Spot missing, repeated and changed analytics requests

Compare the requests, then reveal what to investigate

Open a case below and compare its before and after columns. Decide what you would check, then open Show what to investigate. These are fixed teaching examples from the free practice kit, not a parser for uploaded HAR files. No account or purchase is needed.

What the three differences tell you

• Missing event: add_to_cart goes from one supported request to none. Check the tag trigger, browser errors and the capture conditions. A missing request is a clue, not proof that the release caused it.
• Repeated request: add_to_cart goes from one request to two identical requests. Investigate a duplicate trigger or handler. One extra occurrence does not prove the analytics service counted it twice.
• Changed parameter: the fictional view_item request changes its cu value from EUR to USD. Check the intended currency and compare the payload; the CLI output deliberately withholds the values.

Run the same examples on your computer

Download the free MIT-licensed HAR practice kit and extract it. The ZIP includes source, tests, a README, a diagram and four originally authored fictional HAR files. Node must already be installed; the tested version is 22.22.2 on Linux. Open a terminal in the extracted kit folder and run:

node har-diff.mjs fixtures/baseline.har fixtures/duplicate.har
node har-diff.mjs fixtures/baseline.har fixtures/missing.har
node har-diff.mjs fixtures/baseline.har fixtures/parameter-change.har

• The duplicate comparison reports add_to_cart changing from 1 to 2 supported requests and repeated_after of 1.
• The missing comparison reports add_to_cart removed.
• The parameter comparison reports view_item changed. Inspect the fictional HAR to see EUR change to USD; the reduced CLI report withholds parameter values.

The CLI reads a narrow GA4-style query/form format with one event per supported request. It counts unsupported, malformed and outside-scope requests. Adobe, GA4 batches and custom collectors are outside its coverage. Its limits are 8 MiB and 20,000 entries per file. See the bundled README for parsing rules and the example output.

Files are compared locally without upload or rewriting. Reports withhold raw URLs, parameter keys and values, and custom event names. Real customer captures, other Node versions and other platforms were not part of the test. Review even a reduced report before sharing it.

Keep a real before/after test comparable

1. Choose one short journey on a site you are authorised to test, with a test account and test data. Write down the pages, clicks and consent choices.
2. In Chrome DevTools, open Network and enable Preserve log. Walk the journey on the current build and select Export HAR (sanitized).
3. Repeat the same journey on the next build using the same consent choices and browser conditions. Export a second HAR and name the files clearly, for example checkout-before.har and checkout-after.har.

Chrome’s Network reference explains that sanitized HAR export excludes Cookie, Set-Cookie and Authorization headers. URLs, query strings, form fields, response bodies and analytics parameters may still contain sensitive data. Keep captures private, use authorised test data and delete files when you no longer need them.

Separate request evidence from analytics results

A HAR shows browser request evidence. It does not prove analytics receipt, tracking correctness, consent compliance or equivalence between two journeys. A blocked request, an extra click or a different consent choice can change the result.

For Adobe Analytics, inspect the relevant requests and payloads manually in Network or use an appropriate separate tool. The free CLI on this page does not analyse Adobe requests. Unsupported traffic needs its own check; it is not a clean result.

Turn a difference into a useful investigation

• Confirm the same steps and consent choices were used in both captures.
• Check removed events and changed parameters on the important steps first.
• Investigate repeated requests and confirm that added events were intended.
• Review unsupported and malformed traffic separately.
• Verify collection in the relevant analytics service before concluding that tracking works.

Start with the fictional lesson when learning the result format. For your own captures, record the difference, the test conditions and what you checked next; keep sensitive payloads out of shared notes.