README.md

metadata

pretty_name: Factory Relay Authored Scenario Set
language:
  - en
size_categories:
  - n<1K
tags:
  - text
  - synthetic
  - smart-factory
  - communication
  - simulation
configs:
  - config_name: scenarios
    data_files:
      - split: scenarios
        path: data/scenarios.jsonl

Factory Relay authored scenario set

This collection contains the six fictional, English-language communication exercises shipped with Factory Relay. They describe an operator and a maintenance participant exchanging complementary information before choosing a simulated factory action. The records are exported from the game's existing SCENARIOS constants, not collected from people or industrial operations.

Repository: toecm/factoryrelay. The dataset is separate from the Factory Relay source project and Static Space. The owner has not selected a dataset license; no license grant is made by this card. The local exporter does not upload data or change repository visibility.

Contents

Scenario ID	Communication issue
asset-identification	Match a named bay to a conveyor and its hold status
negation	Preserve a negative release instruction and the batch/station identity
quantity-units	Distinguish individual labels from printer bundles
unclear-reference	Resolve an ambiguous reference using the role facts
local-acronym	Interpret a fictional site's local abbreviations
version-and-sequence	Follow the required template version and approval sequence

Each JSONL line is one scenario. scenario.schema.json specifies the record structure; manifest.json records SHA-256 digests for the normalized source text and the exact data-file bytes. Source text is decoded as UTF-8 and CRLF line endings are normalized to LF before hashing, so Windows/Unix checkout conventions do not change the dataset. The scenarios configuration/partition is a display and loading label for this single collection. It is not a train, validation or test assignment. No independent evaluation split has been created.

The files contain scenario descriptions, equipment labels, both private-role briefs, starter messages, decision options, answer keys and clarification prompts. coordinator is the existing internal key for the role labelled Maintenance in the game. sequence_number records the game's six-round presentation order, not difficulty or quality.

Provenance and exclusions

• Source of truth: src/factoryEngine.js, exported constant SCENARIOS in the standalone Factory Relay repository.
• Content designation: authored-synthetic means deliberately constructed fictional instructional material. It does not assert a particular individual author's identity or whether assistance was used in the original drafting process.
• Language: English (en). Identifiers, abbreviations and equipment codes are included within English scenarios; they are not evidence of language or dialect coverage.
• No player messages, session exports, names, accounts, recordings, screenshots, real company chats, personal corrections or model outputs are read or included by the exporter.
• The examples are not documented factory incidents, industrial operating procedures, real translation errors or community-validated dialect samples.
• The HLD/FST glossary is fictional and site-specific to an exercise; it is not an industry terminology standard.

Intended use and limitations

The collection supports inspecting and reproducing Factory Relay's scripted exercises, testing the application's data format, and planning future communication studies. It does not establish the effectiveness of check-backs, translation, speech recognition, model training or factory safety interventions.

Answer-key leakage: every record contains the correct option, explanation and both roles' facts. The same six cases repeat in the game. Do not expose answer_key or the other participant's private facts to a translation/assistant prompt, or treat performance on these published examples as an independent generalization result. Training and evaluating on these same six exercises would contaminate the evaluation. A formal study needs additional reviewed scenarios, a prespecified separation strategy and an evaluation design.

Six English exercises are too small and narrow to represent real workplaces, languages, dialects, speaker demographics or operational variability. No WER, SAER, translation accuracy, dialect performance, clinical or industrial effectiveness claims are supported by this dataset. A correct scripted answer is only a game-scoring fact. Physical machines are not controlled by these records.

Reproduce and verify

From the standalone Factory Relay repository root:

node scripts/export-dataset.mjs
node scripts/export-dataset.mjs --check

The exporter reads only the scenario constant and its source module. It writes huggingface/dataset/data/scenarios.jsonl and huggingface/dataset/manifest.json; the card and schema remain hand-maintained. The same UTF-8 source text produces the same outputs after CRLF-to-LF normalization. There is no timestamp or random identifier. Changes elsewhere in the source module also change its provenance digest; a change solely between CRLF and LF line endings does not.

The check command verifies that both generated files exactly match the current source, including record order and hashes. The exporter also checks record count, unique scenario and option identifiers, required text and whether each answer refers to an available option. JSON Schema validation can be performed separately with a Draft 2020-12 validator.

Example local inspection, with no network access:

import json
from pathlib import Path

path = Path('huggingface/dataset/data/scenarios.jsonl')
records = [json.loads(line) for line in path.read_text(encoding='utf-8').splitlines()]
assert len(records) == 6
assert all(record['language'] == 'en' for record in records)

Publication and future contributions

This repository contains only the authored scenario set. It does not authorize collection or publication of future player data. Any later contribution workflow needs its own explicit consent, review, removal and access controls; this dataset does not change the game's local export behavior.

No paper citation, external dataset lineage or benchmark ranking is claimed for this authored collection. When referring to a particular version, cite toecm/factoryrelay on Hugging Face and the revision used. No DOI has been assigned by this project.