The dataset viewer is not available because its heuristics could not detect any supported data files. You can try uploading some data files, or configuring the data files location manually.

AURORA-Workflow-1 — Rule-grammar dataset

The primary structured-workflow dataset for Stage-1 of the AURORA research programme. This repository hosts the rule grammars and dataset specification, not the materialised training data — the 60 000 episodes are generated programmatically from the grammars by the AURORA training pipeline.

What is in this repo

Path	Role
`spec.md`	Dataset specification (v1.0.0): six domains, splits, sample sizes, generation pipeline, manual-audit rules.
`domains/<slug>/grammar.md`	Per-domain rule grammar. Six domains: `invoice-triage`, `appointment-scheduling`, `inventory-reorder`, `lab-sample-routing`, `issue-ticket-escalation`, `household-maintenance-planning`.
`aurora-federated-1-spec.md`	Sister dataset (federated schema-exchange). Used by H6.

Why grammars, not raw events

AURORA-Workflow-1 is generator-defined, not collected. The grammars are seeded simulators; the same git-pinned grammar + the same seed always emits the same events. This makes the dataset:

Bit-exact reproducible across replication partners,
Schema-versioned (a grammar change is a dataset_version bump per the spec),
Storage-cheap (~92 KB instead of ~80 GB).

Any partner who wants to consume the materialised data downloads the grammars from this repo and runs the AURORA generator scripts.

How to use

git clone https://huggingface.co/datasets/Anthril/aurora-workflow-1 aurora-workflow-1
# Then in an AURORA checkout:
python scripts/generate-enriched-corpus.py \
    --grammar-dir aurora-workflow-1/domains/ \
    --output data/baselines/lora-llama-8b/<date>-enriched/

Headline numbers (per `spec.md`)

Property	Value
Domains	6
Workflows per domain	50
Episodes per workflow	200
Total episodes	60 000
Adversarial-exception rate	10 %
Temporal rule-replacement rate	20 %
Splits	train / validation / calibration (5 %) / test / OOD / compositional-holdout (20 %)
Seeds	100 master; 20 per Stage-1 experiment

Hypothesis coverage

Used by H1 (event-vs-token), H2 (sparse routing), H3 (continual learning), H4 (episodic-semantic memory), and benchmark families LDA, EUT, ESC, HUB.

Provenance

Source-of-truth path in repo: data/aurora-workflow-1/ and data/aurora-federated-1/spec.md.
Spec anchor: architecture/engineering-spec/training-methodology/training-fairness-controls.md §"Information equivalence".
Local SHA-256 manifest: see hf-publish-manifest.json in the source tree.
Dataset version: v1.0.0 (frontmatter of spec.md).