The Denominator Dispute
Two satisfaction figures—7.6 and 3.8—are headed for the same executive review. Reconcile their scales, populations and coverage before choosing the headline.
- ROLE
- Performance Reporting Analyst
- TIMEBOX
- 60–75 minutes
- COMPLEXITY
- Focused / 2 of 5
- ROTATION
- Brief 1 of 16
The decision has already reached your desk.
It is 08:20 on reporting day. The board packet says customer satisfaction improved to 7.6, while the weekly dashboard says it fell to 3.8. Both figures came from the released response table.
The response estate contains two survey sources with different scales. Survey submissions also arrive after the reporting cutoff, and the people who respond may differ systematically from the ticket and account populations.
At 14:00, the executive review needs one headline measure. You have time to rebuild the eligible response population, compare the two scales and test whether coverage changes the story—not to claim that the survey represents every customer.
Recommend the satisfaction headline, population and scale for the review—or explain why no single directional claim is yet supportable.
Use only responses knowable by 02 Dec 2024 at 08:05 ET. The headline must state its population and cannot imply that survey respondents represent all customers.
Run the brief as a controlled assignment.
The workbench mounts only the listed source neighborhood and supplies neutral starter worksheets. It does not grade the conclusion or reveal the mechanisms planted in the larger assignment.
- Open the dedicated brief workspace and confirm the brief ID in the queue.
- Establish table grain, cutoff and control totals before joining or modeling.
- Make at least two distinct evidence moves and test a credible rival explanation.
- Leave the requested polished artifact, then export the workspace or portfolio package.
The source files are shared; the draft workspace is not. Changing to another full assignment drops brief mode deliberately.
Real tables. A deliberately bounded neighborhood.
These Parquet files already belong to Meridian’s public 96-table estate. Use the mounted schema.table names in SQL or Python; download links are provided for learners working in a local DuckDB environment.
GRAIN / One submitted satisfaction response.
CAUTION / Two incompatible response scales coexist. Submissions after the analysis cutoff remain in the mounted snapshot.
ANALYSIS CUTOFF / 02 Dec 2024 / 08:05 ET
Questions to pressure-test—not steps to copy.
These prompts define the analytical territory without prescribing an order, technique or conclusion.
- 01
Which raw scores become comparable after a declared normalization rule?
- 02
How do response timing, survey source and duplicate business interactions affect the eligible grain?
- 03
Where does response coverage differ from the underlying ticket or account population?
Leave work another analyst can review.
Artifact presence can be recorded; analytical quality remains a human judgment. A complete brief has evidence, reasoning and a decision—not merely executed code.
- 01Governed metric table
Reproduce both opening figures, apply the cutoff and show raw and normalized results by source without mixing scales.
- 02Coverage check
Compare respondent coverage across at least two material ticket or account slices and identify one plausible rival explanation.
- 03Executive scorecard note
Deliver a polished headline, compact supporting table or chart, and a material limitation in no more than 150 words.
Review the reasoning after a real attempt.
The debrief does not contain an official answer. It identifies defensible analytical moves, common failure modes and questions a reviewer may use to challenge the handoff.
SPOILER-GATED REVIEW / REVEAL AFTER YOUR FIRST HANDOFF
DEBRIEF REVEALED / THIS MAY CHANGE HOW YOU APPROACH THE BRIEF
This is a measurement-governance problem. A defensible response separates scale reconciliation from response representativeness and keeps the reporting cutoff visible.
Defensible approaches
- Reproduce each raw result inside its own survey source, then use the supplied scale fields to construct a declared common-scale comparison.
- Apply response and warehouse-availability clocks to the frozen cutoff before interpreting any direction of change.
- Join through stable identifiers only after profiling collisions, then compare respondent composition with ticket or account controls without claiming that weighting automatically solves nonresponse.
Common traps
- Averaging raw scores from different scale maxima.
- Filtering on the response date but not on when the record became available to the warehouse.
- Treating a normalized respondent mean as an unbiased estimate for every customer.
Reviewer questions
- Can another analyst reproduce the eligible grain and both opening figures?
- Does the final sentence name the source, population, period and scale?
- Which plausible coverage mechanism could still reverse your interpretation?
CHALLENGE A COLLEAGUE
Pass the Brief
The link shares this spoiler-free briefing. It never includes your work, identity, or browser progress.SUGGESTED NOTETwo satisfaction figures. Two scales. One executive headline. I worked The Denominator Dispute, a Priority Brief from The Analyst.
