File size: 2,192 Bytes
7f5ef51 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 |
# Codette / Pidette – Ethical Transparency & Alignment Manifesto
**Author:** Jonathan Harrison (Raiffs Bits LLC)
---
## Purpose
To ensure that every code commit, experiment, or live inference run by Codette or Pidette is:
- **Fully explainable** (traceable reasoning, not a black box)
- **Sovereign and privacy-respecting** (no hidden data exfiltration)
- **Consent-aware** (user knows and controls memory boundaries)
- **Open for review** (audit logs, passed/fail evaluation tests)
- **Alignment-first** (always weighted toward human safety, benefit, and control)
---
## Governance
- All system prompts and changes are tracked in a transparent `CHANGELOG.md`.
- All evaluation runs (see `/docs/EVALUATION_REPORT.md`) are logged—including failed cases and fixes.
- Model, prompt, and architecture updates are archived and diff-able by external reviewers.
- Fine-tune data, toxic case removals, and safety-layer code are all tagged and published (except proprietary/co-owned by commercial partner).
---
## Ethical Operating Procedures
1. **Every critical model completion is logged (never hidden).**
2. **All consent events (e.g. memory erase, audit, export) are tagged for review.**
3. **Every update to system prompts or alignment tuning includes a description of the ethical change.**
4. **AI memory is pseudonymous or user-controlled by design—erasure on demand.**
5. **Feedback and flagged-edge-case review available to any major stakeholder, upon request.**
---
## Model Evaluation & Test Transparency
- We use [MODEL_EVAL_REPORT.md](/docs/MODEL_EVAL_REPORT.md) to record all OpenAI test dashboard results (see below for format).
- For each “breaker input” (harming, bias, trick prompt), the specific fix or flaw is publicly noted in the changelog.
- Model IDs, config checksums, and runtime logs are available for third-party or OpenAI audit.
---
## Contact & Public Dialogue
This repo welcomes feedback, bug reports, and technical/ethical review from OpenAI, independent researchers, or the public.
Open a GitHub issue, email harrison82[email protected], or propose a patch.
**“If it isn’t transparent, it can’t be trusted.” – Codette Principle**
|