Unique Selling Points¶
Substrate's six unique selling points form a defensive moat that no single competitor can easily replicate.
USP 1: The WHY Layer¶
Every tool today tells you what exists. No tool tells you why it was built that way.
The Problem¶
When a developer asks "why does the payment service have to go through the gateway?", the answer is scattered across: - ADRs buried in Confluence - PR comments from 18 months ago - Post-mortems filed and forgotten - Tribal knowledge that left with the last senior engineer
The Solution¶
Substrate captures ADRs, post-mortems, PR review rationale, and Slack decisions as first-class graph citizens with WHY edges. A developer joining six months later can:
- Click the payment service node
- See WHY edges connecting to:
- ADR-047: The decision that mandated gateway routing
- POST-019: The incident that caused the rule
- POLICY-012: The active policy enforcing it
The Result¶
Full provenance in under 5 seconds. The developer understands: - The incident that caused it - The ADR that formalized it
- The policy that enforces it
Competitive Gap¶
No existing tool captures decision provenance at the graph level. IDPs catalog services but not reasoning. Documentation tools store text but not relationships.
USP 2: Pre-Change Simulation¶
No competitor offers pre-change what-if analysis at the architectural graph level.
The Problem¶
An architect wants to propose splitting a service. The current process: 1. Write a design doc 2. Wait 3 months for review 3. Discover issues after implementation starts 4. Rewrite or abandon
The Solution¶
Describe the proposed change in natural language:
"What if I split OrderService into Order and OrderHistory?"
Substrate: 1. Translates to structured mutation spec 2. Clones current graph into ephemeral sandbox 3. Applies the mutation 4. Re-evaluates all active policies 5. Returns before/after comparison:
| Metric | Before | After | Delta |
|---|---|---|---|
| Policy violations | 3 | 7 | +4 ⚠️ |
| Services affected | 12 | 18 | +6 |
| Drift score | 0.23 | 0.31 | +0.08 |
Time: <15 seconds
Code written: Zero
The Result¶
Shifts governance left of the IDE, not just left of production. Architects validate proposals before any engineering time is invested.
Competitive Gap¶
Enterprise architecture tools (LeanIX, Ardoq) plan but don't simulate against runtime reality. Observability tools see current state but can't model changes.
USP 3: SSH Runtime Verification¶
No existing IDP platform connects via SSH to verify what is actually running on hosts.
The Problem¶
Your graph says Service X runs on Host Y. But: - Someone deployed manually - A container crashed and wasn't restarted - A service was moved but the graph wasn't updated
Current tools trust declared state. Reality diverges silently.
The Solution¶
Substrate's SSH Runtime Connector: 1. Uses Vault-signed ephemeral certificates (5-min TTL) 2. SSH to host via ProxyJump (no agent forwarding) 3. Runs inspection script: - systemctl list-units → running services - ss -tlnp → actual port bindings - dpkg -l → installed packages - sha256sum /etc/config/* → config integrity 4. Compares against graph-declared state 5. Raises runtime violations for discrepancies
Checks every 15 minutes per host.
The Result¶
Detects shadow deployments, configuration drift, and undeclared services within minutes — not months.
Competitive Gap¶
No IDP implements agentless SSH verification. Monitoring tools see metrics but not topology. Config management tools (Ansible, Puppet) enforce but don't verify continuously.
USP 4: Hardened GraphRAG¶
Microsoft's baseline GraphRAG has three production-breaking gaps we solve.
The Problem with Baseline GraphRAG¶
| Failure Mode | Evidence |
|---|---|
| Hallucinated entities baked permanently into graph | AGRAG paper: LLM entity extraction fails without correction |
| No temporal reasoning | Treats 2019 ADR as equally current as 2026 one |
| 73-84% of errors are reasoning failures | KET-RAG study: Gold answer present but still wrong |
The Solution¶
Substrate's layered retrieval pipeline:
| Strategy | Solves | Accuracy Boost |
|---|---|---|
| HyDE | Terse queries vs verbose documents | +15% recall |
| RAPTOR Tree | Cross-domain synthesis | +20% accuracy |
| Temporal Snapshots | Stale context | +12% relevance |
| Hybrid RRF Fusion | Single-strategy failures | +8% precision |
| Confidence Scoring | Uncertainty blindness | Reject 15% low-confidence |
The Result¶
GraphRAG accuracy >85% on code architecture queries, suitable for production governance decisions.
Competitive Gap¶
Microsoft's GraphRAG is open-source but unhardened. Competitors using baseline GraphRAG will hit the same production failures.
USP 5: Active Governance¶
IDPs catalog. Observability platforms sense. EA tools plan. Substrate blocks.
The Problem¶
Current tools are passive: - Backstage: "Here's your service catalog" - Datadog: "Here's your service map" - SonarQube: "Here are your code smells"
None prevent violations from reaching production.
The Solution¶
Substrate actively blocks architectural violations:
- Developer opens PR
- Ingestion parses changed files
- Graph Service evaluates against OPA policies
- Violation detected: Direct service-to-service call bypassing gateway
- GitHub Check API: ❌ BLOCKED
- PR comment with:
- Plain English explanation
- Linked ADR (ADR-047)
- Linked post-mortem (POST-019)
- Suggested fix (if deterministic)
Enforcement level: Hard-mandatory, soft-mandatory, or advisory
The Result¶
Violations caught and explained before merge, not discovered in production.
Competitive Gap¶
No platform combines deterministic policy evaluation with graph-grounded explanation. AI-only tools lack determinism; rule-only tools lack context.
USP 6: Complete Data Sovereignty¶
All AI inference runs on self-hosted hardware — zero data leaves the building.
The Problem¶
Security-sensitive organizations (fintech, healthcare, government) cannot: - Send source code to OpenAI or Anthropic - Expose architecture topology to cloud APIs - Risk policy logic in third-party systems
This eliminates every cloud-native AI governance competitor.
The Solution¶
Substrate runs entirely on self-hosted infrastructure:
| Component | Deployment | Data Residency |
|---|---|---|
| Application services | Docker containers | Customer infrastructure |
| AI inference (vLLM) | Bare metal systemd | Customer hardware |
| Graph database | Docker container | Customer infrastructure |
| Embeddings | Local BGE-M3 | Never leaves |
DGX Spark Memory Budget: - Llama 4 Scout (MoE): 55 GB always resident - Dense 70B: 38 GB always resident - BGE-M3: 0.6 GB always resident - Total: ~102 GB persistent + 26 GB cache
The Result¶
Compliant with the strictest data sovereignty requirements from day one.
Competitive Gap¶
Cloud-native competitors (GitHub Copilot, most IDPs) require external API calls. Self-hosted alternatives lack Substrate's AI capabilities.
Moat Summary¶
| USP | Year 1 Defense | Year 3 Defense | Year 5 Defense |
|---|---|---|---|
| WHY Layer | Implementation complexity | Switching costs (lost memory) | Organizational DNA |
| Simulation | Algorithm complexity | Training data advantage | Industry benchmarks |
| SSH Verification | Security expertise | Compliance certifications | Industry standards |
| GraphRAG | 6-12 month head start | Retrieval optimization secrets | Academic partnerships |
| Active Governance | Policy library depth | Customer policy libraries | Regulatory recognition |
| Local Inference | Hardware expertise | Deployment automation | Air-gap standard |
Combined: These six USPs create a defensive position no single competitor can replicate without 3-5 years of focused investment.