Compliance tool pricing
Priced for the
risk you remove.
No fixed licence. No upfront cost. You pay when a change has been verified, so the bill tracks the risk we actually took off your desk.
Sandbox pilot
No cost
Run Refinery against your own COBOL in an isolated sandbox for 30 days. Your code never leaves your estate; the engine runs where you put it.
- Your programs, your infrastructure
- Full audit engine, no feature gating
- Signed Change Contract on every run
Per verified change
Risk-share
Once you are in production you pay against changes Refinery has actually verified. It is priced as a share of the compliance risk removed, not as a seat licence.
- No fixed licence, no seat count
- You pay when a change is verified
- Stops when you stop shipping changes
Estate-wide
Bespoke
For governing change across an entire estate rather than one team’s programs. Every dependency is mapped in advance, so what a change touches is known before anyone proposes it.
- On-premise or private cloud
- Multi-tenant estate scoping
- CRO portal and evidence retention
Why Refinery is different
Five strengths, and what each one actually gets you.
The same claims the homepage makes, stated so you can check them one at a time.
Change behaviour equivalence
Checked, with formal proof for supported arithmetic
Verification strategy
PIC-range proof and adversarial execution
Regulator deliverables
Signed PDF, SHA-256, evidence pack
Regulatory mapping
DORA Art. 25; SS1/23 opt-in; SR 26-2 cited
Who authors the code it checks
Any tool, or a person
How others compare
A green pipeline, a quality badge, an email.
Or a signed certificate.
A quick look at what you get from AI migrations, AI code gates and manual review. These are good tools, honestly described. The difference is that their verification is downstream of their own code generation. It stops at the point where a risk function needs something to file.
AI migration platforms
watsonx Code Assistant for Z · AWS Transform · Imogen
- What it checks
- Whether its own output behaves the same
- Who authors the code it checks
- Itself
- Verification strategy
- Generated tests, over the inputs they sample
- Regulator deliverables
- A green pipeline
- Regulatory mapping
- Not produced
AI code-quality gates
SonarQube AI Code Assurance
- What it checks
- Rules, bugs and vulnerabilities
- Who authors the code it checks
- Any tool
- Verification strategy
- Not attempted
- Regulator deliverables
- A quality badge
- Regulatory mapping
- Not produced
Manual review
Reviewer plus regression suite
- What it checks
- Whatever the reviewer thinks to look at
- Who authors the code it checks
- Any tool
- Verification strategy
- Reviewer judgement
- Regulator deliverables
- An email
- Regulatory mapping
- Assembled by hand
Every column is drawn from the vendor’s own published documentation, current as of September 2026.
Included at every stage
There is no cut-down tier.
- Every core check, plus the semantic engine
- Z3 formal proof of arithmetic equivalence, with counterexamples
- A signed, tamper-evident Change Contract PDF per audit
- Downstream impact analysis across CALL, COPY and JCL
- DORA Article 25 evidence mapping, with SS1/23 for regulated models and SR 26-2 cited for reference
- The compliance agent your risk team can question directly
Questions procurement asks
Does our COBOL leave our estate?+
No. Refinery runs where you deploy it, on-premise or in your own cloud. There are no mandatory external dependencies, and the engine works fully offline.
What does "a share of the risk removed" actually mean?+
You pay against changes Refinery verifies, not against how many engineers you have. If a quarter goes by without an AI tool touching production COBOL, that quarter costs you nothing.
Are we locked in?+
No. The evidence Refinery produces is a standard signed PDF plus a structured record; both are yours, readable without us, and remain valid if you stop using the product.
Do we need to replace our AI modernisation tooling?+
No. Refinery is deliberately independent of whatever writes the code. It checks the output of any tool, vendor patch or human commit, which is the whole point of an independent control.
Start with the sandbox.
Thirty days, your own COBOL, no cost. If it does not find anything worth showing your risk team, there is nothing to discuss.
Book a call →