Start here · Orientation
The reading anchor
ROOTVIEL is a dense body of work. This page is the way in: the handful of ideas everything else rests on, a map of what each part contains, and the order to read it in. Grasp the ideas below and the rest of the lab reads cleanly.
The anchor
The ideas everything rests on
Six load-bearing ideas. They recur on every page; each links to where it is developed in depth.
Reasoning is not authorizing
A model may reason, propose and coordinate — but a model’s reasoning or confidence is never authorization. Authority lives outside the model, in deterministic code, behind a frontier no decision branch can cross.
PROVEN / NOT PROVEN / UNKNOWN
Claims are kept honest by an epistemic discipline: what is proven is marked proven, and what cannot be proven is marked UNKNOWN — never “safe”. “It looks vulnerable” must never become a reported fact.
Evidence first
A finding exists only when an independent verifier confirms it with evidence that can be re-derived. Provenance and corroboration decide what is trusted — not how confident the model sounds.
Deterministic enforcement
The controls that keep the system safe and honest are deterministic and live outside the model: tools pass a single deny-by-default gate, and the agent cannot self-authorize.
The epistemic ladder
A claim climbs one rung at a time — Observation → Hypothesis → Experiment → Evidence → Verification → Finding. Only a verified result becomes a Finding; anything unproven stays UNKNOWN.
An honest maturity vocabulary
Every capability and every vulnerability carries its real state — Implemented, Experimental, Lab-only, Verified, Heuristic, Hypothesis/UNKNOWN — so nothing is dressed up as more finished or more certain than it is.
The map
What each part contains, in reading order
New here? Read top to bottom. Already oriented? Jump to what you need.
One lab, read six ways. The Academy explains the ideas, Research grounds them, the Knowledge Base documents them, the Labs let you experiment, the Catalog inventories what exists, and ROOTRAVEN applies the method on-chain.
- 1
Academy
The free, rigorous library — from the cryptography of verifiable agents to agent cognition, agentic architectures, on-chain zero-knowledge, and securing & verifying agents. Start here for the theory.
- 2
Agentic Security
The threat models for AI systems: prompt injection, RAG, MCP/tool safety, agent authority and the lethal trifecta — why enforcement must live outside the model.
- 3
Knowledge Base
The vulnerability encyclopedia across AI, agentic, Web3, application-code and infrastructure domains — each entry with its honest support level.
- 4
Research
Methods, agent architectures and results — how the lab investigates, attacks, verifies and produces evidence.
- 5
Technology
How ROOTVIEL works end to end: the cognitive cycle, the evidence model, signed intent and deterministic enforcement.
- 6
RootRaven
The Web3 product — the ROOTVIEL method carried onto the chain, promoting a finding only when a verifier re-derives it from evidence.
- 7
Security Labs
Interactive, reproducible experiments on synthetic, isolated scenarios — see the principles run.
- 8
Catalog
All capability families and overlays with their real states and declared limits — the honest inventory.
Who is behind it
Raven — builder of the ecosystems
Everything here is designed and built by Raven, the single author behind the whole stack — the research line, the deterministic engine that enforces it, the ROOTRAVEN product that applies it on-chain, and this lab and Academy that teach it. One author, one method, one standard of evidence. The identity is a public pseudonym on purpose: the work should stand on its evidence, not on a name — no inflated claims, no invented results, and a clear line between what is proven and what is UNKNOWN.
It has also been put to work: run at no cost in demanding, production-grade environments across organizations, scoped to those specific engagements rather than offered as a general guarantee. Where a fixed-playbook scanner reports where a pattern matches, these agents form a hypothesis, test it in an isolated sandbox and promote a finding only when an independent verifier re-derives it from evidence — a different method, and honest about what stays UNKNOWN.