Run the preparation path
Follow the workflow through the join that connects account records to subscription activity. Choose the checkpoint relevant to the suspected issue.
Scientist · Interactive runs
Inspect the step before trusting the summary.
Follow execution and examine intermediate results in Scientist. Use the evidence at a meaningful checkpoint to decide whether preparation is doing what the study intends—before interpreting the final analytical output.
Interactive runs
Inspect a join before comparing renewal cohorts
Run the renewal preparation workflow and inspect the account-to-subscription join before continuing the analysis.
Execution point
Join accounts with subscription records
Intermediate view
Inspect account keys and matching subscription rows
Research check
Review repeated accounts before cohort aggregation
Examine an intermediate result at the step that produced it.
Follow the investigation
A concrete path through the research.
An account-to-subscription join can legitimately repeat an account. If the analysis later treats every joined row as a separate account, the cohort comparison can answer the wrong question.
Examine the record relationship
Inspect account keys and matching subscription rows. Distinguish expected multiple subscriptions from unexpected duplicate source records.
Revise or continue deliberately
Check whether the next aggregation uses the correct unit of analysis. Revise the preparation logic if needed, then inspect the relevant output again.
What to examine
Put the analytical choices in view.
The useful output is not just a completed run. It is an analysis whose inputs, assumptions and limits another researcher can question.
Checkpoint
The operator where the unexpected record relationship first becomes visible.
Evidence to inspect
Keys, repeated rows and the fields used by the downstream aggregation.
Next analytical move
A specific workflow correction or a reasoned decision to continue.
Scientist in the DO Stack
Take the research to the right next owner.
Sense helps investigate instrumented services when execution problems arise outside the analytical logic. Keep a data-method question separate from a runtime incident and confirm which telemetry can be connected to the investigation.
What should we validate for implementation?
Confirm intermediate-result availability, run controls and resource limits. Breakpoints, pause/resume, partial recomputation and restart-free debugging are implementation requirements to validate, not guarantees made here.
What should the team bring to a first discussion?
Bring one research question, its source requirements, a relevant workflow and the review criteria. Confirm available capabilities, access and operating requirements before using sensitive material or applying a finding operationally.
Start with a question worth investigating.
Bring the question, the available data and the people who understand it. We’ll work through the analytical path, the review points and the implementation requirements.