DigitalQuill Community Model

Purpose: collect safe, consented examples for teaching small open-weight models the DigitalQuill Trinity: Outcome, Agency, and AI Hygiene.

Status: foundation only. No model training is happening yet.

TLDR

This folder is the starting point for a future small-scale fine-tuning experiment.

The immediate goal is not to train a model. The immediate goal is to collect clean examples, build evals, and protect contributors from accidentally submitting private data.

The DigitalQuill Trinity

Every example should teach the model to ask:

1. Outcome: what useful, testable result are we moving toward?
2. Agency: does this make the human clearer, safer, more capable, or more in control?
3. AI Hygiene: are context, scope, privacy, verification, tools, and cost being handled responsibly?

What This Folder Contains

• CONTRIBUTION_RULES.md: what contributors can and cannot submit.
• DATA_SAFETY.md: privacy, consent, and review rules.
• templates/trinity-example-template.md: safe structured example format.
• dataset/seed_trinity_examples.jsonl: tiny seed dataset for future fine-tuning.
• evals/trinity_eval_cases.jsonl: tiny starter eval set for measuring model behavior.

What Comes Later

Do not fine-tune until these exist:

• a redaction scanner
• 20-50 reviewed seed examples
• a real eval runner
• a chosen base model and license review
• a written hardware/training plan

Safe Contribution Rule

Do not submit raw private chats. Submit structured, fictionalized, consented examples only.

If an example contains private data, credentials, private customer details, medical/legal/financial secrets, or copyrighted long-form text the contributor does not own, it does not belong here.