- Use at least 2 independent sources, including at least 1 credible source not on the track list.

- Minimum volume: 50,000 rows, or 2 years of hourly data, or 5,000 labelled documents or images.

- Record every raw file in manifest.csv with source, query, fetch time, SHA-256 and licence.

Not allowed

- Files whose only source is a re-upload on Kaggle, GitHub or a blog, with no original publisher

- Synthetic or AI-generated data for training or evaluation (stress tests only, clearly labelled)

- Personal, patient or other identifiable data

- Data collected against a site's terms of use or robots.txt

- Datasets so small or so clean that there is nothing to engineer

Generative AI rules

1. No number from the LLM. Every number comes from a gold table and the insight stores the row IDs.
2. No claim without a source. Every recommendation cites a trusted document and section.
3. No synthetic data in evaluation.

Conduct

- Outputs are information, not professional advice. Health outputs must point users to official services in an emergency.

- AI assistants are allowed. Disclose material use and be able to explain all your code.

- Acknowledge all external models, APIs, libraries, datasets and code, and follow their licences and rate limits.

- Do not share code, data or outputs with other teams during the event.

- Label anything simulated or unfinished.

Disqualification

Hard-coded outputs, faked logs, leaked test data or misrepresentation.

Deadlines (PKT)

- Hackathon opens: Mon 19 Oct, 12:00 AM

- Hackathon closes: Wed 21 Oct, 11:59 PM

Check the official channel at least every 4 hours whileawake.