SBL-CORR-2026-001: The historical input-probe disclosure overstates coverage. Unload-only transmissions were not reliably recorded. Focus and blur handlers could run, and per-field failures were not fully counted. These reports cannot establish absence of input transmission outside their retained observations.
Not measuredabsence here is not evidence of absence
Other pages on this site
Signed-in visitors, other browsers and other regions
Anything that loads only after interaction or consent
Services the catalog does not name
Checks with incomplete evidence
one request on record, same site
one request on record, cross-site
not measured
11 distinct catalogued tracking-service entities recorded responses across 41 cross-site domains. A catalog match identifies the service domain, not why the request occurred or what it carried. In the visit labeled GPC off, it also looks like keyboard-input listeners were registered through a call chain that included a cross-site script and 1 browser-fingerprinting heuristic matched. SBL-CORR-2026-001: The historical input-probe disclosure overstates coverage. Unload-only transmissions were not reliably recorded. Focus and blur handlers could run, and per-field failures were not fully counted. These reports cannot establish absence of input transmission outside their retained observations.
The scan observed signals a non-expert should not have to decode from raw request tables. Some detector evidence was incomplete; each affected detector card states its own retained or unavailable scope, while completed request, cookie, and storage measurements keep their recorded exactness.
Where severity is ranked, the cohort is a curated set of popular, mostly commercial sites, not a random sample of the web.Details
The cards below translate the evidence into plain language. Where a measured distribution exists, severity ranks this visit against percentiles from this report's cohort, with each percentile card naming its metric-specific measured-site denominator; otherwise it uses fixed reference thresholds. That cohort is drawn from a curated set of popular, mostly commercial sites, not a random sample of the web. The request log, domain table, and methodology remain below for verification.
285 requests observed in one controlled visit.
What was checked
Page
https://www.amazon.com/
When
Aug 14, 2026, 7:12 PM UTC
Setup
Both visits used headless Chromium in a desktop window of 1440 × 980, with en-US language and UTC time zone. Both ran from a scanner host the report does not name; the network region was not recorded.
Visits
Two visits configured to differ in the Global Privacy Control (GPC) signal: GPC off and GPC on.
Interactions
Typed a made-up test value into 1 form field in each visit, never submitting the form. No scrolling, clicking or consent interaction.
Outcome
Both visits completed, but some evidence was cut short.
What these visits could not see
Other pages on this site. Only the requested page's activity was recorded.
Signed-in visitors, other browsers and other regions. Both visits were automated, from a fixed en-US / UTC profile. Sites can behave differently for real users, browsers, regions, accounts, or network locations.
Anything that loads only after interaction or consent. Trackers that load that late are not observed, so counts are a lower bound.
Services the catalog does not name. Service labels use a US-biased hand-curated catalog, so regional services may be under-labeled.
Checks with incomplete evidence. In both visits: the typing test and the privacy policy check.
Worth checking
54 of 209 requests the engine evaluated matched Brave Shields filter lists
54 requests matched the default filter lists of Brave Shields, the ad and tracker blocker built into the Brave browser, while loading normally.
Blocking can also prevent follow-on requests, so this number is neither a measured block count nor what a Brave visit would block.Details
Computed with Brave's own ad-block engine and default filter lists in classification mode: matched requests were not blocked by the scanner and remain in this report's observed request counts. This is Site Behavior Lab's list-based simulation, which applies every list to the site's own requests too. Brave's default Standard setting skips most of these lists for such requests, so its results can differ.
The hand-curated service catalog separately matched 53 request rows in this visit.Show matched requests
Context
This GPC comparison supports no comparable delta
The pair did not record the network egress region for both visits, and an unrecorded condition never counts as matching.
The raw request logs of both visits remain below for transparency, but the diff between them supports no claim about what the compared condition changed.
1 disqualifying condition named by the report's claim policy.
Review-worthy
Catalogued service domains recorded responses during this visit
Amazon Ads, Google and OpenX, plus 8 others appeared in the request log.
A catalog match identifies a maintainer-reviewed service/domain mapping; it does not establish why an individual request occurred, what it carried, or whether profiling happened.Details
Functional catalog labels include advertising, advertising / exchange and analytics / advertising. This scan could not identify 30 of the 41 third-party hosts recorded here, so it cannot say who operates them. That is a limit of identity coverage, not evidence about the site.
232 cross-site requests across 41 third-party hosts.At or above the 90th-percentile mark for third-party domains across the 51 sites measured for this metric.Show third-party requests
Context
Requests were dispatched to catalogued major-platform domains
This visit dispatched requests to catalogued domains for Google.
The domain matches identify services, not the requests' purpose, payload, or whether any profile linking occurred.
53 third-party cookies showed up during the visit.
This report does not retain cookie values or partition keys, so it does not establish whether a cookie was a persistent identifier or could recognize a visitor across sites.Details
The scanner observed cookie metadata whose domain crossed the scanned site's registrable-domain boundary.
64 cookies total in this report.At or above the 90th-percentile mark for third-party cookies across the 51 sites measured for this metric.
Context
Input-monitoring signal involving a script from another domain
Listener coverage that could observe typing-related input events was registered through call chains that included a script served from another registrable domain. This scan did not establish the operator of that script.
This is a behavioral instrumentation signal from listener registration, stack-attributed script origins, and known-vendor requests: it shows the page was instrumented to observe interaction, not that anything was transmitted.Details
Stack attribution records the script origins present in the bounded registration call chain. A cross-site origin in that chain does not by itself establish that the cross-site script registered the listener. That is because first-party code can register its own listeners through a helper served from elsewhere. A different registrable domain is also not by itself a different operator: sites commonly serve their own scripts from a separate asset domain. The card therefore says third party only when the catalog or reviewed ownership attributes an origin to another organization. On scanners that run the active keystroke-capture probe, actual transmission is tested separately (a synthetic value is typed, never real input, and no typed values are collected). Treat this card as a review prompt rather than proof.
Session-recording listener coverage: 307 interaction-listener registrations with https://m.media-amazon.com/ in the registration call chain. Input-monitoring listener coverage: 122 input-listener registrations with https://m.media-amazon.com/ in the registration call chain.
They are review prompts for this visit, not proof of cross-site identity tracking.Details
These heuristics look for behavior patterns such as canvas readback after drawing, repeated canvas font measurement, WebGL entropy reads, offline audio rendering, or WebRTC peer-connection setup.
Canvas fingerprinting heuristic: 1 canvas matched; reads: canvas.getImageData and canvas.toDataURL.
Domain evidence
43 recorded domains
The busiest recorded domains of the GPC off visit, with role, request count, request start times and catalog match.
Domain
Requests
When, 0 to 9.4 s
Catalog match
Role
Resource types
m.media-amazon.com
138
No catalog match
third-party
document, stylesheet, script, image, font, fetch, media
www.amazon.com
34
No catalog match
first-party
document, xhr, image
*.amazon-adsystem.com
30
Amazon Ads: advertising
service
script, document, image
*.amazon.com
19
No catalog match
first-party
image, ping, xhr
*.*.doubleclick.net
5
Google: advertising
service
image
*.openx.net
4
OpenX: advertising / exchange
service
image
*.adnxs.com
3
No catalog match
third-party
image
*.mediarithmics.com
3
No catalog match
third-party
image
*.smartadserver.com
2
Equativ: advertising / supply-side platform
service
image
*.casalemedia.com
2
Index Exchange: advertising / exchange
service
image
*.demdex.net
2
Adobe: analytics / advertising
service
image
*.scorecardresearch.com
2
Comscore: audience measurement
service
image
*.3lift.com
2
TripleLift: advertising / supply-side platform
service
image
*.pubmatic.com
2
PubMatic: advertising / supply-side platform
service
image
*.360yield.com
2
No catalog match
third-party
image
*.bidswitch.net
2
No catalog match
third-party
image
*.connatix.com
2
No catalog match
third-party
image
*.zeotap.com
2
No catalog match
third-party
image
*.fwmrm.net
2
No catalog match
third-party
image
*.semasio.net
2
No catalog match
third-party
image
*.exelator.com
2
No catalog match
third-party
image
*.adscale.de
2
No catalog match
third-party
image
pixel.rubiconproject.com
1
Magnite: advertising / supply-side platform
service
image
*.taboola.com
1
Taboola: advertising / recommendations
service
image
*.ssl-images-amazon.com
1
No catalog match
third-party
script
*.postrelease.com
1
No catalog match
third-party
image
*.agkn.com
1
No catalog match
third-party
image
*.yahoo.com
1
No catalog match
third-party
image
*.dmxleo.com
1
No catalog match
third-party
image
*.ads.*.thisisdax.com
1
No catalog match
third-party
image
+13 more in the evidence explorer below.
Raw evidence
Open the interactive evidence explorer
The findings above are built from the stored report. Load the request rows, domains, cookies, storage keys, browser signals, visit phases, charts and CSV export when you want to check them; the versioned report JSON is always available below without this interface.
This versioned report is currently retained in the public corpus. Follow the currently retained site history, repeat the same public route when it is available, or request a transparent evidence correction.
cut short: detector output evidence was censored before completion: the synthetic keystroke probe did not fully test every field it found, or could not read every request in full (recorded loss count: 1); 1 privacy-policy visit did not produce usable evidence; recorded by the scanner
cut short: detector output evidence was censored before completion: the synthetic keystroke probe did not fully test every field it found, or could not read every request in full (recorded loss count: 1); 1 privacy-policy visit did not produce usable evidence; recorded by the scanner
Recorded provenance is self-reported measurement metadata, not a cryptographic attestation. The public versioned report JSON contains the retained public evidence; see the methodology for interpretation and retention limits.
This report covers what the scanner measures, not everything a site can do. 17 browser surfaces are not instrumented at all (Every page except the one that was scanned; Anything that only loads after scrolling or interaction; Why a request happened, as a causal chain; and 14 others), 3 capabilities are declined by policy, and 3 are outside what any single visit can see. Their absence from this report is not evidence that they did not occur. Read the published coverage boundary.
Detector accuracy is unmeasured. No calibration study currently publishes a precision or recall for the detectors on this page, so a detector match here is a recorded observation, not a calibrated classification. How detector accuracy is measured.
Severity percentiles come from this report's exact schema, methodology, tracker-catalog, ServiceRole-taxonomy, metric-contract, producer, and Global Privacy Control cohort. Each percentile card's measured sites are among 99 sites with a successful load across all cohorts. Other methodologies and post-choice consent visits are excluded from each card's denominator, while failed or block-page attempts are outside that coverage. Visits with capped or incomplete request evidence are left out of each mark. They tend to be the heaviest pages, so a mark can sit lower than the cohort's true mark.
Approved use boundary: This report is investigative evidence that requires independent corroboration. It is not a standalone legal determination and not the sole exhibit in a legal proceeding. Recorded in this project's release manifest on .