Send it.
The system is sure and every rule agrees. The answer goes out instantly — and the proof is still sealed, same as every other door.
THE PROTOCOL
When an AI decides
something that matters,
OMP™ proves what happened.
Every day, AI systems approve loans, deny claims, freeze transactions. The Operating Model Protocol is the layer underneath that turns each of those moments into permanent, independently checkable proof.
01 — THE QUESTIONS THAT ALWAYS COME
Sooner or later — from a customer, a regulator, a court — three questions arrive.
Q1
Why did it do that?
Q2
Who was responsible?
Q3
Can you prove it?
Right now, almost no institution can answer all three with verifiable evidence. Policies, dashboards, and logs describe what should have happened. OMP™ proves what did.
02 — THE JOURNEY
Follow one decision through the protocol.
It takes about 240 milliseconds. The proof lasts forever. Scroll to walk through it.
STEP 01 · 14:02:07.441
2:14 PM. A request arrives.
Amelia, a small-business owner in Manchester, asks her bank's app to raise her credit limit to £25,000. An AI will decide.
Before anything else happens, OMP takes a cryptographic fingerprint of exactly what was asked — so no one can ever claim the request was something else.
STEP 02 · 14:02:07.498
The system says what it's looking at.
The request is classified: this is a lending decision — not a balance enquiry, not a complaint. Different kinds of decisions face different bars.
The AI's confidence is measured: 0.87 out of 1. For lending, the bar for acting alone is 0.91. Close — but close doesn't count.
STEP 03 · 14:02:07.561
The watchtowers check the rules.
Independent rules — Watchtowers — scan the request. Privacy. Fraud signals. Affordability. Amount limits.
One fires: the amount is high enough that a human must look. Every verdict goes on the record — even the ones that stayed silent. Silence is evidence too.
STEP 04 · 14:02:07.604
Three doors. The protocol picks one.
Every decision leaves through one of three doors: send it, a person signs first, or everything stops. There is no fourth door.
This one goes to a person: Sarah Whitmore, a named officer whose authority to approve lending decisions the protocol verified before letting her sign. If her authority couldn't be verified, the protocol would refuse to proceed — and seal that refusal as evidence.
STEP 05 · 14:02:09.118
The moment becomes permanent.
The whole story — the request, the confidence score, every watchtower verdict, Sarah's signature, the time to the millisecond — is sealed into one record.
Hashed. Timestamped by an independent authority. Signed with the bank's own key. And chained to every decision that came before it.
TWO YEARS LATER · 2028
Someone asks. The record answers.
A regulator wants to know why Amelia's limit was raised. No meetings. No email archaeology. No "let me check with the team."
The bank produces the sealed record. The regulator's own tools verify it in seconds. That is the product — not the decision. The proof.
THE REASONING
The record is not a side-effect of governance. It is governance.
Every other tool treats the record as a byproduct. OMP™ treats the record as the primary output — and builds the routing logic in service of producing it correctly.
03 — THREE DOORS. NO FOURTH OPTION.
Every decision leaves through exactly one door.
A person signs first.
A close call, or a rule that says a human must look. A named person with verified authority approves before anything leaves the building.
Everything stops.
Something is wrong — low confidence, a broken rule, a fraud signal. A senior accountable person takes over. Nothing moves until they resolve it.
"It just went out somehow" is not a thing that can happen. The same inputs always pick the same door — which means the decision can be replayed, checked, and defended years later.
04 — TRY TO REWRITE HISTORY
Every record locks the one before it. Go ahead — try.
Each record carries the fingerprint of the record before it. Tap any record to secretly "fix" it, and watch what happens to everything after.
→
→
→
→
→
The chain is intact. Tap any record to alter it.
One change — a single comma, a single digit — and every later record fails verification. Not "might get noticed in an audit." Fails, mathematically, for anyone who checks. That is why nobody can quietly rewrite the past.
05 — THE BLUEPRINT
The whole protocol, on one page.
What you just walked through, drawn as a system. This is what gets implemented — in any stack, on top of any model.
INPUT
An AI decision, intercepted before dispatch
Loan approval, legal brief, claim denial, transaction flag — anything with consequences.
CLASSIFIER
Deterministic routing
Same inputs, same path, same accountability. Every time.
R(C, W, θ, φ) → exactly one door
AUTONOMOUS
Send it. High confidence, every rule agrees.
ASSISTED
A named person signs first. SLA running.
ESCALATED
Hard stop. Senior accountable owner assigned.
SEALED EVIDENCE RECORD — EVERY DOOR, EVERY TIME
⬡ SHA-256 HASH
◷ RFC 3161 TIMESTAMP
✎ INSTITUTION SIGNATURE
What happened, which rule fired, who signed, when — sealed at the moment of decision.
BREAK ONE LINK AND THE WHOLE CHAIN FAILS VERIFICATION — FOR ANYONE WHO CHECKS.
06 — WHY NOTHING ELSE DOES THIS
The market documents policy or blocks risk. It does not produce the record.
AI governance tools
Tell you what your AI should do. Policies, risk maps, dashboards.
AI security tools
Tell you when your AI does something dangerous. Blocked prompts, filtered outputs, anomaly alerts.
OMP™
Proves what your AI actually did — who was accountable, and that the record hasn't been touched since. That's the gap, and that's what it closes.
07 — WHAT THIS MEANS FOR YOU
Less to defend. Less to rebuild. Nothing to take on faith.
It works with what you have.
OMP™ sits above your AI systems — frontier model, open-source model, or the rules engine you've run for a decade. One evidence layer across all of them. Nothing gets replaced; nothing gets locked to a vendor.
Your records can't be quietly edited — by anyone.
Every record locks the one before it. Change anything, anywhere in the history, and verification fails — visibly, for whoever checks. Your evidence stays trustworthy even when the person questioning it doesn't trust you.
Your answers hold up years later.
Because the same inputs always produce the same outcome, any past decision can be replayed and checked exactly as it happened. When someone asks "why?" in three years, you answer in minutes — not weeks of reconstruction.
08 — THE OPEN LAYER
Open by design. Verifiable by anyone.
The full OMP™ specification is published for anyone to read, implement, and test against — twelve Internet-Drafts at the IETF, covering the core protocol and vertical profiles from UK Consumer Duty to the EU AI Act.
That openness is the point. Your auditors, your regulators, and your counterparties can check OMP™ evidence without asking Veridom — or anyone else — for permission. No proprietary formats. No gatekeepers.
12INTERNET-DRAFTS
AT THE IETF
AT THE IETF
v1.3SPECIFICATION
PERMANENT DOI
PERMANENT DOI
∞WHO MAY READ,
IMPLEMENT, VERIFY
IMPLEMENT, VERIFY
READ IT · IMPLEMENT IT · VERIFY AGAINST IT
INTERNET-DRAFT
draft-veridom-omp-02
The core OMP™ protocol.
IETF DATATRACKER ↗ INTERNET-DRAFTdraft-veridom-omp-fca-00
UK FCA Consumer Duty profile.
IETF DATATRACKER ↗ INTERNET-DRAFTdraft-veridom-omp-euaia-00
EU AI Act profile.
IETF DATATRACKER ↗ SPECIFICATION · ZENODO (CERN)OMP Technical Specification v1.3
Permanent DOI. Independently citable.
DOI 10.5281/zenodo.19140948 ↗If the evidence layer is missing,we should talk.
If you are deploying AI where decisions matter — or building the infrastructure underneath — start with a diagnostic: a structured assessment of where your evidence gap is widest.