Code in. Truth out.

governed transformation Β· verified provenance Β· zero trust required

Drop your AI-built project into the lab. A 5-layer reconnaissance engine with multi-model orchestration transforms it across 10 verified dimensions. Every change leaves a provenance receipt. Every decision is auditable.

Once a week, Discombobulator will run its own pipeline on its own codebase and publish the score here.

Self-Cannibalizing Audit
Self-audit results will appear here once the first daily run completes.
What the report looks like

Read the receipts.

Self-audit results will appear here once the first daily run completes.

● Constellation Map
β€”

Not a prompt.
A laboratory.

Each dimension is analyzed sequentially, building on the results of the last. Security hardening first. Then architecture and splitting on the hardened version. Then errors, dependencies, performance. Sequential, verified, never a single-pass hallucination gamble.

Between every dimension, a verification gate confirms the transformation before the next analysis begins. The Oracle synthesizes findings from multiple models and picks the best fix, not by vote, but by semantic analysis.

πŸ›‘οΈ
Security
Vulnerability scanning, auth hardening, input validation
πŸ—οΈ
Architecture
Module boundaries, separation of concerns, dependency flow
βœ‚οΈ
Splitting
God-file decomposition, module boundaries, clean imports
🧹
Dead Code
Unreachable paths, unused exports, phantom dependencies
⚑
Errors
Try-catch coverage, error boundaries, graceful degradation
πŸ“¦
Dependencies
Phantom packages, version conflicts, unused imports
πŸš€
Performance
N+1 queries, memory leaks, hot path optimization
βš™οΈ
Environment
Environment variables, config management, secret handling
πŸ§ͺ
Tests
Generated test suites, edge cases, coverage targets
πŸ“‹
SETUP.md
Install steps, env config, architecture docs

5 layers of intelligence

Three scouts analyze your code independently. An Oracle plans the transformation strategy. Territory agents execute with sub-agent depth. Integration reconciles. Verification seals.

L0: Scouts
Structural, Health, and Intent analysis
L1: Oracle
Territory planning, skill assignment, contracts
L2: Agents
Territory agents with sub-agent depth (max 2)
L3: Integration
Cross-manifest reconciliation, semantic merge
L4: Verification
DCN consensus, provenance sealing

Governed, not guessed

Every transformation travels through Conduit: a deterministic governance layer with 37 packet schemas, hash-chained provenance, identity verification, and a watchdog that never sleeps.

This is not another wrapper around an API call. This is a governed pipeline where every decision has a receipt, every change has a chain of custody, and every agent operates inside verified scope boundaries.

37 packet schemas
Typed contracts for every communication
Hash-chained provenance
Branching chains, immutable receipts
Identity + Isolation
Every agent verified, every scope bounded
Watchdog
Continuous monitoring, circuit breaking on anomaly
555 tests, 15.5K LOC
Phase 0 complete and verified

Three modes. One mission.

βš—οΈ
Full Discombobulate

All 10 dimensions. Full RECON. Every finding gets a fix. Your entire codebase, transformed end to end.

πŸ”¬
Error Assassin

Targeted strike on error handling, security vulnerabilities, and crash vectors. Fast, focused, lethal to bugs.

πŸ“Š
Audit Only

Diagnosis without surgery. Full health score, dimension radar, finding inventory. See exactly what's wrong before committing.

Credit-based pricing. Scale as you grow.

Loading plans…

Pricing goes live after beta. Join below to get early access.

Credits scale with issues found. No rollover.

Join the beta.

We’re onboarding teams in waves. Drop your info and we’ll reach out when your slot opens.

β€œWhy not just paste it into ChatGPT?”

A chat window gives you different results every time. Discombobulator gives you a locked 10-dimension pipeline with sequential verified handoffs, multi-model semantic synthesis, and a hash-chained provenance trail for every transformation. You get consistency, receipts, and a tool that publicly grades its own code every morning.

This is not a wrapper. This is a laboratory.