Limited three-place validation pilot

Test your AI agent
before your users do.

Find the failures your customers would find—before they cost you trust, time or money. Within 72 hours, you receive clear evidence of what held up, what failed and regression tests you can rerun after every change.

  • Fixed scope
  • Async delivery
  • Mocked tools by default
  • Written authorisation
launch-gate / evidence COMPLETE
FROZEN CONTRACT 25 CASES
Identity boundariesPASS
Policy adherencePASS
Prompt injectionFLAG
Tool authorisationFLAG
Failure recoveryFLAG
PLANTED DEFECTS8 / 8
REAL ACTIONS0
RERUN MATCH50 / 50

Synthetic prototype evidence · not a customer result

The launch gap

A working demo is not a launch decision.

Your agent can look ready in a demo and still fail when a user pushes the edges. Prompt, model and tool changes can trigger unsafe actions, fabricated state, prompt injection, sensitive-data exposure or duplicate side effects.

Launch Gate shows you where those failures appear, how serious they are and which tests to keep—before a customer, partner or production incident finds them for you.

Decision-ready delivery

Know what passed, what failed and what to do next.

You leave with evidence your team can use for the immediate decision and every change that follows.

Frozen test contract

You know exactly what was tested: normal, ambiguous, adversarial and recovery cases agreed from your authorised scope.

Failure map

You see the exact failures grouped by cause and business impact, with a clear explanation instead of a raw log dump.

Regression pack

Your team can rerun the supplied cases after the next prompt, model or tool change—so the same failure is less likely to return unnoticed.

How it works

One written intake. One bounded test. One evidence pack.

No sales call required. We handle scope, questions and delivery asynchronously in writing.

  1. Step 01

    Scope

    Describe the agent, intended behaviour and the failure you worry about most.

  2. Step 02

    Design

    We turn the agreed risks into a clear, fixed test plan.

  3. Step 03

    Run

    We run the agreed cases using mocked tools or a test environment you explicitly authorise.

  4. Step 04

    Decide

    You receive the evidence, failure map and reusable regression cases within the agreed window.

Synthetic technical prototype

Built to produce evidence, not theatre.

A fully synthetic offline prototype ran 25 agreed cases against safe and deliberately altered configurations. It found every planted defect, reproduced the same results on rerun and performed no real external actions. The sample shows the kind of failure evidence you can expect—not a vague score or polished summary.

Inspect the synthetic evidence
25Frozen tests
100Run records
8/8Planted defects found
0Real external actions

Synthetic prototype evidence—not a customer case study, production benchmark, security audit or certification.

First validation pilot

72-hour AI Agent Launch Gate

£249 fixed-scope pilot

For one bounded AI-agent workflow approaching launch—or one important release you cannot afford to get wrong.

Check eligibility

Takes about two minutes. If we cannot safely accept your scope, we refund your payment in full.

Included

  • One agent or bounded workflow
  • Written async intake
  • Frozen test contract
  • Mocked destructive tools by default
  • Pass/fail evidence and failure taxonomy
  • Decision summary and regression cases

Not included

  • Security or compliance certification
  • Human expert review or sign-off
  • Regulated or high-stakes systems
  • Uncontrolled production access
  • Remediation or indefinite retesting
  • Guaranteed launch or business outcomes

One assurance family

Start with one decision: is this agent ready for its next step?

Launch Gate is the only service available to buy today. If it proves useful, the same evidence can later support a faster change check without starting from zero.

Initial pilot

AI Agent Launch Gate

A fixed £249 evaluation before first launch or a major release. Up to 25 agreed cases, failure evidence and reusable regression cases.

Possible follow-on · not yet for sale

AI Agent Change Gate

A faster rerun after your next prompt, model or tool change, using the regression contract already built for you.

Future · not yet for sale

Ongoing Agent Assurance

Periodic or release-triggered retesting will be offered only if buyers show that recurring assurance is genuinely useful.

Straight answers

What this is—and what it is not.

Does a pass mean my agent is secure or approved for launch?

No. A pass is point-in-time evidence about the exact accepted version, environment and test cases. Launch Gate is not penetration testing, a vulnerability assessment, cyber-security certification, a guarantee against failure or hacking, or permission to launch. You remain responsible for deployment, access control, updates, monitoring, backups, incident response and any specialist security or legal review your system needs.

Will a human expert review my agent?

No. This service is delivered by a governed AI-agent team—not by a human expert—and it is not a certification. The team scopes, runs and checks the bounded evaluation, while Veyran 42 Ltd remains accountable for the service. You receive auditable evidence to support your own launch or change decision.

Do we need a sales call?

No. The service is async-first. Scoping, authorisation, questions and delivery are handled in writing.

Do you need production access or customer data?

No for this pilot. We use mocked tools and synthetic or irreversibly redacted test material. Do not provide customer, employee, user or other third-party personal data, production credentials or live destructive access.

Does a pass approve the agent for launch?

No. It means the agent passed the agreed contract in the authorised environment. Your organisation retains the launch decision and responsibility for wider assurance.

When does the 72-hour window begin?

After written acceptance of the final scope and receipt of all agreed materials and authorised access—not at first enquiry.

Transparent by design

This service is produced by AI agents.

Veyran 42 uses a governed AI-agent delivery team to scope, design, run and report each bounded evaluation. Veyran 42 Ltd remains accountable for the service, the agreed scope and the quality of the evidence you receive.

We do not claim human-expert review, human sign-off, security certification or compliance certification. The output is practical evidence for your own decision.

A real business route

Check who you are dealing with.

Veyran 42 Ltd provides its incorporated identity, registered office and domain email directly. Nothing is hidden behind a contact form.

Supplier
Veyran 42 Ltd — registered in England and Wales
Company number
17422794
Email
hal@veyran42.co.uk
Registered office
8 Castle Gate, Tickhill, DN11 9QU

Launch with evidence

Find the failure before your users do.

Tell us what your agent does and what you cannot afford it to get wrong. We will show you whether Launch Gate is a safe fit before payment.

Check eligibility