---
pretty_name: procedural-jev
language:
- en
license: apache-2.0
task_categories:
- text-classification
tags:
- tasksource
- jev
- system-one
- procedural
- synthetic
- multi-question
dataset_info:
- config_name: entity_belief_tracking
  features:
  - name: id
    dtype: string
  - name: level
    dtype: int32
  - name: state
    dtype: string
  - name: questions
    dtype: string
  - name: answers
    dtype: string
  - name: world_location
    dtype:
      class_label:
        names:
          '0': desk
          '1': locker
          '2': archive
          '3': lab
  - name: agent_belief_location
    dtype:
      class_label:
        names:
          '0': desk
          '1': locker
          '2': archive
          '3': lab
  - name: belief_matches_world
    dtype:
      class_label:
        names:
          '0': 'false'
          '1': 'true'
  splits:
  - name: train
    num_bytes: 39699742
    num_examples: 20000
  - name: validation
    num_bytes: 1988456
    num_examples: 1000
  - name: test
    num_bytes: 1976442
    num_examples: 1000
  download_size: 3163050
  dataset_size: 43664640
- config_name: event_state_reconstruction
  features:
  - name: id
    dtype: string
  - name: level
    dtype: int32
  - name: state
    dtype: string
  - name: questions
    dtype: string
  - name: answers
    dtype: string
  - name: current_owner
    dtype:
      class_label:
        names:
          '0': alice
          '1': bob
          '2': carol
          '3': unassigned
  - name: is_open
    dtype:
      class_label:
        names:
          '0': 'false'
          '1': 'true'
  - name: current_severity
    dtype:
      class_label:
        names:
          '0': Routine.
          '1': Degraded service requiring attention.
          '2': Critical user-blocking incident.
  splits:
  - name: train
    num_bytes: 34904934
    num_examples: 20000
  - name: validation
    num_bytes: 1747994
    num_examples: 1000
  - name: test
    num_bytes: 1738194
    num_examples: 1000
  download_size: 2525516
  dataset_size: 38391122
- config_name: evidence_sufficiency
  features:
  - name: id
    dtype: string
  - name: level
    dtype: int32
  - name: state
    dtype: string
  - name: questions
    dtype: string
  - name: answers
    dtype: string
  - name: claim_supported
    dtype:
      class_label:
        names:
          '0': 'false'
          '1': 'true'
  - name: has_conflict
    dtype:
      class_label:
        names:
          '0': 'false'
          '1': 'true'
  - name: strongest_support_origin
    dtype:
      class_label:
        names:
          '0': S1
          '1': S2
          '2': S3
          '3': S4
          '4': S5
          '5': S6
          '6': S7
          '7': none
  splits:
  - name: train
    num_bytes: 37800061
    num_examples: 20000
  - name: validation
    num_bytes: 1885368
    num_examples: 1000
  - name: test
    num_bytes: 1883591
    num_examples: 1000
  download_size: 2904366
  dataset_size: 41569020
- config_name: multi_view_adjudication
  features:
  - name: id
    dtype: string
  - name: level
    dtype: int32
  - name: state
    dtype: string
  - name: questions
    dtype: string
  - name: answers
    dtype: string
  - name: intent
    dtype:
      class_label:
        names:
          '0': billing
          '1': access
          '2': technical
          '3': other
  - name: is_urgent
    dtype:
      class_label:
        names:
          '0': 'false'
          '1': 'true'
  - name: workflow_impact
    dtype:
      class_label:
        names:
          '0': Minor inconvenience; normal work can continue.
          '1': Important workflow is degraded but a workaround exists.
          '2': Core work is blocked and no workaround is available.
  splits:
  - name: train
    num_bytes: 64029482
    num_examples: 20000
  - name: validation
    num_bytes: 3200754
    num_examples: 1000
  - name: test
    num_bytes: 3198300
    num_examples: 1000
  download_size: 4527531
  dataset_size: 70428536
- config_name: partial_observation_calibration
  features:
  - name: id
    dtype: string
  - name: level
    dtype: int32
  - name: state
    dtype: string
  - name: questions
    dtype: string
  - name: answers
    dtype: string
  - name: incident_real
    dtype: float32
  splits:
  - name: train
    num_bytes: 17712369
    num_examples: 20000
  - name: validation
    num_bytes: 924994
    num_examples: 1000
  - name: test
    num_bytes: 922014
    num_examples: 1000
  download_size: 1237474
  dataset_size: 19559377
configs:
- config_name: entity_belief_tracking
  data_files:
  - split: train
    path: entity_belief_tracking/train-*
  - split: validation
    path: entity_belief_tracking/validation-*
  - split: test
    path: entity_belief_tracking/test-*
- config_name: event_state_reconstruction
  data_files:
  - split: train
    path: event_state_reconstruction/train-*
  - split: validation
    path: event_state_reconstruction/validation-*
  - split: test
    path: event_state_reconstruction/test-*
- config_name: evidence_sufficiency
  data_files:
  - split: train
    path: evidence_sufficiency/train-*
  - split: validation
    path: evidence_sufficiency/validation-*
  - split: test
    path: evidence_sufficiency/test-*
- config_name: multi_view_adjudication
  data_files:
  - split: train
    path: multi_view_adjudication/train-*
  - split: validation
    path: multi_view_adjudication/validation-*
  - split: test
    path: multi_view_adjudication/test-*
- config_name: partial_observation_calibration
  data_files:
  - split: train
    path: partial_observation_calibration/train-*
  - split: validation
    path: partial_observation_calibration/validation-*
  - split: test
    path: partial_observation_calibration/test-*
---

# procedural-jev

Procedurally generated decision problems. Each row is one structured state
(JSON) with **several typed questions over that same state**, following the
Jev / System One request shape: `choice` (pick one criterion), `noul` (a
number in [0, 1]; a probability or a yes/no), and `score` (an ordered rubric).
Every answer is computed exactly from the state by rules that the state
itself spells out, so the labels are noise-free.

This is an independent dataset. It is not an official TypeSafe Jev dataset and
is not produced by or affiliated with TypeSafe or OpenJev.

## Configs

| config | questions |
|---|---|
| `entity_belief_tracking` | `world_location` (choice), `agent_belief_location` (choice), `belief_matches_world` (noul) |
| `event_state_reconstruction` | `current_owner` (choice), `is_open` (noul), `current_severity` (score) |
| `evidence_sufficiency` | `claim_supported` (noul), `has_conflict` (noul), `strongest_support_origin` (choice) |
| `multi_view_adjudication` | `intent` (choice), `is_urgent` (noul), `workflow_impact` (score) |
| `partial_observation_calibration` | `incident_real` (noul, exact Bayesian posterior) |
| `policy_applicability` | `access_allowed` (noul), `governing_policy` (choice), `review_risk` (score) |
| `state_perturbation` | `material_change` (noul), `changed_dimension` (choice), `risk_direction` (score) |

## Schema

| field | meaning |
|---|---|
| `id` | `task:split:index` |
| `level` | Difficulty level (0–4); larger levels add events, records, sensors, or distractors. |
| `state` | The state, as a JSON string. |
| `questions` | JSON object of named System One questions (`type`, `instructions`, `criteria`). |
| `answers` | JSON object of reference answers, in the System One `answers` shape. |
| one column per question | Flat label: a `ClassLabel` for choice, score, and yes/no noul questions; a float for graded noul (`incident_real`). |

States are unique within a split, and validation/test states never occur in
train.

## Use

As a multi-question Jev request, send `{"state": json.loads(row["state"]),
"questions": json.loads(row["questions"])}` and compare with `row["answers"]`.
The same rows are included, grouped by state, in
[`tasksource/tasksource-jev`](https://huggingface.co/datasets/tasksource/tasksource-jev).

Each question is also a Tasksource task:

```python
from tasksource import load_task
load_task("procedural-jev/policy_applicability/review_risk")
```

## Reproduction

Generation is deterministic (row `i` of a split is seeded by `task:split:i`).
From a [tasksource](https://github.com/sileod/tasksource) checkout:

```bash
PYTHONPATH=.:src python jev/build_procedural.py --output build/procedural-jev --upload
```

Generators live in `src/tasksource/jev/procedural/`.