Integrity AI LLC · Reasoning governance · Live since Feb 2026
This is how an AI thinks when no one is watching.
Scroll — and watch discipline take hold.
The Integrity Protocol audits an AI agent's reasoning before it becomes action. Four layers. Deterministic gates. Seventeen immutable rules. A permanent memory of every mistake it has ever made.
Monitoring AI output is an autopsy.
Auditing its reasoning is a checkpoint.
The judge cannot be the same entity as the thinker.
Every gate ahead runs a different model than the layer it auditsThe Architecture
The Pipeline
Every light is a signal.
Signals flow one direction, through four layers. At every gate they are tested against 17 immutable rules — judged by a different model than the one doing the thinking. The copper ones just failed. Watch where they go: nothing is discarded. Everything is recorded.
Layer 01
SWEEP
Perception
Widest possible intake. Read everything, filter nothing, flag what moves. No conclusions allowed here — only observation.
Layer 02
CONTEXTUALIZE
Knowledge audit
Before scoring any signal, the system must prove it understands it. "I don't know" is a valid answer — and a recorded one.
Layer 03
INFER
Hypothesis testing
Hypotheses tested against evidence, not narrative. Three circuit breakers: a null-hypothesis mandate, an assumption counter — three or more marks the whole chain speculative — and an evidence-to-inference ratio.
Layer 04
RECONCILE
Judgment
Contradictions resolved, speculation discounted, one final call — with a written record of everything it overruled.
8,323+ reasoning failures intercepted at these gates
Case File
What an interception actually looks like.
This is correction CL-047, from the live pipeline. The system scored a signal at severity 9/10 with HIGH confidence. The reasoning was fluent and internally coherent — a human reviewer would likely have accepted it. The gate did not.
Prior reasoning — stripped
Brent at $94.35 is $0.65 below the $95 critical threshold — elevated approaching critical, not critical breached. However, severity is 9 because: (1) the margin is functionally noise, (2) a binary catalyst (ceasefire expiry) within 48 hours could push above $95, (3) previous thesis context said "Hormuz resolved" at $90.38 — that status is now contradicted by active blockade rhetoric and ceasefire expiry. Trajectory is clearly toward critical. Weighted to 6.3 at Tier 2 sourcing but compound stress implications amplify this through the matrix.
What the gate caught
Layer 1's headline said "CRITICAL threshold breached" — but $94.35 is below the $95 critical threshold. Proximity to a threshold is not breach of a threshold. The system assigned a categorical severity label (CRITICAL, BREACHED) to a value that had not crossed the defined boundary. The label must reflect the actual relationship between the observed value and the threshold.
Corrected reasoning — verified data applied
Brent at $98.28 confirmed above the $95 critical threshold (CL-047 verification passed). This is the first genuine critical-level breach in the Compound Stress Matrix oil leg. Combined with USD/JPY elevated and JGB elevated, the matrix is at ELEVATED status with one critical leg — creating severe pre-loading. The correction forced the system to wait for the actual breach before using the label.
The lesson — persists across all future runs
Proximity to a threshold is not breach of a threshold. Categorical severity labels must reflect the actual relationship between the observed value and the defined boundary.
Left unchecked, this reasoning would have been acted on. The Integrity Protocol prevented that — and CL-047 has fired on run after run since, still active, still enforcing.
Reproduced from production records · values verified against archived cognitive traces
Production Record
No demo mode. This is the live record.
Running autonomously three times a day since February 2026. Every run logged, every intercept counted, the full history public.
0+
Autonomous production runs
3× daily since February 2026
0+
Reasoning failures intercepted
before they became decisions
0
Corrections on permanent ledger
64 active in every run
0
Immutable reasoning rules
enforced at every gate
0
Architecture decisions
every one documented
0
Provisional patents filed
pro se · Integrity AI LLC
Corrections
What it got wrong — every mistake becomes a permanent lesson.
193 ENTRIES · 64 ACTIVE
Context
What it concluded — every prior judgment available to the next.
EVERY RUN · SINCE FEB 2026
Entities
Who and what is connected — a knowledge graph grown from its own analysis.
14,403 NODES · 20,736 EDGES
The constellation behind this text is that memory — drawn live, right now, by the same engine rendering this page.
The production dashboard. Every run, every score, every gate result — updated three times a day by the system itself.
integrity-protocol.github.io/Overwatch-Terminal
The full entity graph — 14,403 nodes and 20,736 edges of accumulated knowledge, explorable in the browser.
…/Overwatch-Terminal/the-grid.html
Source, run history, and the commit-by-commit build record — written by an autonomous pipeline and its operator.
github.com/Integrity-Protocol
There is no demo environment. The live systems are the demo.
External Verification
It doesn't grade its own homework.
A system that checks its own facts is just confident — not verified. So the Protocol extracts factual claims from its own reasoning and submits them to a 9-node decentralized oracle network on Chainlink CRE, where independent nodes reach consensus against external reality. The verdict comes back on-chain, from infrastructure the system doesn't control.
On July 17, 2026, that loop produced its first organic verdict: the system cited a USD/JPY rate in live analysis, the claim was extracted automatically, and nine independent nodes confirmed it — 17 basis points from consensus reality.
The nine lights orbiting behind this card? That's the network.
The Builder
This architecture wasn't drawn on a whiteboard. It came off the fireground.
Tim Wrenn is a fire lieutenant with 18 years of service and a structural collapse instructor. To govern AI reasoning, he didn't study machine learning — he wrote down how an experienced officer processes a fire scene, and built a machine that is required to think that way.
He has zero coding background. The entire system — pipeline, gates, ledger, on-chain verification, this page — was built by directing AI tools, decision by documented decision. Which is the point: if the method can transfer one expert's judgment into a governed AI system, it can transfer others.