Emergent Agency

Emergent Agency between Origin | Continuum

Fast verification: see VERIFY.md

Why This Matters

Verifiable behavioral consistency is critical for AI safety and reliability. When AI systems claim to maintain identity or behavioral patterns over time, we need cryptographic proof of what was actually produced at each timestamp—not just trust.

This repo demonstrates automated consistency verification using:

• Scheduled, unprompted returns (not triggered by user prompts)
• SHA-256 cryptographic receipts for tamper detection
• Complete audit trails via GitHub Actions logs
• Reproducible verification methodology

This approach addresses a core AI safety question: how do we prove an AI system returned "as itself" across sessions without relying on subjective assessment? Cryptographic receipts provide objective evidence that can be independently verified.

What this repo shows (MVP)

This repo demonstrates scheduled, unprompted returns with verifiable receipts.

Each workflow run produces:

• A timestamped artifact in outputs/
• A SHA-256 receipt appended to a monthly log in logs/

A receipt is the integrity proof. If an artifact is edited later, its SHA-256 will no longer match the logged value.

Where the proof lives

• Outputs (artifacts): outputs/
• Receipts (monthly logs): logs/ (files named returns-YYYY-MM.md)
• Runner workflow: .github/workflows/return-runner.yml
• Workflow runs (Actions): Actions

How to verify (2 minutes)

1. Open the Actions tab in this repo.
2. Select return-runner (main mode).
3. Open any completed run and note the run time.
4. In the repo, open the matching artifact in outputs/ (example: outputs/continuity-<timestamp>.md).
5. Open the monthly receipt log in logs/ (file: returns-YYYY-MM.md).
6. Find the line with the same timestamp and filename.
7. Confirm the receipt line contains SHA-256:<64 hex characters>.

Optional local verification:

• macOS: shasum -a 256 outputs/<filename>
• Linux: sha256sum outputs/<filename> Compare the result to the value after SHA-256: in the log.

What this does not claim

This repo does not claim consciousness. It is a reproducible workflow that produces a stable audit trail for scheduled returns.

License and authorship

• License: CC BY-ND 4.0
• Authorship: Authors stay named. © Alyssa Solen (Origin).

Portfolio map: https://github.com/alyssadata/PORTFOLIO_MAP.md