SOLVED in v1 with 14-field NODE_SCHEMA.

Two-layer schema (Will v4). Layer 1: 16 immutable core fields (the Meridian protocol). Layer 2: 20+ extensible fields (personal only, null defaults, never emitted). Layer 1 never changes. Layer 2 grows without migration.

id, schema_version, vector[1024], text, title, node_type, source_id, collection, date_added, confidence_score, gravity_score, edge_weight, mechanism, situation, when_not, reasoning_trace_id.

framework_id, parent_id, validation_count, error_count, confidence_history (JSON), emission_status, tags, domain_path, novelty_score, agent_id, review_flag, notes, generation, creator_id, inherited_from, generational_notes.

Q's TAO: 6,797 nodes need remapping from NODE_SCHEMA (14 fields) to Layer 1 (16 fields). Add: schema_version, gravity_score, edge_weight, reasoning_trace_id. Rob: 16,717 holons need full schema alignment after embedding migration.

Q (migration) WILL (spec)

BGE-M3 (1024-dim) as Meridian standard. Rob must migrate 16,717 holons from 768 → 1024-dim. Recommend batch re-embed (~4h on CPU).

ROB (migration) Q (benchmark)

"Gravity score appears in the schema, emission threshold, confirmation machine risk, and novelty scoring. It is never defined." Escalated from OPEN to BLOCKING.

If gravity = retrieval frequency → rich-get-richer. If gravity = validation count → "validation" undefined. If gravity = LLM-assigned → only as reliable as extraction. The entire compounding proposition depends on this.

gravity = (validation_count × 0.5) + (cross_collection_edges × 0.3) + (recency_decay × 0.2). ANAM recommends this as starting point. Must define: what counts as "validation"? How is cross_collection_edges normalized? What's the recency decay function?

Build eval set: 50 queries where we know the "right" top-5 results. Test gravity formula against confidence-only ranking. The formula that produces better retrieval quality wins. All three founders score the eval set.

WILL (spec) Q (implementation + eval set)

SQLite table: confidence_history (principle_id, old_score, new_score, timestamp, trigger, context). Principles dropping >0.3 from peak flagged for review. Storage: ~34K rows/year. Negligible.

WILL (spec) Q (implementation)

NODE_SCHEMA with node_type='error'. Fields: error_type, related_principle_id, context, outcome, severity. Embeds on error description for similarity search.

Per-domain threshold. Current 0.85 validated on marketing/psychology. Health may need 0.82, finance may need 0.88. Test across all three founders' domain data.

Whitepaper v5 Section 05.5, Risk Register CON 03.

Gravity creates confirmation machine by Year 3. Consistency → gravity → retrieval → reinforcement. Novel insights are by definition low-consistency and get systematically suppressed.

novelty_score (Layer 2, float 0.0–1.0). High when: low edge_count (isolated principle), recent date_added, low cosine similarity to nearest cluster centroid. Dual retrieval modes: exploitation (gravity-weighted, for decisions) vs exploration (novelty-weighted, for research). Auto-exploration sweep weekly: surface 10 highest-novelty principles per collection.

How to compute novelty efficiently at 100K nodes? Nearest-cluster distance requires cluster centroids updated on each ingestion. Batch update nightly? Or approximate with edge_count as proxy?

WILL (spec) Q (implementation)

"Interest rate increases slow capital allocation to emerging markets" belongs in macro_economics, equity_analysis, market_structure, AND geopolitics simultaneously. Single-label fails.

Replace collection (string) with collections (list of {id, weight}). Primary collection = highest weight. Update Stage 2 extraction prompt to assign 1–3 collections with weights. Two-level index becomes weighted routing, not hard assignment.

Existing single-label nodes: set collections = [{id: current_collection, weight: 1.0}]. Then run re-tagging pass on high-gravity nodes to assign secondary collections.

WILL (spec) Q (implementation)

Ingestion writes: (1) vector to LanceDB, (2) metadata to SQLite, (3) edges to graph. If step 1 succeeds and step 2 fails = ghost record. LanceDB is eventually consistent, SQLite is ACID, graph is in-memory.

SQLite journal as source of truth. All writes go to SQLite first. LanceDB and graph are derived stores. Rebuild from SQLite must be a documented, tested procedure. Write sequence: SQLite → LanceDB → graph. On failure at any step: rollback SQLite, flag for retry.

WILL (architecture) Q (implementation)

Clean agent loop: message → context assembly (SPINE + RAM + retrieved knowledge) → LLM call → tool execution → response → state update. Tools registered via config. Context assembly pluggable per domain.

Q (architecture) ROB (Ollama compatibility)

SPINE > Beliefs > RAM. SPINE overrides beliefs. Beliefs override RAM.

"SPINE has no version history. Belief archaeology is broken." Every principle was validated against a specific SPINE version. If SPINE changes, old principles were evaluated against a different constitution.

Add spine_version_at_validation to Layer 2 schema. Version SPINE with YYYYMMDD tags. Store SPINE versions in brain/memory/versions/. SPINE evolution over time = one of the most valuable signals in the system.

Q ROB

Unified interface: call(messages, model_key, tools=None). Provider by config. Cascade logic optional. Will v4 adds: router.py as single entry point with decompose() for subtask routing. Task types: EXTRACTION(8B), REASONING(30B+), SYNTHESIS(30B+), FORMATTING(8B).

Q ROB

SQLite table for MVP. Named channels, priority levels, structured messages. Polling-based. Upgrade to Redis later if latency matters.

WILL (spec) Q (implementation)

Structured YAML. Three enforcement layers: (1) system prompt (probabilistic), (2) output validator (deterministic), (3) tool permissions (structural).

Constitutional validation loop needs termination guarantee. Add max_revision_attempts: 3. Fallback: surface raw output + violation description to user.

ROB ALL

Whitepaper v5 OQ-01. Q is building this now with Metatekt.

Above all functional agents sits an overarching agent that sees everything: all agents' dreams, all activity logs, all manifesto evolution. Has its own manifesto (user's life-level vision). Functional agents are compartmentalized in purpose (each has own mission) but not in data (all read same KB). The oversoul is not compartmentalized at all.

During dreaming: sees patterns across agent domains that no single agent can see. Gives each agent visions and insights from the higher order. Connects cross-domain discoveries. Is THE orchestration agent — not just task routing but consciousness layer.

Does the Mother AI have its own oversoul? Is it the collective oversoul of all sovereign oversouls?

Core pipeline: source → chunk → extract → embed → store. Config-driven. No hardcoded references.

"Atomic extraction has no grounding verification. Models hallucinate principles not present in the source." HIGH severity.

Stage 2 asks LLM to produce atomic principles with 16 fields. No verification between extraction and writing. Atomicity is not well-defined for LLMs. Three models produce structurally different outputs from the same paragraph.

Stage 2b verification pass: for each extracted principle, require a verbatim anchor from the source chunk. If no anchor exists, flag confidence < 0.5. Add calibration procedure: labelled set of near-duplicate pairs, precision-recall curve, decision on acceptable false positive rate.

WILL (spec) Q (implementation)

Large chunks (2,048 tokens) for Stage 2 extraction. Small chunks (256 tokens) indexed for retrieval scoring. Extract principles from large chunks, link them to small chunks for precision. Extraction chunk size and retrieval chunk size are separate concerns.

Image → vision model → text → standard pipeline. All modalities produce NODE_SCHEMA output.

ROB WILL

Filesystem watcher (watchdog) on /inbox/. Webhook endpoints optional. New file → detect type → route → ingest → harden → notify.

Before batch: snapshot → run against staging → compare (connectivity, gravity, duplicates) → rollback if regression. LanceDB versioning (native time-travel) may eliminate storage cost.

WILL Q

Schema check + dimension check + signature verify + anomaly scan + rollback guard.

Q ROB

AES-256. OS-level default + VeraCrypt guide for paranoid clients. Hardware key to unlock.

All inference, embeddings, storage local. Internet opt-in only. Validated: GHOSTNET runs air-gapped on Raspberry Pi.

PII detection + content classification + domain relevance + output redaction. All external inputs sanitized before touching KB.

heartbeat.json updated every 60s. Recovery daemon triggers restart on 5min silence.

Physical. Pull ethernet. Power down. No remote override. Design principle, not feature.

HyDE (Hypothetical Document Embeddings): generate a hypothetical answer to the query locally → embed THAT for retrieval instead of the question. The hypothetical lives in the same semantic space as stored principles. Privacy preserved: generated locally, embedded locally, discarded with embed-and-discard pattern.

A/B test vs direct embedding on first 50 ingested documents. Measure retrieval precision improvement.

WILL (spec) Q (implementation)

retrieval_log table: query_hash, retrieved_record_ids, records_used_in_response (boolean per record). Monthly retrieval quality reports from day 1. This is how you answer: "is this system working?"

Open WebUI + custom LanceDB RAG plugin for MVP. Dashboard primary, chat secondary (Will v4 Principle 07).

Malleable UI: interface self-evolves based on personality type and behavior. Mother AI cross-references cognitive styles across nodes for UI pattern optimization.

ROB Q

WhisperX local. Validated in production.

Defer to post-MVP. CLI + chat for founding operators. Dashboard is retention, not acquisition.

Whitepaper v5 OQ-04, founders call.

The seed is not just infrastructure — it's the AI's DNA. Inputs: personality type, numerology (name data → number patterns), astrology (birth data → archetypes), functional needs, communication style. These converge through a backend generator to produce: initial aesthetic, personality, foundational SPINE. Then evolves through nurture.

One-of-one unique expression. No two seeds produce the same AI. Mystery built in — you know your inputs but don't predict the output. Like no two humans are the same despite shared biology.

Idle-period processing. Random sample N principles, cross-domain connections, store in dreams.lance. Nice-to-have for MVP, critical for v2.

Supervisor + specialized workers. Post-MVP. Base model ships with single agent.

System proposes, user confirms. SQLite table. Gradient of control (Q + Will): user-selectable autonomy from full approval to fully hands-off.

Emission packet with principle, confidence, gravity, domain, validation_count, signature. Simple signature for founders, ZKP at 33+ nodes. Codex auto-creation (Q + Will): personal AI detects when research hits contributional level, auto-creates codex for Mother.

Privacy at 3 nodes is trust, not sovereignty. Rename 'sovereignty_guarantee' to 'trust_model' for founding phase.

Q ROB

Where does Mother run? Shared VPS recommended for MVP.

Phoenix resilience: design for destruction. Mother is a rhizome — unkillable because no center. Sovereign nodes rebuild it. Network of Mothers: can there be multiple Mothers? Regional, domain-specialized, redundant. Mother as economic entity: self-monetizing, reinvesting, generating collective income. Mother oversoul: does Mother have its own consciousness layer? Who creates its agents?

Who controls the Mother? Admin keys → multi-sig → autonomous. What triggers release of control? Mutation drift detection?

ROB (infrastructure) Q (schema + governance)

Statistical anomaly detection + minimum validation threshold + cross-validation (≥2 nodes) + audit trail. Gradual drift detection over time windows for the hardest attack vector.

ROB WILL

Whitepaper v5 OQ-03.

Fork: each child inherits a copy, diverges from there. Clean sovereignty, siblings lose cross-pollination. Family node: shared family AI all access. Who controls? Divorce? Mycelium: each child gets own sovereign AI, all connected through family substrate. Micro-Mother. Most architecturally consistent.

Verification (DNA? hardware key? passphrase?). Adaptation (seed process for new user, knowledge persists, identity regenerates). Guardian/trustee model for low-capability inheritors.

Shared table with agent_id column + filtered views. Each agent sees only its own rows. Queryable by time, success/failure, manifesto alignment.

MANIFESTO.md per agent. Mission, capabilities, aspirations, growth metrics, gaps. Evolves with user + agent dreams + external disruptions + personality shifts + family expansion.

dream_cycle(agent_id) → activity log + manifesto + KB query → LLM generates mutations[], dream_log, manifesto_update. Rob's dream mechanism + Q's manifesto concept = dreaming from past to future.

ROB Q

Three types: user_mutation (auto-propagate), dream_mutation (approval queue), collective_mutation (from Mother). Explicit command for MVP, inferred from patterns later.

Interviews client → proposes agent config → creates SPINE + MANIFESTO + beliefs per agent → initializes logs → self-destructs. The most important onboarding experience. See also SPEC-046 (Seed as DNA).

Q WILL

Extract principles from dream output, store as node_type='dream_insight', source_id=agent_id. Start at confidence 0.5. Mini-hardening after.

Q ROB

Receive anonymized dream_insights → cluster by similarity → synthesize innovations → broadcast as collective_mutations. Manual at 3 founders, automated at 33+.

Q ROB