Research record · June 11, 2026

What we checked, what we counted, and where the data stops.

This is the methodology and limitation record for one bounded scan of 50 affiliate publisher sites. It documents observable link behaviour—not market-wide prevalence, customer outcomes, or revenue impact.

One observed sampleChecked-link denominator onlyNo revenue estimates

6,550

outbound links checked

11,676 found; 5,126 outside the time budget

4.5%

visibly broken

295 non-blocked 4xx, 5xx, or timeout observations

5.7%

could not be verified automatically

372 links where the destination blocked or throttled the checker

34/50

sites with substantive coverage

The other 16 returned fewer than three public pages and limit site-level conclusions

All rates above use the 6,550 links actually checked within budget. The 5,126 skipped links are not silently treated as healthy.

Collection method

A repeatable crawl, with the shortfalls left visible

The process was deliberately bounded. Partial coverage is part of the result, not a reason to convert missing observations into zeros.

  1. 1. Build a fixed research panel

    The panel contained 50 established affiliate publisher sites across several niches. It was curated, not randomly sampled, so it does not represent every publisher or the whole market.

  2. 2. Discover a bounded set of pages

    The crawler checked sitemap entries first, then used a same-site crawl when needed. It respected robots.txt and capped each site at 20 public pages.

  3. 3. Enforce the same time limits

    Each site received a 100-second soft budget and a 300-second safety stop. 23 sites returned partial results; skipped links never entered a rate denominator.

  4. 4. Follow each checked link

    The checker started with HEAD, retried suspicious responses with GET, allowed up to 5 redirect hops, and used a 10-second request timeout.

  5. 5. Keep blocks separate from breaks

    372 links returned 403, 405 or 429 after the fallback, landed on a bot or captcha page, or were throttled by Amazon with a 5xx. They were labelled unverifiable and excluded from the visibly-broken count. Share buttons and photo credits are not outbound content and are excluded from findings.

Counting rules

Six labels, three reporting groups

A status-code failure and a tracking failure answer different questions. They are shown separately and never converted into a financial claim.

ObservationCountChecked linksReporting group
Non-blocked 4xx1502.3%Visible break
5xx server response180.3%Visible break
Timeout1271.9%Visible break
Affiliate link to an expired partner program60.1%Affiliate dead end
Tracking tag never reached network or merchant00.0%Affiliate dead end
Deep link redirected to homepage280.4%Lost context
Bot-blocked or rate-limited3725.7%Reported separately

Correction, September 21, 2026

This page originally reported 597 attribution failures, 569 of them tracking parameters that disappeared by the final URL. That was wrong. Our classifier counted any missing parameter as lost — including social share buttons, photo credits, and affiliate networks that consume their own parameters and record the click, which is how they are meant to work. Re-checking the same stored observations with corrected rules found no case of a tracking tag failing to reach the network or merchant. What did hold up: 6 affiliate links sent readers to an expired partner program while still loading normally, and 295 links were visibly broken (originally reported as 380; Amazon throttling responses are now counted as unverifiable). The 6,550-link denominator still includes the excluded share buttons, so the corrected rates are slightly understated.

Read before citing

What this study cannot establish

The honest use of a bounded study is narrow. These findings describe the links observed on this panel, on this date, under these crawl limits.

Not a representative sample

The 50 sites were curated. The percentages are not a universal rate for publishers, niches, or the web.

Not complete site inventories

23 sites exhausted the soft budget, and only 34 returned at least three pages. Results are bounded observations, not full audits.

Not proof of revenue impact

The crawler did not have commission rates, traffic, conversion data, account access, or customer outcomes. No dollar inference is supported.

Not permanently current

Links, redirects, servers, and bot rules change. The record describes June 11, 2026 and should not be presented as a live measurement.

Apply the same evidence standard

Check a site without pretending the crawler knows its revenue.

The free leak snapshot reviews a limited set of public pages and reports observable link behaviour. Fit for deeper work is reviewed separately.

Get a free leak snapshot