The three-twin architecture
Signal builds your operational model in three progressive phases. Each produces a deliverable. Each transition requires your approval. The system never acts without permission.
Three phases. Three deliverables. Two human gates. The system never acts without permission.
Principle
Deterministic before probabilistic.
File classification, work scoring, gap detection, and dependency mapping use deterministic algorithms. No language model. No hallucination. No inference cost. AI models are layered on top for generation and analysis, never in place of the deterministic foundation. You choose which models: local (on your hardware), external (Claude, GPT), or both.
Phase 1
Raw Twin: what exists
The system recursively scans your document environment. Every file is classified by type, indexed with a content hash, and linked to its source path. Multi-server support means distributed environments (multiple offices, cloud drives, local machines) merge into a single view.
No data is copied. The raw twin is a map of what exists, not a duplicate. Placeholders link back to your living documents. If a file changes, the next scan detects it.
Output: Raw twin report, file index with classifications, activity signals, cross-machine detection.
Human gate: You review what was found before anything enters the operational model.
Phase 2
True Twin: what the system perceives
From the approved raw data, Signal builds a complete operational model in a standalone database. Initiatives and their steps. Pipeline and leads. Financial position. Decision history. Milestones. Content calendar. Corpus index. Your business as the system perceives it.
The true twin is standalone. It can run independently. It does not depend on your live systems after creation.
Output: True twin report, your operational reality as a structured model. 3-model AI comparison (local, base, external) with deterministic scoring.
Human gate: You confirm the model accurately represents your business before optimization.
Phase 3
Optimized Twin: what matters
The same data, reorganized by structural importance. Every work item scored on four deterministic factors: revenue proximity, foundation alignment, dependency weight, and effort-to-impact ratio. Structural gaps detected, not just what is present, but what is absent. Dependency chains mapped from foundation to revenue.
The comparison report shows the delta: what the optimization revealed that was invisible in the raw perception. The raw twin saw files. The optimized twin derived intelligence. The difference is the deliverable.
Output: Optimized twin report, scored, gapped, dependency-mapped with executable opportunities ranked. Twin comparison report.
Human gate: You approve the optimization scope before any action is taken.
AI Architecture
Three tiers. The right tool at the right layer.
Deterministic
Classification, scoring, gap detection, dependency mapping. Zero hallucination. Zero cost. Zero latency beyond computation. This layer never changes regardless of which AI models are available.
Local AI
Models fine-tuned on your approved corpus. Run on your hardware. Optimized for narrow, specific tasks: operational Q&A, domain inference, context-aware analysis. You own the model.
External AI
Claude, GPT, or any API for high-quality generation when you choose. All outputs scored deterministically against ground truth. No vendor lock-in.
This is not theoretical.
We deployed this methodology on our own company. The build log shows the measured results.