Data break to repair, before and after: what a Snowflake-only export loses about a pipeline fix

Short answer

A Snowflake-only export keeps what ran and when: query text, start and end times, the user or role that ran it. It loses almost everything that explains a repair: which dbt test failed and how many rows it caught, the Jira ticket that diagnosed the cause, the commit that changed the model, the rerun that passed, and whether finance accepted the corrected number. Query history records execution. The reasoning that turns a rebuild into a repair lives in dbt artifacts, Jira and often a chat thread.

Visual briefFull-size diagram
A timeline of four Snowflake query events with blank gaps beneath them; tiles drop in from dbt and Jira to fill each gap: the failed relationships test, ticket BI-217, the commit and the finance sign-off.
Illustrative incident at a fictional subscription software company. The queries show what ran; the tiles show why.

This before-and-after case is part of the data-break-to-repair records guide. To describe the same chain for an introduction, use the metadata packet; to judge how complete it is, use the completeness scorecard.

Before: the incident as Snowflake query history shows it

Illustrative. A fictional 290-person subscription software company runs dbt on Snowflake and tracks data work in Jira. One week, its customer churn model breaks after a CRM account merge. Here is what an export of Snowflake's query history alone would show about that week:

TimeUser or roleWhat the query did
Monday 05:40Transformation service roleRan a scheduled build of the churn model
Monday 05:52Transformation service roleRan a further set of queries against the churn model
Tuesday 14:18An analytics engineerRan several ad hoc selects against subscription tables
Tuesday 16:05Transformation service roleRebuilt the churn model with new SQL
Tuesday 16:09BI service roleRefreshed the churn dashboard extracts

Nothing in that table says anything went wrong. Monday's 05:52 queries were dbt tests, but query history does not mark which returned failing rows. Tuesday's rebuild looks like any other deployment. A reader cannot tell that a number on the board's churn dashboard was wrong for a day and a half.

After: the same incident with dbt, Jira and the sign-off around it

Now place the other records beside the same timestamps.

TimeSourceRecordWhat it adds
Monday 05:52dbt run resultsrelationships test on the churn model's account key: status fail, failures count above zeroThat something broke, where, and how many rows
Monday 09:15JiraBI-217 opened: "Churn model has subscriptions with no parent account"The symptom in business words, and who owned it
Tuesday 14:18Jira commentDiagnosis: merged CRM accounts left subscription rows pointing to retired account IDsWhy it broke; explains the ad hoc selects
Tuesday 15:50Git, linked to JiraCommit "BI-217 map retired account IDs to surviving account"What changed, tied to the ticket by its key
Tuesday 16:05dbt run resultsSame test: status passThat the fix worked against the test
Wednesday 11:30Jira commentFinance lead confirms churn for the month matches billingThat the business number was accepted

Every row in the "after" table explains a row in the "before" table. dbt's run results file lists each executed node by unique_id with a status, and test results carry a failures count. Jira links a commit to a work item when the commit message contains the key, which is how BI-217 reaches the code. The finance comment is the only record that the repair ended in an accepted number.

What went missing, and where it actually lives

Lost contextWhere it livesKept how long, typicallyWho can confirm it exists
Which test failed, and how badlydbt run_results.json per invocationOnly if the team archives artifacts from each runAnalytics engineering lead
What the test was checkingdbt manifest.json and the project codeAs long as the project is in version controlAnalytics engineering lead
The symptom in business termsJira ticket descriptionUsually for the life of the Jira projectJira project admin
The diagnosisJira comments, pull request threads, chatTickets long-lived; chat often notEngineer who fixed it
The code changeGit commit, linked by the ticket keyLife of the repositoryAnalytics engineering lead
Acceptance of the numberJira comment, email or a meetingVaries; often only memoryController or finance lead
What ran and whenSnowflake ACCOUNT_USAGE.QUERY_HISTORYLast 365 daysData platform owner

The last row is the irony: the system that holds the least explanation is also the one with a fixed window. Snowflake's QUERY_HISTORY view covers the last 365 days, so even the "before" version of an incident from two years ago may survive only if someone copied it out.

The context no system holds: a change in meaning

Six weeks later, the same team changes how the churn model treats paused subscriptions. Every dbt test still passes, because a data test is a query for failing rows and passes when it returns none; none of the four generic tests checks what "churned" means. The run is green, Snowflake logs a rebuild, and churn for closed months moves.

In the "before" version, that change looks identical to the BI-217 repair: a rebuild by the service role. In the "after" version it looks like nothing at all unless someone opened a ticket or wrote the definition change down. This is the context most often missing even from well-connected histories, and the one most worth asking about. A company that logged metric definition changes, or reopened tickets when finance disputed a number, holds a record of judgment that tests do not capture. The pattern resembles the reasoning in reconciliation break notes and is part of what AI training data usually lacks. Demand for any given company's records is judged only in SourceX's review.

Who can describe it, and who can authorize

The analytics engineering lead can say which of the "after" sources exist and for which years. The Jira project admin can describe the ticket history, and the controller can say where finance sign-off was recorded. A licensing discussion needs the CEO's authorization, usually with the CFO or general counsel involved, because warehouse tables can include customer records under contract terms the data team does not control. The Jira system page and the Snowflake MCP resource give background on those tools; SourceX does not claim a connector to either.

Confirm the company baseline: 50+ full-time employees at peak (contractors excluded), documented operating history, rights to license, and an authorized contact. Companies in the U.S., Canada and the EU qualify, as do companies in the UK, Australia, New Zealand, India, Singapore, South Africa, Latin America, the UAE and Saudi Arabia when at least 70% of the data is in English. Sign the Channel Partner Agreement first.

Unknown until SourceX reviews it under the company's own agreement: whether the rights cover these records, how many incidents have a full "after" version, and whether a buyer wants them.

Next step

Refer a company. No cap on rewards. Default reward on a company's first qualifying transaction: the greater of $50,000 plus 5% and 15% of Actual Supplier Payout, with no cap; further qualifying transactions first agreed within 12 months of the first Closing pay 15%. Eligibility and payment conditions apply, including the USD 100,000 Minimum Transaction, and a signed Schedule A or addendum can differ. A lead, meeting or signature alone does not trigger payment.

If a data team you advise keeps the "after" version of its incidents, not just the query log, and an officer will take the call, register as a partner, check who qualifies, and read the terms before you submit.

  1. Step 1Share your linkSend your personal link to a company you know.
  2. Step 2Company appliesThe company applies itself at /apply.
  3. Step 3Company signs and is paidThe company signs, delivers its data and is paid by SourceX under its agreement.
  4. Step 4You get your rewardthe greater of $50,000 plus 5% and 15% of the actual supplier payout, with no cap. A reward becomes payable after SourceX has paid the referred company under its signed agreement and the payout conditions in the program terms are met, including finance approval and your tax and payment details.

Common questions

Is query history useless on its own?

No. It is the most reliable record of what ran and when, which anchors every other piece. It simply cannot say why something ran or whether the result was accepted.

What if the team uses a different warehouse or tracker?

The same split usually holds. The warehouse logs execution, the transformation tool logs tests, and the tracker or chat holds the reasoning. Name the actual systems in the description.

Should I ask the data team to export these records so I can check them?

No. A partner never handles warehouse logs, artifacts or tickets. Ask which systems hold each piece and for how long, and leave any review to SourceX under the company's agreement.

Sources

  1. docs.snowflake.com/en/sql-reference/account-usage/query_history
  2. docs.getdbt.com/reference/artifacts/run-results-json
  3. support.atlassian.com/jira-software-cloud/docs/reference-issues-in…
  4. docs.getdbt.com/docs/build/data-tests
  5. refer.sourcex.si/terms

Free resources

Know a company with valuable proprietary data?

Become a referral partner, get your link and introduce the company’s CEO (in an owner-run company, usually the owner). Referral partners can join from most countries. We can't work with partners in countries or regions under U.S. sanctions or on U.S. restricted-party lists. The companies you introduce should be based in the U.S., Canada and the EU, or in the UK, Australia, New Zealand, India, Singapore, South Africa, Latin America, the UAE and Saudi Arabia with at least 70% of their data in English.

Refer a company →

I own a business

Explore licensing your company's data directly to SourceX for AI training and evaluation. Start a short assessment; no uploads needed.

Start an assessment