Semantic Infrastructure Lab

Version: 1.0
Date: 2025-12-15
Status: Working document

Overview

This document compares the different layer models that have emerged across SIL documentation. The goal is to understand the differences, evaluate trade-offs, and converge on a canonical model.

Model Summary Table

Model 1: Canonical (Glossary v2.2)

Source: SIL_GLOSSARY.md
Status: Current canonical reference

L6: Intelligence    - Agents, Planning, Adaptation (Agent Ether, BrowserBridge)
L5: Intent          - Goals, Constraints, Roles (Pantheon validation)
L4: Dynamics        - Time, Behavior, Execution (Morphogen scheduler)
L3: Composition     - Graphs, Routing, Topology (Pantheon IR, SUP)
L2: Structures      - Semantic Units, Types (TiaCAD, GenesisGraph)
L1: Primitives      - Irreducible Operations (Morphogen domains, RiffStack)
L0: Substrate       - Physical/Computational Reality (Philbrick hardware)
─────────────────────────────────────────────────────────────────────────
Meta: Observability - Cross-cutting (Reveal)

Characteristics:
- Hardware-grounded (Philbrick at L0)
- Observability as meta-layer
- Product-centric assignment
- 7 layers + meta

Patron Saints (from OSI_LAYER_MAPPING):
- L6: Marvin Minsky (Agents)
- L5: Douglas Engelbart (Augmentation)
- L4: Alan Turing (Computation)
- L3: Claude Shannon (Information)
- L2: George Philbrick (Modularity)
- L1: Harold Black (Feedback)
- L0: Richard Feynman (Physics)

Model 2: Original Semantic OS (6-Layer)

Source: SIL_SEMANTIC_OS_ARCHITECTURE.md (pre-alignment)
Status: Historical, partially updated

L5: Human Interfaces
L4: Deterministic Engines (Morphogen)
L3: Agent Ether
L2: Domain Modules
L1: Pantheon IR
L0: Semantic Memory

Characteristics:
- Memory-grounded (Semantic Memory at L0)
- Human interfaces at top
- Agent Ether as middleware (L3)
- 6 layers, no meta

Model 3: Feedback Loops (6-Layer)

Source: SEMANTIC_FEEDBACK_LOOPS.md
Status: Historical

L6: Applications (Scout, Morphogen)
L5: Agent Orchestration (agent-ether)
L4: Semantic Primitives (USIR, knowledge graphs)
L3: Feedback & Reflection
L2: Tool Infrastructure (reveal, tia)
L1: Storage & Indexing (Beth, Gemma)

Characteristics:
- No L0 defined
- Feedback as explicit layer (L3)
- Tool infrastructure prominent
- Application-focused top layers

Model 4: Observability (7-Layer + L4.5)

Source: SEMANTIC_OBSERVABILITY.md (pre-alignment)
Status: Historical, had unique L4.5

L7: Applications (Scout, Reveal, Agent-Ether)
L6: Agent Orchestration
L5: Pantheon
L4.5: SEMANTIC OBSERVABILITY  ← Unique!
L4: TIA
L3: Beth
L2: Domain modules
L1: Semantic primitives

Characteristics:
- Observability as full layer (L4.5)
- Tools as layers (TIA at L4, Beth at L3)
- 8 effective layers
- Most tool-centric model

Model 5: Provenance-First (Proposed)

Source: sessions/heating-snow-1214/README_2025-12-15_08-52.md
Status: Proposed alternative

L6: Reflection      - Learning from execution (observability)
L5: Execution       - Doing work under constraints (agents)
L4: Composition     - Cross-domain integration (Pantheon IR)
L3: Intent          - What we're accomplishing (contracts)
L2: Trust           - Who can do what (TAP, Authorization)
L1: Meaning         - Embeddings, types, similarity (Beth, Pantheon)
L0: Provenance      - Everything has lineage (GenesisGraph)

Characteristics:
- Problem-centric (solves LLM coexistence)
- Provenance as foundation
- Trust as explicit layer
- Philbrick becomes optional backend
- Tools span layers (like Unix utilities)

Design Rationale:
If core problems are:
1. Decomposing intent into trackable work
2. Trust relationships for LLM/AGI coexistence
3. Cross-domain tooling
4. Meaning manifolds for analogies

Then layers should reflect those problems, not products.

Unix Philosophy Parallel:
- Unix insight: "Everything is a file"
- Semantic OS insight: "Everything has meaning and provenance"

Component Assignment Comparison

Key Architectural Questions

1. What is L0?

Question: Is the foundation hardware, memory, or trust?

2. Where do tools fit?

Question: Should TIA/Beth/Reveal be layers or cross-cutting utilities?

3. Is Observability a layer or meta?

4. Where does Trust/Authorization fit?

Cross-Cutting Concerns (All Models)

Regardless of layer assignment, these span the stack:

1. Provenance - Who made what, when, from what
2. Trust - Who can do what to whom
3. Observability - What's happening, how well
4. Feedback - Learning from execution

The Provenance-First model makes two of these (Provenance, Trust) explicit layers rather than cross-cutting.

Critical Missing Subsystems

The coral-shine-1212 and brewing-sleet-1212 sessions identified three subsystems that don't yet exist but are critical for any model:

1. Intent Verification Subsystem

Problem: No mechanical way to verify that an action preserves the original intent.

Solution: Intent as cryptographic primitive - asymmetric verification where it's easy to check but hard to fake.

Components needed:
- intent.py - Intent object schema
- signature.py - Intent signature verification
- contract.py - Contract enforcement
- amendment.py - Intent versioning

2. Uncertainty Tracking Subsystem

Problem: Uncertainty compounds geometrically through operations; system can't detect or prevent runaway uncertainty.

Solution: Uncertainty as first-class field - track assumptions, coupling, and propagation gradients.

Components needed:
- uncertainty.py - UncertaintyProfile schema
- propagation.py - Gradient tracking
- brakes.py - Abort semantics, checkpoints

3. Cross-Domain Translation Subsystem

Problem: Domains translate meaning ad-hoc; invariants not preserved; loss untracked.

Solution: Semantic operators with contracts - like FFT preserving information while changing representation.

Components needed:
- operator.py - SemanticOperator base class
- invariant.py - Invariant schema
- registry.py - Operator catalog
- verification.py - Round-trip testing

Subsystem Layer Mapping Summary

Key Insight: These aren't new ideas - we're applying proven mathematical patterns to semantic computing.

Evaluation Criteria

To decide on a canonical model, evaluate against:

1. Problem fit - Does it help with LLM coexistence, cross-domain work?
2. Conceptual clarity - Can you explain it in 2 minutes?
3. Component coherence - Do assignments make sense?
4. Extension path - Can new projects find their place?
5. Implementation guidance - Does it help developers?

See MODEL_EVALUATION.md for detailed analysis.

Session References

• enchanted-centaur-1214 - Discovery of 4 competing models
• heating-snow-1214 - Provenance-first proposal
• noble-kraken-1125/COGNITIVE_OS_MASTER_MAP.md - Comprehensive mapping attempt
• temporal-fractal-1214 - Documentation audit

Next Steps

1. [ ] Complete MODEL_EVALUATION.md with criteria scoring
2. [ ] Draft PROVENANCE_FIRST.md with full rationale
3. [ ] Get stakeholder input on key questions
4. [ ] Propose canonical model
5. [ ] Update source documents to align