CDR: Cognitive Derivation & Reverse Engineering

Your codebase contains business rules, state machines, and behavioral patterns that aren't written down anywhere. They're implicit in the code — in how functions call each other, how state transitions work, how data flows through the system.

CDR (Cognitive Derivation & Reverse Engineering) is Fabriq's pipeline that scans your source code and extracts this knowledge into structured, navigable assets.

What CDR produces

After a scan, you get a set of cognitive assets:

Asset	What it captures	Example
Behavior	How components interact — which functions call which	SearchBar.onSubmit → ProductList.filter
Business Rule	Explicit business logic encoded in code	"Orders over $100 get free shipping"
State Machine	State transitions and their triggers	Order: Pending → Paid → Shipped → Delivered
Evidence	Sourced, validated observations	Code references that support each finding
Domain	Business domain boundaries and entities	"Order Management", "User Profile", "Inventory"

The CDR pipeline

CDR runs as a multi-stage pipeline:

Bootstrap → Entries → Behavior → State → Domain → Generate
             Mining   Machine

Stage	What happens
Bootstrap	Reads project metadata, configures scanners
Entries Candidate	Identifies entrypoints (routes, event handlers, API endpoints)
Behavior Mining	Traces call chains between components — who calls whom and why
State Machine	Detects state transitions (enums, status fields, conditional branches)
Domain Analysis	Groups related entities into business domains
Document Generation	Produces navigable cognitive assets

Evidence levels

Every finding in CDR is tagged with an evidence level:

Level	Meaning	Example
Fact	Sourced and validated against actual code	"Order status transitions from 'pending' to 'paid' in OrderService.pay()" — supported by a code reference
Inference	Derived from patterns, plausible but not directly validated	"This authentication middleware likely applies to all /api/admin routes"
Unknown	Knowledge gap — the scanner detected something but couldn't classify it	Need human review to determine what this pattern represents

Facts are the most reliable. Inferences are useful but should be verified. Unknowns are opportunities for investigation.

Why CDR matters for agents

Without CDR, an agent must read the entire codebase from scratch for every feature — expensive and error-prone. With CDR:

• Agent queries the cognitive index instead of re-scanning
• Business rules are pre-extracted — the agent knows the rules before it modifies code
• State machines are navigable — the agent understands state transitions without tracing code
• Evidence is sourced — every finding links back to specific code locations

The cognitive index is available for context injection — you can send it to the agent before starting a feature.

When to run CDR

When	Why
After significant code changes	Keep the cognitive index up to date
Before starting a complex feature	Give the agent structured context
When onboarding to an unfamiliar codebase	Understand the system without reading everything
Periodically (weekly)	Catch knowledge drift before it becomes a problem

CDR scan time depends on your codebase size. A small project (50 files) takes under a minute. A large monorepo (5000+ files) may take several minutes.

Next steps

• Running Your First CDR Scan — step-by-step tutorial
• Injecting Context — send cognitive assets to your agent
• Audit Trail — every decision, recorded