This is a sanitized reference harness, not the exact internal build used for the report. It shows the launch, check, and reporting contract so other teams can reproduce the method or audit the data.
Files
| File | Purpose |
|---|---|
| benchmark_harness.py | Connects to a browser over CDP, runs the check suite, and writes JSONL results. |
| aggregate.py | Aggregates JSONL results into check-level and full-profile rates. |
| requirements.txt | Python dependencies for the reference harness. |
| checks.json | The 47 check definitions used by both scripts. |
Quick start
python -m pip install -r requirements.txt
python benchmark_harness.py --manifest profiles.json --out results.jsonl --workers 4
python aggregate.py results.jsonl --out aggregate.csv
The manifest is a JSON array of profiles:
[
{
"profile_id": "ADI26-00001",
"browser": "Incogniton",
"cdp_url": "http://127.0.0.1:9222",
"platform": "Facebook",
"proxy_country": "NL"
}
]
What the harness does
- Connects to each browser profile over the Chrome DevTools Protocol.
- Collects the fingerprint surface used by the 47 checks.
- Records pass/fail, launch time, and resident memory.
- Writes one JSON object per profile to JSONL.
- Leaves account login and platform behavior out of the harness by design.
What it does not do
- It does not bypass platform access controls, automate logins, or manage accounts.
- It does not include vendor-specific launch flags or proprietary kernel patches.
- It does not guarantee account survival. Fingerprint consistency is one layer of a much larger operational problem.
Reproducibility notes
- Keep the browser version, OS build, proxy country, and hardware constant across browsers.
- Run each performance measurement at least 10 times and report the median.
- Separate check-level from full-profile reporting. Do not describe a check-level rate as if every profile passed.
- Publish the limitations and the sample size alongside the ranking.
Questions about the harness: research@antidetectdata.com.