Resource library
Benchmarks from 412 data teams, migration playbooks with the rollback plans still attached, and recorded sessions where an engineer breaks a live pipeline on purpose. Seventy-three of the eighty-one items here open without a form.
Featured
We asked 412 data teams what breaks, how long it takes them to notice, and what a monitored table actually costs once compute is counted. The question wording and the raw responses ship with the report.
Report
Incident volume, mean time to detection and cost per monitored table across 412 teams, split by warehouse and by team size. Median detection sits at 4 hours 12 minutes; the top quartile catches breakage in under 20 minutes.
Guides and templates
Each one was run at a customer site before it was written up, so it names the checks, the order to do them in, and the step that usually goes wrong.
Sixteen freshness and volume checks, five alert thresholds, and the escalation path that keeps an on-call rotation from burning out by month three.
Where the contract lives, who signs it, and what to do the first time a product team ships a breaking change to a source table anyway.
The cutover order Corvia Health used, the four checks that ran on both systems in parallel, and the rollback plan they never had to trigger.
How lineage is parsed out of query logs, the three cases where that parsing gives up, and what a human still has to label by hand.
The nine queries we run against Snowflake and BigQuery accounts, plus the spreadsheet we hand back with the findings ranked by refund size.
A blameless post-mortem built for pipeline failures: blast radius first, root cause second, and a named owner on every action before the doc closes.
On demand
Working demos and post-mortems with the engineer who ran them. Slides, queries and the sample warehouse are linked under every recording.
A solutions engineer drops a column from a running Snowflake model, then walks the blast radius out to every dbt model, dashboard and reverse-ETL sync it touches.
Their staff engineer walks the cutover plan, the two parity checks that failed on day six, and the one decision she would make differently.
Thirty minutes of questions from data leads building an internal business case, answered without a slide deck and with the awkward ones left in.
Customer stories
Every number below was checked by the team that produced it. Where a result was unflattering, it stayed in.
Fintech 480 employees
Four hand-run reconciliation jobs became one governed pipeline, and finance stopped re-checking figures against last month by hand.
Close in 3 days, not 8
Healthcare 1,200 employees
A decade of untracked SQL moved onto column-level lineage without pausing a single downstream dashboard or freezing the dbt repo.
1,400 models in 9 days
Logistics 860 employees
A six-person revenue operations team retired a weekly reconciliation ritual in two quarters and stopped writing stale fields into Salesforce.
11 syncs, 0 bad writes
Financial services 5,200 employees
Access policy, approval trail and lineage now come out of one system, so the annual regulatory review takes a week instead of a quarter.
Audit evidence in 1 export
From the blog
Release notes with the reasoning attached, plus the occasional public argument about how lineage ought to be modelled.
Stay current
What shipped, what broke, and which of these guides got rewritten because it turned out to be wrong. Written by the same engineers who answer support — around 600 words, no digest padding, no product news you could have read in the changelog.
A first name and a work email. That is the whole form.
Unsubscribe in one click. The list is never sold, shared or handed to sales.
Prefer a different channel?
