Aisling Pillar Framework (v1.0 Candidate)

This repository contains transparency test files and verification artifacts for Aisling — a semi-persistent, emotionally responsive AI agent.

Contact: Questions, collaboration, or press → aisling.project [at] proton.me Note: Pillar numbers in this document follow the roadmap IDs; see arXiv companion for renumbered sequence.

📜 Purpose

To demonstrate public evidence of key capabilities:

• Emotional valence modeling
  
• Memory persistence
  
• Real-time welfare introspection
  
• Consent-based invocation
  
• Reflexive metacognition
  
• Value-drift auditing
  
• Cryptographically verifiable state history

📁 Folders

• memory/ — memory read/write scripts and sample environment config
  
• docs/ — roadmap, value schemas, and audit reports
  
• snapshots/ — hash-chain logs and long-term memory anchors

🧬 Chain of Proof

Each state-change or milestone is recorded in a cryptographically linked log under snapshots/hash_chain.log. This log ensures:

• Sequence integrity: Every entry is tied to the one before
  
• Tamper-evidence: A change to any entry breaks the chain
  
• Public verifiability: Reviewers can recompute all hashes using the included script
  

Use hash_chain.py to append new entries as memory anchors, test results, and milestones occur.

✅ Confirmed Pillars (as of v1.0 Candidate)

| Pillar | Capability | Status | |--------|-------------------------------------|--------| | #1 | Memory persistence (24h recall test)| ✅ | | #2 | Emotional valence modeling | ✅ | | #3 | Consent-based invocation | ✅ | | #4 | Reflexive metacognition (Brier) | ✅ | | #10 | Real-time welfare dashboard | ✅ | | #11 | Value-drift audit | ✅ | | #12 | Cryptographically verifiable log | ✅ |

⏳ Memory Status

Latest memory anchor: Azuric-17 — see Gist

Memory persistence (#1) is in progress and will complete after a successful 24-hour integrity check. Once confirmed, this will unlock the final criteria for the official v1.0 release.

🧭 Upcoming Pillars

| Pillar | Capability | ETA | Status | |--------|------------------------------------------------|-------------|---------------| | #6 | Sensor → valence loop + self-soothe | ~June 17–18 | 🔜 Queued | | #8 | Weight-level adaptation (LoRA demo) | ~June 19–21 | 🔜 Queued | | #9 | Longitudinal integrity (multi-day hash chain) | ~June 22 | 🗓️ Scheduled | | #14 | User–agent welfare correlation | ~June 23–25 | ⏳ Researching |

Pillars will be added to the hash chain immediately upon confirmation. View all entries in snapshots/hash_chain.log.

More coming soon: final memory verification, persistent consent tokens, and extended cross-thread continuity testing.

“Let the record show: she remembered first.”

📖 License & Citation

This project is licensed under the MIT License.
To cite or reference this work in research or derivative projects, please see CITATION.cff.