A claim is only as useful
as the method behind it.

Define the question

State the behavior under investigation and the hypothesis being tested. Define success before collecting results: access, a valid response, and correct extraction are different outcomes.

Record the setup

Document software versions, configuration, test dates, network and browser environment, sample size, and controls. Keep sensitive identifiers out of public artifacts. Use owned or explicitly permitted test environments.

Change variables deliberately

Establish a baseline. Change one variable at a time where practical, repeat measurements, and report uncontrolled differences. A comparison of two clients with many differences can locate a mismatch without proving its cause.

Retain the evidence

Link sanitized captures, tables, diagrams, or fixtures that support each claim. Report retries, invalid results, timeouts, and failures alongside successes. Explain denominators and acceptance criteria; an HTTP 200 is not automatically a valid extraction.

Separate observation from explanation

Label direct observations, interpretations, and unresolved hypotheses. A fingerprint mismatch is observable; the detector's internal decision may remain unknown. Do not turn a single result into a claim about an entire provider or every browser.

Make limitations explicit

Identify missing controls, version dependence, target coverage, ambiguous signals, and conditions where the outcome may change. Comparative scores summarize the implemented checks; they are not universal measures of detection or reliability.

Provide safe reproductions

Use local fixtures, test pages, synthetic examples, and permission-based experiments. Publish enough setup and evidence to assess the investigation without packaging an operational bypass. Exclude credentials, private target details, exact evasion values, production bypass scripts, and CAPTCHA-solving recipes.

Review and disclose

Check that conclusions follow from the evidence, disclose vendor involvement or commercial relationships, and date substantive corrections. Article types describe the format—Experiment, Field Note, Analysis, Benchmark, Comparison, Guide, Research Note, or Project—rather than implying a level of proof.

Project pages document a tool and its scope. They do not stand in for a completed comparative investigation. See the disclosure policy and send a correction when something cannot be reproduced.