Future InTech

Log queries and report integrity · FIELD GUIDE

Use row weights in Workers Analytics Engine metrics

Calculate represented counts and weighted averages using each row’s sample interval instead of applying one assumed sampling factor.

By Future InTech · Published
Evidence cutoff: 2026-11-02T23:15:29.254292+00:00 · UTC

What the evidence can establish

Cloudflare documents _sample_interval as a per-row value related to the inverse sample rate and the number of original data points represented by a sampled row. The interval can vary between rows, so multiplying a total by one constant is not sufficient. The documentation also shows a weighted average using the sum of value times interval divided by the sum of intervals. Define the measure first, then apply the appropriate row weights consistently.

Hypothetical arithmetic: one row has value 2 and weight 10; another has value 8 and weight 30. Their represented count is 40 and weighted mean is (2×10 + 8×30) / 40 = 6.5. The unweighted mean is 5 and answers a different question about the two returned rows. This calculation is an instructional fixture, not data from a Cloudflare account.

Row A: value 2, weight 10; Row B: value 8, weight 30; Represented count: 10 + 30 = 40; Weighted mean: 260 / 40 = 6.5
Original explanatory diagram based on the dated sources below. Any example values are hypothetical; this is not a screenshot, customer measurement or AI platform result.

The saved evidence used here

This guide interprets dated official documentation. It does not report traffic from a customer or a measured AI answer. Each material claim is mapped to a source below.

c1 · Explanation and verification steps
The sample interval is associated with the inverse sample rate. Source 1
c2 · Explanation and verification steps
A sampled point represents a number of unsampled data points. Source 1
c3 · Explanation and verification steps
Sample intervals can vary by row, so a constant factor is insufficient. Source 1
c4 · Explanation and verification steps
The documented weighted average divides the weighted sum by the sum of sample intervals. Source 1
c5 · Explanation and verification steps
Sample interval alone does not determine result accuracy. Source 1

The diagram explains the investigation or a clearly labeled hypothetical example. No measured counts or platform outcomes are added.

AI answer visibility: Not measured. No original platform answers or verified citation records were collected for this guide. A request count is not an audience, ranking or citation count.

What to do next

  1. Confirm that the chosen dataset and query use the documented _sample_interval field. Preserve the measure, its unit and the returned per-row weights in a restricted verification sample.
  2. For a represented count, reason from the number of original points each row represents. For a sum or average, apply the documented weighting formula to the intended numeric field.
  3. Check a tiny hand-calculated fixture before running the production query. Include unequal weights so an implementation that ignores weights cannot accidentally pass.
  4. Compare the query result with the fixture and retain the formula in the report method. Label sampled estimates appropriately instead of describing them as an unsampled raw-event export.

How to check the result

The unequal-weight fixture produces the expected result, the query names its measure and units, and the method accounts for weights that vary by row.

Limits of this conclusion

A larger sample interval alone does not determine accuracy. This guide covers arithmetic and query design, not a guarantee about the accuracy or completeness of any production dataset.

This is a saved, dated edition. Source documentation can change. This edition retains the original retrieval dates and claim mapping; follow the sources before applying version-specific guidance.

Sources and publication checks

Automated checks validate source availability, supported fields, duplicate wording, numeric figures and page configuration. This guide was prepared with AI assistance and checked against saved source passages in the same authoring session. The batch importer checks source fingerprints, quotation locations, wording overlap and required fields. No paid generation API or separate model-review pass was used for this batch. These checks can miss errors and are not expert or human review.

Sampling with WAE

Obtained: 2026-11-02T23:15:29.254292+00:00
Location: How to read sampled data / b19; How to read sampled data / b20; How to read sampled data / b21; How to read sampled data / b26; How to read sampled data / b28

Saved source fingerprint6b4025546fe66b06ae652cf8737252d84f8ecdc4260bc451cedee91a800af59d

Download this edition's evidence and action record (JSON)