The quarterly exploration produces a number. A shorter date range produces another number with a different quality indicator. The reporting draft puts both into the same comparison.
Before interpreting the difference, record what each request asked.
A shorter window can be useful. It also changes the period represented by the result. Treat that change as part of the analysis.
The worksheets below are proposed review methods for GA4 reports and explorations. Every property label, report observation, and record is synthetic. No account was queried. The examples explain comparison boundaries without claiming a business outcome.
Name the Constraint Before Choosing a Response
Sampling uses part of the available data to produce a scaled result. GA4 indicates the percentage used in the data quality icon. Google's sampling documentation explains this behavior.
Several other constraints need different labels. Thresholding withholds information to protect individuals. Google's threshold guide defines that purpose.
An (other) row groups less common dimension values when a supporting table exceeds its row limit. Cardinality describes how many distinct values a dimension has. Google's explanation of the other row covers both concepts.
Approximate distinct counting is another mechanism. Google documents HyperLogLog++ estimation for metrics including users and sessions. Its developer explanation describes that approximation.
Use the following proposed labels in a review record:
| Evidence under review | Working question |
|---|---|
| Sampling percentage | How much available data informed this request? |
| Threshold warning | Which requested breakdown cannot be established here? |
| Other-row warning or grouping | Which detail has been condensed? |
| Distinct-count metric | Which counting definition and approximation apply? |
Do not treat these labels as mutually exclusive. Preserve the messages you observe. Avoid replacing several mechanisms with a single note called “GA4 error.”
Record Current Limits With Their Configuration
As checked on 2026.10.28, Google lists event-level query limits of ten million events for standard properties and up to one billion for Analytics 360. Analytics 360 initially defaults to one hundred million. Its Explore interface offers more detailed results through the quality icon. The current sampling guide supplies those limits.
The same guide notes a country-filter processing exception for large datasets. Property type and query shape therefore belong in the request record. Do not turn a headline limit into a promise about every report.
Record the documentation check date separately from the report run date. A remembered limit is not configuration evidence.
This worksheet does not calculate whether an unseen property should sample. It asks the analyst to retain the property's type, request definition, and displayed sampling evidence.
If two runs disagree with your expectation, preserve both observations. Name the unresolved question for the analyst reviewing the property. Avoid changing the numbers to fit an assumption about the limit.
Leave account upgrades and configuration changes outside the comparison decision. A report reviewer can document a limitation without approving a purchase or changing the property.
Capture One Fictional Exploration Request
The fictional service company wants to compare assessment-page use across two quarters. Its analyst begins with one quarter's exploration.
The following record is illustrative. Its sampling percentage is invented to demonstrate the field, not derived from a real event count.
| Field | Fictional request A |
|---|---|
| Business question | How did assessment-page active users differ between quarters? |
| Property | Synthetic service site; standard GA4 property |
| Surface | Free-form exploration |
| Period | 2026.07.01 through 2026.09.30 |
| Reporting time zone | UTC for this example |
| Dimensions | Page path and device category |
| Metric | Active users |
| Filter | Assessment page path |
| Segments | None added |
| Sampling evidence | Fictional quality indicator shows 40% of available data used |
| Other constraints | Not established by the example; inspect separately |
| Capture reference | Request A screenshot in the analyst's controlled review folder |
| Comparison status | Previous quarter not yet reviewed |
The forty-percent figure describes the invented indicator. It is not the share of users who visited the page. Do not multiply the displayed metric to compensate for it.
Name the metric precisely. Keep users, sessions, and events in separate fields if the request includes them. A generic “traffic” label cannot establish what two results count.
Record filter operators and values in your working copy. Retain enough detail for another analyst to reproduce the intended request. Exclude private identifiers from a broadly shared presentation.
Preserve the Original Before Shortening the Window
Create request B as a separate record. Keep A intact.
For the fictional example, B covers September alone. It retains the property type, dimensions, metric, page filter, and reporting time zone.
| Comparison field | Request A | Request B |
|---|---|---|
| Period | July through September | September only |
| Calendar days | Ninety-two | Thirty |
| Synthetic indicator | 40% of available data | 100% of available data |
| Analytical scope | One quarter | One month |
| Original comparison ready? | Previous quarter still unreviewed | Does not answer the quarterly comparison |
July has thirty-one days, August has thirty-one, and September has thirty. Request B removes sixty-two days from the original period.
Those arithmetic facts describe the invented date windows. They do not establish how activity was distributed across them.
The analyst can inspect September as a monthly question. The analyst cannot substitute September for the quarter without changing the brief.
Write that choice beside the new result. If the decision owner needs the original quarter, keep the quarterly question open. A cleaner indicator on a different period does not complete it.
Do Not Add Distinct Counts Across Windows
Breaking the quarter into months introduces another question: can those metric values be added?
For distinct people, repetition across periods matters. Use this tiny synthetic record set to see the issue without any GA4 estimation:
| Invented record | July membership | August membership |
|---|---|---|
| A | Included | Included |
| B | Included | Absent |
| C | Absent | Included |
July contains two distinct records: A and B. August contains two: A and C. Together the periods contain three distinct records: A, B, and C.
Adding two plus two produces four period memberships. It does not produce four distinct records across the combined interval.
These are abstract set members. They are not GA4 user identifiers or an implementation of Active users. The example isolates why an analyst must understand a metric before combining windows.
Approximation does not remove that definition problem. Neither does an unsampled indicator. Request the intended combined-period metric when that is the question under review.
If a report requires additive quantities, define those quantities explicitly. Do not relabel distinct-count values to make the spreadsheet reconcile.
Decide Whether Two Results Support a Comparison
Use a readiness checklist before calculating a change. This proposed checklist concerns the interpretation of the two retained requests.
- Write the decision and the two periods being compared.
- Confirm the property, reporting time zone, and report surface for each request.
- Compare metric names, dimensions, filters, segments, and relevant configuration.
- Copy each quality indicator and its sampling information.
- Record threshold and other-row evidence separately.
- Identify definition changes and incomplete evidence.
- Ask the decision owner whether the remaining limitation permits the intended use.
A completed checklist can still end with “comparison deferred.” That is a valid finding when one request has not been inspected.
For the fictional quarterly question, the earlier quarter remains missing from the record. Neither the quarter-to-month contrast nor the September result can establish quarterly growth.
If another dashboard supplies the missing quarter, freeze its definition before comparing values. The dashboard reconciliation worksheet helps document those differences.
Keep an unexplained residual visible. Do not assign every discrepancy to sampling because sampling appears in one view. That explanation needs evidence specific to the compared requests.
Put the Caveat Beside the Decision
For request A, a proposed report note reads:
“The quarterly exploration uses sampled results. The retained synthetic indicator reports forty percent of available data. The previous quarter has not been reviewed. No quarterly change is calculated.”
For request B, use a different note:
“The September exploration reports one hundred percent of available data for this request. It replaces the quarter with a monthly scope. It does not establish the quarter's result or guarantee exact distinct counts.”
These statements describe invented observations. Adapt them only after inspecting your own requests.
Google's quality guide describes the unsampled state as using all available data for the selected dimensions and metrics. Read that state in its report context. “Available” does not establish that every intended business interaction was collected.
Keep collection completeness as a separate evidence question. A report indicator does not prove the site's implementation matches the business definition.
For configuration work, use a separate GA4 setup review. Preserve this comparison record as the evidence of what the analyst inspected.
Keep Measurement Claims Within the Evidence
A sampled analytics result cannot establish an accounting result by itself. If the decision concerns return on investment, define the revenue and cost records separately.
The B2B organic search ROI article covers that neighboring definition work. This worksheet establishes neither revenue nor causation.
Before presenting an apparent improvement, ask what changed between requests. A different date range, filter, or metric is part of the explanation. Record it before assigning a business meaning to the difference.
When the evidence remains incomplete, describe the next inspection. For example: review the earlier quarter with the same recorded dimensions and metric, then assess whether comparison is suitable.
Do not promise that the next request will be unsampled. Do not promise exact counts after a warning disappears. The decision can remain provisional while the analyst gathers the necessary evidence.
Leave a Comparison Record Someone Can Continue
Store the original question, request A, request B, indicator captures, caveats, and owner decision together. Give each request a distinct reference.
Record the chosen reporting scope in the presentation draft. Mark the quarterly comparison deferred if the owner selects the September review instead.
At follow-up, append the new request definition and observation. Avoid overwriting the earlier percentage or replacing its screenshot. A later reader needs to know which evidence supported the earlier decision.
Have another reader reconstruct the conclusion from the record. They should identify the unanswered quarterly question, the monthly alternative, and the counting restriction.
If they describe September as the quarter, revise the heading. If they add distinct counts across periods, revise the calculation. If they interpret an indicator as proof of complete collection, revise the caveat.
Keep the query attached to the number.
Sources
- Google Analytics Help: About data sampling. Sampling, indicated percentage, event-level limits, property distinctions, and query caveats. Checked 2026.10.28.
- Google Analytics Help: Data quality. Available-data meaning and indicator context. Checked 2026.10.28.
- Google Analytics Help: About data thresholds. Disclosure protection differs from sampled analysis. Checked 2026.10.28.
- Google Analytics Help: About the other row. Table limits, dimension cardinality, and aggregation. Checked 2026.10.28.
- Google for Developers: Unique count approximation in Google Analytics. HLL++ estimation for distinct-count metrics. Checked 2026.10.28; this article uses no precision-parameter claim.