OneTick Blog

Trust & Verify: Takeaways from the TrueTick Webinar

Written by Mick Hittesdorf | Nov 19, 2026, 5:36:38 PM

By Mick Hittesdorf, OneTick Senior Cloud Architect

If there's one idea I hope people took away from our recent webinar, TrueTick – Introduction to AI Ready Market Data, it's this: quants, traders and surveillance analysts will not use data they don't trust. It doesn't matter how fast your platform is or how broad your vendor's coverage looks. Once users find defects, trust erodes quickly, and it is very hard to win back.

Thank you to everyone who joined us live and sent in such sharp questions. For those who couldn't make it, here are the main takeaways.

An assumption is not evidence

At most firms, data quality is assumed until something breaks. The market data team assumes the vendor validates the feed; the quants assume the platform team does. Very few firms actually verify it. "Trust, but verify" applies here: using a vendor's feed is not proof that the data is right.

I learned this the hard way as head of data engineering at a quantitative market-making firm. We built quickly but didn't invest enough in proactive testing. We only had a few issues, but our traders found them before we did, and that was enough to damage trust.

The problem is industry-wide. In one case, a vendor promised sub-second real-time delivery, yet some recipients got data several minutes old during peak volumes, which you'd never spot just by watching ticks. Only two of 31 banks were fully compliant with the BCBS 239 risk data principles a decade after the deadline. About 13% of outstanding derivatives reported under EMIR have missing or outdated valuations. And nearly 10% of US monthly stock returns changed after a routine data update.

Treat data like code, across six dimensions

TrueTick is an enterprise-grade data quality testing and monitoring platform built for capital markets data. It borrows what software engineering learned over decades: write tests, automate them, track results, file tickets and fix bugs. Tests are written in Python and SQL in Jupyter notebooks using pytest, so any modern developer will feel at home.

General-purpose monitoring tools can tell you whether a feed is up or down. They can't easily express the rules market data needs, like relationships within and across feeds, or checks against symbology and exchange calendars.

Every TrueTick test is classified into one of six standard data quality dimensions:

  • Timeliness: is the data available when you need it?
  • Completeness: is all of the data there?
  • Uniqueness: are there duplicates?
  • Accuracy: does it agree with an objective source of truth?
  • Validity: does it conform to business rules and formats, such as ISINs, CUSIPs and RICs?
  • Consistency: does it agree within and across feeds, such as one-minute bars versus raw ticks, or a feed versus your security master?

Make quality a number you can track

To make data quality objective, TrueTick rolls every test into a data quality score: the share of tests that pass or warn, out of all tests run. The score is computed by database, schema and dimension and compared with a configurable KPI, 95% by default. Green, yellow and red show at a glance whether your SLOs are being met.

A score turns quality from a feeling into something you can trend, share with stakeholders and hold yourself to. It also supports what really matters: a daily program of writing tests, watching trends, responding to alerts and fixing problems. That's how quality improves over time, and TrueTick sends a daily report covering both T+1 and real-time tests to keep everyone on track.

What the demo showed

TrueTick started as an internal project to improve OneTick Cloud data quality, and we've been running it as an internal pilot for about six months. In the demo I walked through our own dashboard, covering a subset of the 250+ venues in OneTick Cloud.

  • A heat map of the whole data estate. Each database is scored by dimension. White cells reveal gaps in coverage, like a database with no timeliness tests yet.
  • Trends and drill-down. You can move from a ten-day trend to a single failing test, the notebook that ran it and the data it ran on. One example flagged missing symbols in a futures bars database.
  • Real-time alerts. Just after the open, a latency test failed on a few loader threads that exceeded a one-second threshold. A companion test showed why: a spike in tick volume. The failure triggered Slack and email alerts automatically.
  • Scheduling built in. Tag a notebook cell with "schedule" and TrueTick runs it, with no third-party scheduler needed. Jobs run on spot instances, which can save up to 80 to 90% on compute.
  • Claude built in. TrueTick Studio includes Claude Code, plus a data catalog and MCP server. I asked Claude to build a test plan for an options database from a set of pricing best practices. It discovered the relevant databases, schemas and rates data on its own and proposed three notebooks with dependent tests, in a fraction of the time it would take by hand. Claude can also help investigate failing tests and suggest fixes.

Highlights from the Q&A

How do we know our data is AI-ready? This is the question on everyone's mind. Many AI pilots stall before production because the data isn't ready. AI-ready data is data tested against the failure modes of AI: look-ahead and survivorship bias for machine learning, and contradictory context or temporal confusion for LLMs. A stock split, for instance, can look like a crash to an LLM that doesn't know about the corporate action. Build tests for those conditions and you can turn a data quality score into an AI readiness score. We'll be publishing a white paper on this soon.

How do we avoid false alerts? TrueTick is calendar-aware, so it knows about holidays, half days and session breaks and won't run jobs when markets are closed. You can also downgrade a known or probabilistic failure to a warning with one line of Python. It's still logged and tracked, but it won't page anyone.

Can it test data that isn't in OneTick? Yes. TrueTick can read Parquet, including arbitrary Parquet, and data directly from S3.

I already run my own sanity checks. Why change? Keep them! TrueTick lets you drop those queries into automated, scheduled jobs so the whole firm benefits, with tracking, alerting and diagnostics built in.

Can I hold vendors to their SLAs? Absolutely. Ask vendors how they validate and score their data, and use TrueTick to bring evidence when they miss.

What does it take to get started? We host TrueTick in SaaS mode on OneTick Cloud, with a dedicated stack per client that can typically be deployed within hours. Bring-your-own-cloud deployment is expected as soon as next quarter.

Where do you stand?

Do any of these questions sound familiar? Maybe you aren’t running automated data quality tests. Or you only investigate breaks after they happen. You often hear about data issues from your users first. If so, I'd love to talk. Email me at mhittesdorf@kx.com.

Learn more about TrueTick, or watch the full webinar recording if you missed it.

Join the private preview. We're inviting a select group of capital markets firms to try TrueTick on their own data. Preview participants get their own TrueTick stack and a dedicated engineer to implement it on at least one of their databases. You'll see firsthand how to make your data trustworthy and AI-ready, and help shape where TrueTick goes next. Apply to join the private preview today.

 

Best wishes,

Mick Hittesdorf