NeuraGrowth, a one-person digital products studio, recently uncovered a critical flaw in its click-counting infrastructure that standard rate-limiting failed to catch. The discovery, detailed in a post on DEV.to by Rob Swierk, highlights how relying on HTTP Referer headers for traffic attribution can lead to significant data inaccuracies when modern privacy attributes like rel="noreferrer" are implemented correctly.

The Discrepancy Between Real and Fake Traffic

The investigation began after the team noticed a sudden drop in daily clicks from approximately 1,257 to 200 following the implementation of a per-address limit (P26). While the volume decreased, the pattern of the remaining traffic raised red flags. On November 9, 2026, nginx logs recorded 207 hits to the /api/go/ endpoint from 205 distinct addresses. Crucially, 195 of these requests carried a Referer header of https://neuragrowth.co, while 5 carried the trailing slash variant.

Playwright as the Ground Truth

To determine if this traffic was legitimate, the developer used Playwright to simulate a real user click on a "Buy" button. The test revealed that because the buttons are configured with rel="nofollow noreferrer", a genuine click from the site’s homepage results in a Referer header of "-" (empty). This proved that the 200 requests carrying the site’s own URL as the Referer were not coming from actual user interactions on the site, but were likely spoofed or generated by bots that ignored the noreferrer attribute.

Shipping P26b to Filter Spoofed Clicks

The mismatch meant that per-address limits were ineffective, as the spoofed traffic originated from a wide distribution of IPs, each making only one or two requests. To fix this, NeuraGrowth shipped patch P26b on November 9, 2026. The update, which modified two files with 87 lines added and one removed, now explicitly excludes any click that carries a Referer header from the internal counting logic. This ensures that only traffic without a Referer (consistent with noreferrer behavior) is counted toward the address ledger.

Key Takeaways

  • Verify your analytics assumptions with real browser automation tools like Playwright or Selenium.
  • rel="noreferrer" suppresses the Referer header; if your analytics rely on it, you are blind to genuine traffic.
  • IP-based rate limiting fails against distributed bot traffic that mimics legitimate user patterns.
  • Always log raw HTTP headers to debug discrepancies between expected and observed traffic.

The Bottom Line

If you are building analytics on top of HTTP headers without validating them against real browser behavior, you are likely counting ghosts. Use Playwright to audit your own tracking pipeline before your data becomes meaningless.