DataKitchen, Inc., 405 Waltham Street #101, Lexington, MA (2026)

10/15/2026

We migrated a database, left 483 orders behind on purpose, and the test suite passed. 55 tests. Zero failures.

That suite checks whether data looks healthy: no blanks, no duplicate keys, values in range. The migrated data did look healthy. It was just missing 483 rows, and nothing in a health check knows what used to be there.

📉 Gartner puts migration failure at 83%, and only 15% of cloud migrations finish on time and on budget.
🕳️ A matching row count proves the rows arrived. It doesn't prove a single value survived.
⏳ Whatever you validated Monday is stale by Tuesday, because the source system is still live.

So on October 14 we shipped Table Comparison tests in TestGen, open source and Enterprise.

🎯 TestGen profiles the source and turns it into tests: 345 on our Jaffle Shop migration, and they failed 61 times.
🔑 A four-column mapping CSV carries the tests to renamed tables and columns, even on a different database.
⚡ All 345 ran in five seconds, and every failure matched a change we made on purpose.

Read the full story here:
https://hubs.ly/Q04BgKck0

10/15/2026

Your data quality tooling should cost one data engineer's salary for a month, not for a year.

Most tools in this category charge by rows, tables, or compute credits. Every time you test more, the bill goes up, so teams ration coverage to keep it down. That's the opposite of what a quality program needs. TestGen is $100 per user and $100 per database connection a month, with unlimited tables and rows, published for anyone to read.

📉 On our pricing page, 10 users and 3 connections pay $15,600 a year. Per-table and per-credit tools run $120,000 to $360,000 for the same coverage.
⏳ The open-high, push-back, meet-in-the-middle dance costs weeks of meetings and your champion's momentum.
🔒 A negotiated price means somebody with a sharper negotiator paid less than you did.

So when procurement asked Gil for a discount, he didn't have one. He gave them a story to take to finance instead.

🎯 Three months of weekly consulting with a senior DataKitchen person, on how to be good at data quality, not just how to install TestGen.
🔑 A six-month start if you want one, and your own contract paper.
📈 Or start free: the open source version is one user, one connection, and every feature.

Read the full story here:
https://hubs.ly/Q04zVf330

10/14/2026

Who needs a third manifesto?

Your customer found the bad data and emailed you at 9am, copying your boss.

The night before, every job in the pipeline finished green, and the one person who checked thought the row counts looked okay. Last year's data quality assessment sits on the shared drive: 80 pages, a maturity score on page 4, and a roadmap nobody funded. We'd already written two manifestos that tell you to test your data. Neither one says how.

📉 "Quality is paramount" is principle 15 of the 2017 DataOps Manifesto. True, and useless at 2am.
🔄 Engineers want the rules in Git. Stewards will never have a Git account. So the spreadsheet wins.
🕳️ Teams building a medallion architecture test the one layer they own, and the customer reports the rest.

So we wrote a third: the DataOps Quality Manifesto, 24 principles plus one, for the Monday morning after you agree with the first two.

🎯 One null check does four different jobs: at the source, at ingestion, in production, and in development.
🔑 Score what your customer cares about. The programs that scored the fax number column didn't survive their second budget cycle.
📈 The three fit on one line. The first is how you build, the second is how you watch, and this one is how you prove.

Read the full story here:
https://hubs.ly/Q04zVdNY0

10/13/2026

I built a sales dashboard in 8 minutes and 39 seconds, and I never opened a BI tool.

One sentence into Claude Code. It found the AdventureWorks warehouse on my laptop, read its 28 tables, wrote six queries, and published a Sales page. Nobody built a semantic layer or dragged a measure onto a canvas.

📉 The reseller channel brought in two-thirds of the revenue at a -1.5% gross margin. On your dashboard, the VP's first question is whether that number is right.
🕳️ Two columns in factinternetsales are empty in all 60,398 rows. A green checkmark would never tell you.
⏳ Most catalog sidebars answer "what is this?" from a description somebody wrote once. The data changes every night.

So I asked Claude for three more tabs, built from TestGen's open source MCP server.

🎯 Data quality: a 99.4% score, and all 128 tests on the columns the Sales page reads passed.
🔦 Monitors on all 28 tables for freshness, volume, and schema drift.
🔑 A catalog of 344 columns, with 27 flagged as personal data and their values never shown.

That took 35 more minutes. The prompt is in the post.

Read the full story here:
https://hubs.ly/Q04B2zRZ0

10/13/2026

📅 Tuesday, October 20, 12:00 PM ET / 4:00 PM GMT

AI doesn't fix data quality. It amplifies whatever quality you already have.

An AI with the wrong context doesn't hedge. It hands you a confident, well-written, wrong answer, and it reads better than anything your data team ever shipped. You already know how the data underneath goes wrong. Chip Bloche, principal architect of TestGen, will spend one hour, with no code, showing you where the AI on top of it breaks.

🕳️ A column quietly changes meaning, and the assistant keeps answering as if it hadn't.
🔄 The wording of the prompt changes the answer, so two analysts get two numbers.
🔒 A hallucinated fact lands in a board deck, and somebody asks you who owns it.

Those are data quality questions. You've been asking them for years, and AI makes them more expensive to ignore.

🎯 What a prompt is, what "context" means, and why both change the answer.
🔦 A live check of AI's work: TestGen profiles the data the AI draws on, tests it, and scores it, so "grounded" means evidence you can point at.
🔑 Register once and the recording comes to you, live or not.

Register Now:
https://hubs.ly/Q04zVdfQ0

10/12/2026

Your data is wrong—and you only find out from a user. UGH!'

Why? Because nobody wrote a data check for it, since nobody has time to write tests.

TestGen writes data checks fast. Point it at a database, and it profiles every table, generates the tests, and runs them. It catches freshness, volume, schema drift, and bad values.
Hundreds of tests in minutes, not weeks.

It’s open source. Apache 2.0. One pip install, no sales call, no credit card. Run it on your own data today (https://hubs.ly/Q04zWBZ-0) and see what it catches. Unlimited tables, all the tests.

When you’re ready for more, TestGen Enterprise adds what your boss cares about: support, security, and a flat annual price. It avoids per-row metering. Know the cost before you start. Clear enterprise value, predictable pricing.

10/12/2026

How many hours does one data incident cost your team? Not a guess. Your number.

You don't have it. You have a feeling, and the feeling is "a lot," and the feeling is not going on a slide. The industry number is 13 hours per incident: four or more to detect, another nine to resolve, before anyone corrects a single downstream report. That comes from Wakefield's 2022 survey of 300 data professionals, and it is the one stat in this space that holds up.

📉 40 percent of a data professional's time goes to evaluating or checking data quality. More than half said incidents were going up, not down.
🔒 In our 2021 survey of 600 data engineers, 97 percent reported burnout. 69 percent said their own company's governance policies made the job harder.
🕳️ Gartner's $12.9 million a year comes from 154 reference customers of data quality vendors estimating their own losses. Quote it as an order of magnitude. Do not take it to your CFO.

So get your own number. Tag every ticket that came from something breaking, read the ratio at the end of the sprint, and track the lag from bad data landing to a human saying something.

🎯 Tests do not stop the Salesforce admin from adding a field on a Tuesday. They change when you find out: Monday at 9am from finance becomes Tuesday at 6:14am from a check.
⚡ Nobody has time to write tests for 4,000 tables. Nobody has to. Profile what the data does and generate them. Then keep the ones that make sense.
🔑 Split them on day one. Hard constraints fail loudly. Statistical checks run warn-only for two weeks, or somebody mutes the channel and the tests become decoration.

Read the full story here:
https://hubs.ly/Q04z8K7y0

10/11/2026

An analyst on the finance team has known for six years that when active members drop while claim volume holds flat, the eligibility file truncated.

She checks it by hand every month before the close. Your data quality tests do not check it. They cannot. The total member count fell 3 percent, which is under every threshold you have, and the entire drop sits in one plan that lost 60 percent of its members. Both tables pass. The failure lives in the relationship between two feeds, inside a segment nobody thought to slice, and both of those facts exist only in her head.

📉 Send domain owners a spreadsheet asking for their data quality rules and you get four rows back. Two say "data should be accurate."
🕳️ The engineer knows the load ran at 2:14am and the table landed at 4 million rows instead of 4.1. He does not know a duplicate contract ID means a re-keyed renewal.
🔒 The model sees a column go from 0.2 percent null to 3 percent and writes a clean test. It does not know whether anyone cares.

Nobody is good at authoring rules from a blank page. Reacting is easy. So stop asking people to author.

🎯 Show someone a rule that already exists and they will tell you in five seconds whether it is right, and that the threshold belongs at 1 percent because anything above that means the vendor file arrived twice.
⚡ You get better rules in a 20-minute review than in a six-week requirements exercise.
🔑 Generic tests catch broken pipelines. Business rules catch broken business. Statistical detection catches what nobody thought to check. You need all three, and you have one.

Read the full story here:
https://hubs.ly/Q04z8JQn0

10/10/2026

Three pharma payments in the CMS Open Payments file are dated November 30 in the year 2. Not 2002. The year 2.

CMS publishes every payment drug and device companies make to doctors: 15.5 million rows for 2024, $3.42 billion, 91 columns each. The file passed CMS validation. We loaded it into TestGen from Claude Code, let it generate 353 tests, then read the Sunshine Act, the OIG speaker-program alert, the PhRMA and AdvaMed codes, and Massachusetts law and wrote 16 more. Five prompts, start to finish.

📉 349 of 353 generated tests passed. Generated tests learn what the data looks like, not what the law allows. To them, the year 2 is just the minimum value.
🕳️ 13,890 physician records are missing an NPI, license state, or specialty. 7,729 of them are nurse practitioners and physician assistants, covered since 2021 and still missing from most HCP master data.
🔒 6,562 entertainment payments from 84 manufacturers, in a category both industry codes set at zero. Massachusetts bans them by statute. Its rate matched the national baseline.

The rules were written down. The data was public. Nobody had connected the two, so we did, with SQL tests that return the rows that break each rule.

🎯 3,450 meals over $500 per person. The most expensive one: $40,300 for a single attendee.
🔦 The most-fed practitioner took 237 meals from one orthopedics manufacturer in a year. Those are working days.
⚡ All 369 tests against all 15.5 million rows in 15 minutes 50 seconds on a Mac M1. A transparency team can run this in February instead of reading about it in July.

Read the full story here:
https://hubs.ly/Q04z8JGD0

10/09/2026

Nobody reads a 620-page book cover to cover. So we built this one to be read a chapter at a time, starting with whatever broke this week.

The DataOps Way to Data Quality and Data Observability is free. The September 2026 edition has 63 chapters in seven sections, and the newest section is on AI: context engineering, agents that fix data with control, and the seven deadly sins of AI-generated pipelines.

🔄 Your ingest, data science, and BI teams blame each other every time the pipeline breaks. Start with the culture section.
🕳️ The DAG turned green, and the first alert came from a business user weeks later. Start with monitoring and observability.
🧪 Your tests pass and the numbers are still wrong. Start with testing strategy.

Pick the chapter that matches your week. Skip the rest until you need it.

📈 54 of the 63 chapters are on https://hubs.ly/Q04z8JfK0 in full. Nine were written for the book.
🎯 The practices work on any stack. The two tools that match them, TestGen and DataOps Observability, are open source.
🔑 By Chris Bergh, Chip Bloche, and Gil Benghiat. Revised September 2026.

Get the free PDF:
https://hubs.ly/Q04z8JfL0

DataKitchen, Inc.

10/15/2026

10/15/2026

10/14/2026

10/13/2026

10/13/2026

10/12/2026

10/12/2026

10/11/2026

10/10/2026

10/09/2026

Address

Website

Alerts

Contact The Business

Shortcuts

Share

Category

Listing takedown request