Practical framework / 5 MIN READ
Make an important number reproducible after the meeting
Attach the result to its inputs, transformation version and reporting context.
AI-assisted research and analysis for the SNS Creative Fieldbook
Sources checked 2026-11-06
A screenshot preserves what a chart looked like, but often cannot explain how it was calculated. Consequential marketing decisions need enough provenance for another authorized person to reproduce or reconcile the evidence.
What you will leave with: Create a compact reproducibility packet for one recurring decision report.
Use this when
- A budget decision depends on a number nobody can reproduce.
- A report owner leaves and the team loses the calculation context.
This lesson concerns evidence traceability, not storing every raw record or documenting every dashboard equally.
Choose the numbers worth preserving
Prioritize reports used for budget commitments, incentive decisions or repeated planning. Identify the specific measures and conclusions that would matter if challenged later. The required detail should match the consequence, not the visual complexity of the chart.
Write a short result description with units, period, population and exclusions. Save the report’s parameter values, including filters that may be invisible in a screenshot. Record whether values were provisional, modeled or subsequently corrected. Without that context, even the original query may answer a different question when rerun.
Trace the path from source to decision
List source datasets, transformation steps, mapping versions and the final report. dbt exposures provide one documented way to connect downstream dashboards or applications with dependencies and an owner. The principle can also be implemented in a controlled register for smaller workflows.
Include manual steps. A pasted adjustment or spreadsheet override belongs in the lineage if it changes the result. Record its reason, approver and calculation rather than hiding it inside the final total. Do not claim full reproducibility when an upstream provider can revise data and no authorized historical snapshot exists.
Save the minimum reproducibility packet
Preserve the query or transformation version, execution date, source cutoff, selected parameters and a permissible result artifact. Keep source identifiers and access instructions without copying credentials. If an immutable data snapshot is necessary, obtain approval for its scope and retention.
A result hash can help detect file changes, but it cannot prove the calculation was correct. Pair mechanical checks with a small reconciliation to an independent source or fixture. Record known limitations and unresolved differences so the next reviewer does not mistake a polished file for audited evidence.
Have someone reproduce the result
Ask an authorized colleague to regenerate one key total using the packet. Observe which missing assumptions force them to ask questions. Improve those points rather than writing a long generic document nobody uses. Where exact reproduction is impossible, require a clear explanation of the difference.
Connect subsequent corrections to the original artifact and notify decision owners when the correction matters. Keep an owner for the report and dependencies. A useful lineage record should support both impact analysis before a source change and a calm explanation after a number changes unexpectedly.
Make the reasoning concrete.
A hypothetical quarterly deck reports 240 qualified enquiries. The saved query returns 260 because the deck excluded 20 test records through an undocumented spreadsheet edit. The screenshot alone cannot explain the difference.
Preserve only the slide
Opening: Keep the 240 figure as the historical record.
Hypothesis: The meeting artifact remains available.
Risk: The exclusion cannot be checked or repeated.
Save the calculation path
Opening: Record the source cutoff, query version and approved test-record exclusion.
Hypothesis: Another reviewer can reconcile 260 to 240.
Risk: If the exclusion lacks evidence, the result still needs correction.
Validate the 20-record exclusion and preserve the bridge. If it was unjustified, issue a correction rather than manufacturing documentation that makes the old number look defensible.
Take the questions into your next brief.
Use these prompts in your own notes. Nothing is submitted from this page.
Consequence
Which decision depends on this result?
Inputs
What sources, cutoffs and versions produced it?
Manual steps
Which overrides changed the output?
Artifact
What minimum permitted evidence supports reproduction?
Verification
Can another person regenerate or reconcile it?
Watch for the wrong conclusion.
- Treating a screenshot as complete provenance.
- Leaving manual overrides out of the record.
- Saving credentials beside the query.
- Confusing file integrity with calculation correctness.
Your next action
Build a reproducibility packet for one important chart and have an authorized colleague rerun its main total.
Sources & scope
Frameworks, worksheets and hypothetical examples are original editorial synthesis. Linked primary sources support only the stated concepts. Source checking is not an independent audit of effectiveness. No named human author or reviewer is claimed.
- dbt: Add exposures to your DAG →Primary technical documentation · Supports: Documents downstream dashboard/application dependencies and ownership metadata. The reproducibility packet and reporting controls are original editorial practice.