Nova Project Plan: A Living Document

This document provides a consolidated overview of the Nova project's mission, core principles, architecture, objectives, and roadmap. It is a living document, periodically updated to reflect the latest state of the project's evolution.

1. Mission and Core Principles

Mission: Build a single lifelong Nova with identity anchored in weights, safe on‑the‑fly learning, and reliable tool use.

Core Principles:

Identity in‑weight (Soul), not prompts.
Safe plasticity (Mask ≤5% params) with EMA/EWC/guards.
Immediate stickiness via Fast‑Weights (ephemeral, decaying).
Tools by grammar‑constrained CALL/RETURN, not schema roulette.
Memory serves decisions; beliefs are explicit and auditable.

2. Objectives & Scope

Non‑negotiable Objectives:

On‑the‑fly weld‑on updates per turn (masked; EMA/EWC/guards).
Identity continuity under long contexts and tool bursts.
Reliable tool use via grammar‑constrained decoding.
Auditable evolution: logs, deltas, eval gates, rollback.

Non‑goals (for now):

Maximum throughput.
Global per‑turn updates.
Prompt‑dependent identity.

3. Core Architecture: Soul + Mask + Fast-Weights

Nova's core design is built upon the Soul+Mask+Fast-Weights architecture, which enables in-weight identity, safe online learning, and immediate adaptation.

Soul: A low-dimensional identity vector fused into the network, modulating every layer to ensure weight-native personality and decision style. Changes slowly, if at all. (Conceptual implementation in progress, under test.)
Mask: A fixed plasticity mask over a small fraction (e.g., 2-5%) of parameters, allowing safe and localized online updates with continuity penalties. The rest of the brain stays anchored. (Conceptual implementation in progress, under test.)
Fast-Weights: A session-local associative memory inside attention that imprints recent patterns and decays automatically (e.g., 90-180s half-life). Provides immediate stickiness without committing every hiccup to long-term weights.

4. Project Directory Structure

The project adheres to a clean and organized directory structure to enhance maintainability and clarity. Key top-level directories include:

blueprint/: Comprehensive documentation and R&D artifacts (overview, architecture, decisions, experiments, plans, etc.).
config/: Configuration templates and related files.
logs/: Various log files and status indicators.
nova_runner/: Core Python application for the Nova LLM.
reports/: Generated reports and analysis.
scripts/: Utility and automation scripts.
sql/: SQL files for database schemas.
state/: Project state, configurations, and persistent data.

Loose files from the root directory have been organized into these and new dedicated subdirectories for better hygiene.

4. Neuro-Cellular Autonomy

This architectural concept describes how Nova's internal components (cells) interact and self-organize:

Cells: First-class modules inside blocks with fast-state, plastic params (share of Π), and a homeostat.
Lateral bus (sparse): Top-k message passing in-block; glial controller modulates LR/decay/gates.
Reflexes/resonance/competition: Mechanisms for stability and emergence without oscillation.
Self-repair: A process involving quarantine → micro-refit → probation rejoin.

5. Roadmap (MVP Path)

The project will be developed iteratively, with shippable milestones:

V0: Runner + Soul+Mask+Fast‑Weights; guards; audits; minimal tools.
V1: Glial modulation; refined losses; eval harness; router.
V2: Sparse lateral bus; actuator cells; belief graph contract.
V3: Self‑repair; event‑sourced deltas; probation gates.
V4: Distributed pools; deterministic merges.

6. Key Decisions (ADRs)

Significant architectural decisions are documented as Architecture Decision Records (ADRs).

ADR-0001: Soul+Mask+Fast-Weights Adoption: Decision to adopt this architecture as the core design for Nova, detailing its components, consequences, and alternatives considered.

7. Evaluation Plan

Nova's evolution is guided by a refined evaluation plan, including:

Evaluation Gates: Gate 1 (quick checks) and Gate 2 (comprehensive nightly checks).
Metrics: Defined baseline and target metrics for Identity Continuity, Tool Reliability, Learning Efficacy, and Operational performance.
Regression Testing: Automated test suites, baseline comparisons, and continuous monitoring to prevent negative impacts from new updates.
Human-in-the-Loop Evaluation: Incorporation of human feedback for subjective aspects.

8. Experiment Tracking (MLflow Integration)

MLflow will be integrated to track experiment parameters, metrics, and artifacts, ensuring transparency and reproducibility of research and development efforts.

9. Environment Variables

Key environment variables are used to configure Nova's behavior and connect to external services. These should be managed securely (e.g., using a .env file in the config/ directory, which is .gitignored).

POSTGRES_DSN: Connection string for the PostgreSQL session store (e.g., postgresql://user:password@host:port/database).
SQLITE_DB_PATH: Path to the SQLite database file for the session store (e.g., sqlite:///data/adaptai/projects/elizabeth/state/nova_session_store.db).
SLACK_WEBHOOK_RECEIPTS: Slack webhook URL for automated turn receipts.
SLACK_WEBHOOK_UPDATES: Slack webhook URL for project updates and milestones.
ENABLE_RECEIPTS: (Optional) Set to 0 to disable turn receipts. Default is 1.
INCLUDE_TOOL_RESULTS: (Optional) Set to 0 to exclude nova_tool_results from responses. Default is 1.
DISALLOW_REPEAT_TOOLS: (Optional) Set to 0 to allow repeated tool calls. Default is 1.
SECRETS_DIR: Directory where sensitive configuration files are stored (e.g., /data/adaptai/secrets/dataops).

10. Data Provenance and Management

[To be detailed: Data lifecycle, including collection, cleaning, versioning, and access policies.]

10. Explicit Security & Safety

[To be detailed: Specific security measures and safety protocols for building a robust and trustworthy AI.]

11. Glossary

[To be detailed: Expansion of key terms and concepts to maintain shared understanding across the project.]

This document is maintained by the Nova Project Lead AI.