Bad data stops at the door.
Most tools tell you on Tuesday what went wrong on Friday. OpenDQV Cloud validates every record as it is written, against plain-YAML contracts you own, and seals every decision in a tamper-evident audit chain. The bouncer at the door, at enterprise scale.
14 days · 200 ODUs · no credit card · no code to start
Validation beats inspection.
Inspection is all you have today.
Most data quality tools on your stack today inspect: they scan what already got in and report what they find, after the damage is travelling. Validation happens at the door, before a record enters, while there is still a decision to make.
Manufacturing spent seventy years learning this. It even priced the alternative: the rework burden of inspection-after-the-fact has a name, the Hidden Factory, and it quietly consumes 15 to 40% of capacity. Yours included.
The rules have a cost problem of their own. Without a validation platform they live everywhere: once in Salesforce, again in the lakehouse, again in the ERP, again in a loader script — every rule re-implemented per system, drifting apart from the day they're written. Add a system: implement them again. Swap Salesforce for Dynamics: your data migrates, and every rule you spent years encoding is rebuilt from memory.
OpenDQV Cloud makes the rules an asset that outlives the systems they guard. Contracts live once; every system calls the same gate; a migration touches plumbing, never rules — and governance gets one definition of valid, with one evidence trail, across the whole estate. Plumb once per system. Write the rules once, ever.
And you don't rip anything out. Keep the inspection tools you have; add validation at the front door. Validation + inspection = better data quality.
Said precisely: a synchronous gate wherever your code writes; automatic quarantine scanning where the platform writes for you.
One engine. Three jobs.
Your own Vanguard Compute Engine, per-second metered, doing the three things data quality actually needs: stop bad records, improve your data quality cycle by cycle, and prove what happened.
Block it at write time
A failing record never enters your system. The decision is synchronous, sub-millisecond in the engine, and returned to the caller with the exact rules that failed.
Measure, tighten, repeat
Every contract starts in observation mode: it counts what would have failed on your real traffic without blocking anything. Tighten the contract, watch the reject rate fall, promote to enforcement with the numbers in front of you. That cycle is how data quality actually improves.
An audit trail that can't be quietly edited
Every validation lands on a sealed, hash-chained audit log — every seal countersigned by a second, separate system, so the record can be independently verified and never quietly edited. Built for the conversations you have with regulators.
Your AI assistant already speaks to it
Every contract, metric and audit event is exposed over MCP, the open protocol AI assistants use. Data quality stops being the dashboard someone opens on Tuesdays and becomes a question anyone can ask, mid-conversation, and a check any agent can run before it writes.
balance_non_negative (112). Want the audit rows?
Your regulation is a data contract.
Most data quality problems are a regulation wearing a disguise. 41 starter contracts ship with every trial — each one starts in observation mode, counting what would have failed, until you choose to enforce it. The rule counts below are real: every chip is a rule you can read, edit, or delete on day one — they are starting points you own, not a black box we maintain for you.
Natasha's Law (PPDS)
A wrong allergen flag should never reach a label. The starter contract makes all 14 allergen declarations impossible to omit and checks each one against the reference lists you control — blocked before the record enters, with sealed evidence of your due diligence.
Martyn's Law
Drop your venue and training spreadsheets in and get a site-by-site gap report against all 63 rules — including the enhanced-duty fields: the designated senior individual, SIA notification, documented public protection procedures. Then enforce: records that cannot be saved incomplete, and when the inspector asks how you know, the sealed audit trail.
MiFID II
LEIs check-digit-verified (ISO 17442 mod-97), ISINs validated, timestamps and cross-field mismatches caught before the report leaves the building — replay last month's rejected file through it in observation mode and count what the ARM would have bounced.
DORA
An ICT incident register that is correct as it is written — not repaired the week before the audit.
GDPR
Processing records and DSAR requests validated at the door, with every accept and block decision on the sealed chain.
NHS DSP Toolkit
Patient-record fields validated at write time — NHS numbers checksum-checked, not just present — and the evidence of it sealed for your DSP Toolkit return.
Also in the catalogue: SOX control tests, HIPAA disclosure accounting, Companies House filings, Ofgem & Ofwat meter readings, the building-safety golden thread, telecoms CDRs, insurance claims, clinical trials — and the contracts you write for the rules only your business has. Read every contract, rule by rule, before you sign up →
Contract in, evidence out
From signup to your first validated record is about two minutes. The sequence below is the whole product.
Write what “valid” means
A contract is plain YAML, readable by your engineers, your auditors, and your AI. Start from 41 starters or write your own.
Put the engine in the write path
Call it from your pipeline over HTTPS, or let your AI assistant call it over MCP. Same engine, same contracts, same audit trail.
Not a developer? Drop a CSV in the dashboard instead and read the results — and keep doing it. A weekly spreadsheet run through the same rules lands the same sealed evidence; CSV-in, evidence-out is a supported way to run, not training wheels. The pipeline plumbing can come whenever it earns its keep.
Promote with evidence
Watch observation counts on live traffic, then flip the contract to enforcement: one deliberate action, recorded on the chain like everything else.
Your contracts are portable by construction
OpenDQV Core is the MIT-licensed open-source project that defines the contract format. OpenDQV Cloud runs those same contracts on a managed engine, so the rules you write here run anywhere the standard runs.
If you leave, your YAML goes with you. That is the point of a standard, and the reason you can adopt one without asking permission first.
Plainly: the contract format is open source; the Cloud engine is commercial software. The openness is your exit path, and we won't blur that line.
contracts → plain YAML, exportable any time
audit log → exportable any time, API or dashboard
records → in memory; never in audit, backups, or logs
lock-in → none by design
The prices on the page
are the prices
Metered on engine uptime, like the compute you already buy. No sales call at any self-serve tier. No minimum commitment, no termination fee.
14 days or 200 ODUs, whichever comes first. One XS engine, every feature that runs on it. No credit card.
Start free trialPay for the engine while it is running. Per-second billing, 60-second minimum per start. Stop the engine and the meter stops. Sizes XS (1 ODU/hr) and S (2 ODU/hr).
Upgrade in-productSame uptime billing, engine sizes to L — each size meters ODUs at its vCPU count (M 4 · L 8 ODU/hr) — SSO (OpenID Connect), 365-day audit retention, continuous-run pinning, direct support.
Upgrade in-productEverything in Enterprise, on your own substrate, inside your own network. This one is a conversation, the only one.
Talk to the founderOne ODU = one vCPU-hour of engine uptime. A size meters at its vCPU count: XS 1 · S 2 · M 4 · L 8 ODU/hr. Observation mode costs the same as enforcement: the meter is the engine being on, not what you ask of it. Worked example: an XS engine (1 vCPU) on UK office hours is ≈ 2,100 ODUs a year — about £10,500/yr on Standard or £16,800/yr on Enterprise; running 24/7 it is £43,800 / £70,080 — and a nightly batch window is a small fraction of the office-hours figure. Auto-stop means idle time costs nothing. Trial → Standard → Enterprise is entirely self-serve.
Your first validated record,
about two minutes from now.
Signup → engine provisioned → sample validated. Tell us what breaks; the founder reads every reply.
Start free trialGet to know OpenDQV Cloud
What is a data contract?
What happens to the records I validate?
Where does my validation data live?
You look small. How do I get comfortable trusting you?
What happens if the engine is asleep when a call arrives?
Can I gate writes into Salesforce or another SaaS?
How does it fit Databricks, Snowflake, or a Python pipeline?
mapPartitions step in Spark, a
pre-load step in your loader, one HTTP step in your orchestrator —
then land the passes and quarantine the blocks. Copy-paste snippets
for Python, Node, curl, Apex, Salesforce Flow, and dbt are in the
dashboard, and inspection tools you already run sit alongside
untouched. Full per-platform recipes, readable before signup:
the Connect page.