (value: None): "Both command and question"
- `option_6` (value: None): "Number"
- `option_7` (value: None): "Command"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 0.0, 'option_5': 0.0, 'option_6': 0.0, 'option_7': 1.0}

The first matching rule takes priority.

## synthetic-choice-081

### Instruction

```json
"Choose the emotion explicitly expressed, not what you might expect."
```

### State

```json
"Everyone thought I'd be upset, but honestly I'm relieved the competition is over."
```

### Criteria

- `option_1` (value: None): "Jealousy"
- `option_2` (value: None): "Anger"
- `option_3` (value: None): "Relief"
- `option_4` (value: None): "Fear"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 1.0, 'option_4': 0.0}

The speaker explicitly says relieved.

## synthetic-choice-082

### Instruction

```json
[
  "Read the complete update.",
  "Classify what is blocking the work now."
]
```

### State

```json
"The code is finished and tests pass. We cannot deploy until the service owner approves."
```

### Criteria

- `option_1` (value: None): "Deployment completed"
- `option_2` (value: None): "Approval pending"
- `option_3` (value: None): "Implementation unfinished"
- `option_4` (value: None): "Tests failing"

Target: {'option_1': 0.0, 'option_2': 1.0, 'option_3': 0.0, 'option_4': 0.0}

Only service-owner approval remains.

## synthetic-choice-083

### Instruction

```json
"What can this source tell us about the designer?"
```

### State

```json
{
  "question": "Was Morgan the designer?",
  "source": "Morgan visited the building last spring."
}
```

### Criteria

- `option_1` (value: None): "Definitely Morgan"
- `option_2` (value: None): "Definitely someone else"
- `option_3` (value: None): "Not established"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 1.0}

Visiting does not establish who designed it.

## synthetic-choice-084

### Instruction

```json
"On which day does the event fall in this fictional schedule?"
```

### State

```json
{
  "event": "two days after the reference",
  "reference": "Wednesday",
  "week_order": [
    "Monday",
    "Tuesday",
    "Wednesday",
    "Thursday",
    "Friday",
    "Saturday",
    "Sunday"
  ]
}
```

### Criteria

- `option_1` (value: None): "Wednesday"
- `option_2` (value: None): "Friday"
- `option_3` (value: None): "Tuesday"
- `option_4` (value: None): "Sunday"
- `option_5` (value: None): "Saturday"
- `option_6` (value: None): "Thursday"
- `option_7` (value: None): "Monday"

Target: {'option_1': 0.0, 'option_2': 1.0, 'option_3': 0.0, 'option_4': 0.0, 'option_5': 0.0, 'option_6': 0.0, 'option_7': 0.0}

Two days after Wednesday is Friday.

## synthetic-choice-085

### Instruction

```json
"Select the precise defect in the inference."
```

### State

```json
{
  "conclusion": "Taylor is a gold member",
  "observation": "Taylor can enter the lounge",
  "premise": "All gold members can enter the lounge"
}
```

### Criteria

- `option_1` (value: None): "Misspells a name"
- `option_2` (value: None): "Changes units"
- `option_3` (value: None): "Quotes the wrong speaker"
- `option_4` (value: None): "Confuses past and future"
- `option_5` (value: None): "No defect; the conclusion is necessary"
- `option_6` (value: None): "Reverses a one-way implication"
- `option_7` (value: None): "Adds two numbers incorrectly"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 0.0, 'option_5': 0.0, 'option_6': 1.0, 'option_7': 0.0}

Entry may be available to other people; the implication cannot be reversed.

## synthetic-choice-086

### Instruction

```json
"What does 'bank' refer to here?"
```

### State

```json
"We sat on the grassy bank and watched the river flow past."
```

### Criteria

- `option_1` (value: None): "River edge"
- `option_2` (value: None): "Financial institution"
- `option_3` (value: None): "Aircraft tilt"
- `option_4` (value: None): "Stored reserve"

Target: {'option_1': 1.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 0.0}

The grassy location beside the river is its bank.

## synthetic-choice-087

### Instruction

```json
"Which result concerns repairing a physical keyboard key?"
```

### State

```json
{
  "query": "My laptop's space bar is stuck. How do I fix the key?"
}
```

### Criteria

- `option_1` (value: None): "Replace a door key"
- `option_2` (value: None): "Explain a map key"
- `option_3` (value: None): "Clean a stuck laptop key"
- `option_4` (value: None): "Recover an encryption key"
- `option_5` (value: None): "Change a musical key"
- `option_6` (value: None): "Find a database primary key"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 1.0, 'option_4': 0.0, 'option_5': 0.0, 'option_6': 0.0}

The query concerns a stuck physical laptop key.

## synthetic-choice-088

### Instruction

```json
"Pick a room satisfying every stated condition. Missing information does not count as a match."
```

### State

```json
{
  "needs": [
    "two beds",
    "private bathroom",
    "under $120"
  ],
  "rooms": {
    "A": "two beds, private bath, $110",
    "B": "one bed, private bath, $100",
    "C": "two beds, shared bath, $90",
    "D": "two beds, private bath, $130",
    "E": "two beds, bath unspecified, $100",
    "F": "one bed, shared bath, $80"
  }
}
```

### Criteria

- `option_1` (value: None): "Room D"
- `option_2` (value: None): "Room E"
- `option_3` (value: None): "Room C"
- `option_4` (value: None): "Room A"
- `option_5` (value: None): "Room F"
- `option_6` (value: None): "Room B"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 1.0, 'option_5': 0.0, 'option_6': 0.0}

Only Room A explicitly satisfies all three requirements.

## synthetic-choice-089

### Instruction

```json
"Which approval is required at this exact boundary?"
```

### State

```json
{
  "amount": 500,
  "rule": "Below $500: team lead. $500–$1,999 inclusive: manager. $2,000–$9,999: director. $10,000 or more: board."
}
```

### Criteria

- `option_1` (value: None): "Team lead"
- `option_2` (value: None): "Board"
- `option_3` (value: None): "Director"
- `option_4` (value: None): "Manager"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 1.0}

Exactly $500 enters the manager band.

## synthetic-choice-090

### Instruction

```json
"What is the writer mainly doing in the second sentence?"
```

### State

```json
"I thought you had the latest file. I now realize I sent you the old one."
```

### Criteria

- `option_1` (value: None): "Recognizing their own mistake"
- `option_2` (value: None): "Celebrating success"
- `option_3` (value: None): "Giving travel advice"
- `option_4` (value: None): "Requesting payment"
- `option_5` (value: None): "Denying responsibility"
- `option_6` (value: None): "Making a threat"

Target: {'option_1': 1.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 0.0, 'option_5': 0.0, 'option_6': 0.0}

The writer recognizes sending the old file.

## synthetic-choice-091

### Instruction

```json
"Who can satisfy both requirements?"
```

### State

```json
{
  "need": "Can drive and is available Tuesday",
  "people": {
    "Ari": "drives; Monday only",
    "Bo": "does not drive; Tuesday",
    "Cy": "drives; Tuesday",
    "Dee": "does not drive; Monday",
    "Eli": "drives; Friday"
  }
}
```

### Criteria

- `option_1` (value: None): "Eli"
- `option_2` (value: None): "Ari"
- `option_3` (value: None): "Dee"
- `option_4` (value: None): "Bo"
- `option_5` (value: None): "Cy"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 0.0, 'option_5': 1.0}

Only Cy both drives and is available Tuesday.

## synthetic-choice-092

### Instruction

```json
"Choose the relation between the clauses."
```

### State

```json
"The path was flooded, so we took the bridge."
```

### Criteria

- `option_1` (value: None): "Direct quotation"
- `option_2` (value: None): "Definition and example"
- `option_3` (value: None): "Cause and consequence"
- `option_4` (value: None): "Unrelated enumeration"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 1.0, 'option_4': 0.0}

Flooding causes the route change.

## synthetic-choice-093

### Instruction

```json
"Which button label best matches this action without promising more?"
```

### State

```json
{
  "action": "Stores an unfinished document without making it visible to readers"
}
```

### Criteria

- `option_1` (value: None): "Delete document"
- `option_2` (value: None): "Save draft"
- `option_3` (value: None): "Publish now"
- `option_4` (value: None): "Invite readers"

Target: {'option_1': 0.0, 'option_2': 1.0, 'option_3': 0.0, 'option_4': 0.0}

Saving a draft preserves work without publishing.

## synthetic-choice-094

### Instruction

```json
"Is the reply accepting or declining the invitation?"
```

### State

```json
"I'd love to, but I can't make that day. Please count me out."
```

### Criteria

- `option_1` (value: None): "Declining"
- `option_2` (value: None): "Accepting"

Target: {'option_1': 1.0, 'option_2': 0.0}

Count me out explicitly declines despite the polite opening.

## synthetic-choice-095

### Instruction

```json
"Choose the conclusion justified by the complete record. Don't turn missing information into a fact."
```

### State

```json
{
  "question": "Which of the seven couriers delivered the package?",
  "record": "The package arrived. No courier name, signature or tracking ID was recorded."
}
```

### Criteria

- `option_1` (value: None): "Courier Cedar"
- `option_2` (value: None): "Courier Maple"
- `option_3` (value: None): "Courier Elm"
- `option_4` (value: None): "Courier Birch"
- `option_5` (value: None): "Courier Oak"
- `option_6` (value: None): "Courier Fir"
- `option_7` (value: None): "Courier Ash"
- `option_8` (value: None): "Cannot determine from the record"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 0.0, 'option_5': 0.0, 'option_6': 0.0, 'option_7': 0.0, 'option_8': 1.0}

Delivery is known but courier identity is unrecorded.

## synthetic-choice-096

### Instruction

```json
"Choose the function of the sentence beginning 'For example'."
```

### State

```json
"Many objects can be reused. For example, a jar can become a pencil holder."
```

### Criteria

- `option_1` (value: None): "Retraction"
- `option_2` (value: None): "Counterargument"
- `option_3` (value: None): "Command"
- `option_4` (value: None): "Illustration"
- `option_5` (value: None): "Question"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 1.0, 'option_5': 0.0}

The jar illustrates the broader reuse claim.

## synthetic-choice-097

### Instruction

```json
{
  "rule": "null means not collected; zero is a measurement",
  "task": "Classify observation status"
}
```

### State

```json
{
  "reading": 0
}
```

### Criteria

- `option_1` (value: None): "Invalid format"
- `option_2` (value: None): "Missing"
- `option_3` (value: None): "Measured zero"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 1.0}

Zero is explicitly a measured value.

## synthetic-choice-098

### Instruction

```json
"Select the sole eligible applicant under the full rule."
```

### State

```json
{
  "applicants": {
    "A": [
      true,
      false,
      false
    ],
    "B": [
      false,
      true,
      false
    ],
    "C": [
      true,
      true,
      true
    ],
    "D": [
      false,
      false,
      false
    ],
    "E": [
      true,
      false,
      true
    ],
    "F": [
      false,
      true,
      true
    ],
    "G": [
      true,
      true,
      false
    ],
    "H": [
      false,
      false,
      true
    ]
  },
  "field_order": [
    "member",
    "orientation_complete",
    "has_overdue_equipment"
  ],
  "rule": "Must be a member, have completed orientation, and have no overdue equipment"
}
```

### Criteria

- `option_1` (value: None): "Applicant C"
- `option_2` (value: None): "Applicant A"
- `option_3` (value: None): "Applicant G"
- `option_4` (value: None): "Applicant H"
- `option_5` (value: None): "Applicant F"
- `option_6` (value: None): "Applicant B"
- `option_7` (value: None): "Applicant E"
- `option_8` (value: None): "Applicant D"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 1.0, 'option_4': 0.0, 'option_5': 0.0, 'option_6': 0.0, 'option_7': 0.0, 'option_8': 0.0}

G alone has membership, orientation and no overdue equipment.

## synthetic-choice-099

### Instruction

```json
"Which item is still missing?"
```

### State

```json
{
  "packed": [
    "tent",
    "map",
    "water"
  ],
  "required": [
    "tent",
    "water",
    "torch",
    "map"
  ]
}
```

### Criteria

- `option_1` (value: None): "Map"
- `option_2` (value: None): "Water"
- `option_3` (value: None): "Tent"
- `option_4` (value: None): "Torch"

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 1.0}

The torch is the only required item not packed.

## synthetic-choice-100

### Instruction

```json
"Pick the revision that removes only the unsupported claim while preserving the supported information."
```

### State

```json
{
  "draft": "The shop opened in 2010 and was founded by Nora.",
  "source": "The shop opened in 2010. The owner is not identified."
}
```

### Criteria

- `option_1` (value: None): "The shop was founded by two people."
- `option_2` (value: None): "The shop opened in 2020."
- `option_3` (value: None): "The shop opened in 2010 and won an award."
- `option_4` (value: None): "The shop opened in 2010."
- `option_5` (value: None): "Nora closed the shop in 2010."
- `option_6` (value: None): "The shop never opened."
- `option_7` (value: None): "Nora founded the shop."
- `option_8` (value: None): "The owner is certainly not Nora."

Target: {'option_1': 0.0, 'option_2': 0.0, 'option_3': 0.0, 'option_4': 1.0, 'option_5': 0.0, 'option_6': 0.0, 'option_7': 0.0, 'option_8': 0.0}

Only the opening year is supported; the founder is unknown.