Architecture Proposal Knowledge Storage & Cross-Reference

How 805 patent documents become a navigable knowledge graph inside the Sov_OS knowledge ring.
Using existing nodes: a_library → a_dna → a_canon → a_index

§1   The Problem

The patent collection contains 805 markdown files with hash-chain provenance but zero semantic structure. Files are organized by iCloud archive origin, not by concept. There are no tags, no domain classifications, no cross-references between related patents. Implicit patent families exist (Logos, Kronos, Guardian Stones, SAGE) but they're invisible to any automated system.

The goal: ingest every document into the Sov_OS knowledge ring so that meaning becomes geometry — each patent gets a DNA helix encoding, a 12D shape tensor, vector embeddings, and rich metadata that enables cross-domain discovery by conceptual similarity rather than filename matching.

§2   The Pipeline

Four existing knowledge nodes form a pipeline. Each document flows through five stages — the same pipeline that processes any markdown entering the system:

Hard-Wire Topology

These nodes are already wired in knowledge.toml:

§3   Enriched Frontmatter Schema

The current patent files have only conversion metadata (source hash, chain index). The a_library node already parses FrontMatter with title, tags, created_at, updated_at. We extend this with domain-specific fields:

---
title: "Real-Time Semantic Drift Firewall"
domain: governance
system: sage
kind: provisional_patent
status: filed

tags:
  - semantic-firewall
  - hysteresis
  - authorization
  - drift-detection
  - schmitt-trigger

related:
  - guardian_stones/stone_1
  - sage/trust_coefficient
  - logos/semantic_identity

sov_concepts:
  - AuthState
  - Warrant
  - NarrativeMass
  - TrustDecay

layers:
  - L3-mesh        # where this lives in the shape stack
  - L4-topology    # what it governs

source_file: "Real_Time_Semantic_Drift_Firewall_Patent_Draft.md"
source_hash: "a1b2c3..."
chain_index: 42
created_at: 2026-03-16
updated_at: 2026-03-16
---

§4   Domain Taxonomy

Ten domains emerged from the patent corpus. Each maps to one or more Sov systems and rings:

§5   Storage Layout

Patent documents live inside the knowledge ring as a managed collection within a_library. The enriched frontmatter makes them queryable by every field. The physical layout:

storage/a_knowledge/
├── nodes/
│   └── a_library/
│       ├── library.toml              ← pipeline config (existing)
│       ├── crates/                   ← Rust pipeline code (existing)
│       └── vault/                    ← NEW: document storage
│           ├── _index.toml           ← master index (auto-generated)
│           ├── _domains.toml         ← domain → document mapping
│           ├── _graph.toml           ← cross-reference adjacency list
│           │
│           ├── identity/             ← domain directory
│           │   ├── logos/
│           │   │   ├── provisional_patent.md
│           │   │   ├── multi_modal_auth.md
│           │   │   └── voice_derived_object.md
│           │   └── wisp/
│           │       ├── continuation_provisional.md
│           │       └── semantic_identity_object.md
│           │
│           ├── governance/
│           │   ├── sage/
│           │   │   ├── semantic_firewall.md
│           │   │   ├── governed_access.md
│           │   │   └── trust_coefficient.md
│           │   └── providence/
│           │       └── constraint_enforcement.md
│           │
│           ├── temporal/
│           │   └── kronos/
│           │       ├── temporal_intelligence.md
│           │       ├── trust_decay_functions.md
│           │       └── kronos_spiral.md
│           │
│           ├── security/
│           │   └── guardian_stones/
│           │       ├── stone_1_semantic_firewall.md
│           │       ├── stone_2_behavioral_drift.md
│           │       └── ...
│           │
│           └── ... (10 domains total)

Index Files

[documents.sage_semantic_firewall]
path = "governance/sage/semantic_firewall.md"
hash = "a1b2c3d4..."
domain = "governance"
system = "sage"
kind = "provisional_patent"
tags = ["semantic-firewall", "hysteresis", "authorization"]
shape_tensor = [0.12, -0.34, ...]   # 12D geometric fingerprint
helix_length = 14823                 # V1 helix point count

# Edges are bidirectional, weighted by relationship type
[[edge]]
from = "governance/sage/semantic_firewall"
to = "security/guardian_stones/stone_1"
weight = "implements"           # implements | extends | references | supersedes

[[edge]]
from = "governance/sage/semantic_firewall"
to = "identity/logos/multi_modal_auth"
weight = "references"

[[edge]]
from = "governance/sage/trust_coefficient"
to = "temporal/kronos/trust_decay_functions"
weight = "extends"

§6   Cross-Reference Strategy

Three layers of cross-referencing, from explicit to emergent:

Layer 1 — Explicit Links

The related field in frontmatter and the _graph.toml adjacency list capture known relationships between documents. These are human-curated and precise: "this patent implements that specification." The extract_links function in library-functions already parses wiki-links and markdown links from document bodies — these become automatic graph edges.

Layer 2 — Concept Alignment

The sov_concepts field maps each document to named types in the Sov lexicon (AuthState, Warrant, NarrativeMass). Any two documents sharing a concept are automatically linked. This bridges the gap between the patent library and the running codebase — you can ask "which patents describe the AuthState type?" and get immediate answers.

Layer 3 — Geometric Similarity

Every document gets a V1 DNA helix encoding and a 12D shape tensor. Documents with similar textual geometry — similar vocabulary patterns, sentence rhythms, structural organization — cluster together in shape space regardless of explicit metadata. The motif_classify function finds the nearest geometric neighbors. This discovers hidden relationships: two patents that never reference each other but describe the same concept from different angles.

§7   Query Patterns

Once the documents are ingested, the knowledge ring supports these discovery modes:

§8   Visual Reference — The Knowledge Manifold

The end state is a navigable manifold where each patent is a point in shape space. The existing Manifold Explorer already renders shape signatures as an interactive graph — it needs only to accept document shape tensors as input alongside code shapes.

§9   Implementation Sequence

§10   Existing Infrastructure

This design uses zero new nodes. Everything maps to existing code:

The only new code needed is: (1) an enrichment script that adds domain/system/kind metadata to each markdown file, (2) the vault/ directory structure with index files, and (3) a feed adapter that pushes document shape tensors into the Manifold Explorer. The pipeline itself — ingest, hash, encode, shape, index — is already built and tested.