container audit
GTM-N345CDJX
153 tags, 30 custom scripts, 4 third-party hosts — an ordinary container, near the middle of what we measure (median: 39 tags, 4 custom scripts).
- 30 custom HTML tags carrying 57 KB of JavaScript — more than 80% of the 46,522 containers we have measured. This runs on every page the container loads on.
- 4 third-party hosts reachable from its tags.
- Measurement runs through a server container on its own domain — the mark of a mature setup, and rarer than the tooling vendors imply.
Tags153
Custom scripts30
Third parties4
Delivered681 KB
Versions112
What we observed
A real browser loaded madeleine.com a month ago and recorded every request it made. This is what happened, as distinct from what the container is configured to do. It is one page load — anything that fires only deeper in the site did not fire here.
Container loaded from
mvgt.madeleine.com — the site's own hostMeasurement sent to
mvgt.madeleine.com — not GoogleBoth wires run through infrastructure the site controls.
Forwarded to
GA4 ×9 — read from the measurement id, not the hostname
Consent
Usercentrics answered its own API. Nothing measured until it was granted — no analytics or advertising hit left the page beforehand, and 9 fired afterwards. That is the outcome consent is meant to produce. 10 hosts appeared only after consent was granted. Google recorded
gcs=G111 (all granted).Third parties contacted
25 hosts — bj3yocjsex.kameleoon.io, web.cmp.usercentrics.eu, a.storyblok.com, eu-data.kameleoon.io, vimeo.com, v1.api.service.cmp.usercentrics.eu, i.vimeocdn.com, app.usercentrics.eu, and 17 more
dataLayer events seen
content-view, htf, htf-ags-data-dataLayer-init, htf-ags-data-dataLayer-browsed, consent_status, virtual_page_view, experiment_triggered, content_view, shop_init, view_promotion, gtm.js, gtm.dom, +2 more
Where the data goes
Confirmed — a browser watched this site send measurement data to mvgt.madeleine.com, not to Google. observed on madeleine.com More info »
Findings
Ordered by consequence. review has a concrete cost when it bites; info is real but depends on your setup. Seen is how common it is across the containers we have measured — reported separately on purpose, because common and harmless are different claims.
reviewdocument.write (10)
10 pieces of custom code call document.write.
Whydocument.write blocks the HTML parser while it runs, and browsers ignore it entirely once the page has finished parsing — so an async-loaded tag using it can silently do nothing.
CheckIf these tags fire on page load they are costing render time; if they fire later they may not be working at all. Worth confirming the tag still does what it was added to do.
Seenseen in 18.4% of 46,522 containers measured
reviewWrites cookies directly (1)
1 piece of custom code set cookies in code.
WhyCookies set this way bypass GTM's consent settings, because the consent checks live in the tag templates rather than in raw JavaScript.
CheckIf any of these cookies are used for analytics or advertising, confirm they are still gated on a consent decision.
Seenseen in 21.2% of 46,522 containers measured
infoLoads JavaScript from another host (4)
4 pieces of custom code fetch and run JavaScript from another host.
WhyWhatever that host serves executes with full access to the page — it can read the DOM, cookies and form fields. The code is not in this container, so it can change at any time without any change here.
CheckConfirm you recognise every host below and that someone owns the relationship. A host you cannot name is the one to chase.
Seenseen in 54.4% of 46,522 containers measured
infoConsent configuration detected (2)
The container declares one or more consent signals.
WhyConsent handling is present in the container. This describes what the artifact declares; what actually happened on the page is measured separately, above.
CheckConfirm these settings match the consent policy you intend to enforce, and compare them with the before-and-after observation.
Seenseen in 44.6% of 46,522 containers measured
Third parties it can reach
Each host is a party that can execute code or receive data in a visitor's browser. Follow one to see who else loads it. More info »
What the custom templates can do
Custom templates are third-party code shipped inside the container. Injecting scripts and running page globals is ordinary for them — most tag templates need it to work — so this is an inventory, not a warning. It lists the capability each template declares so you can match it against templates you meant to install.
__cvt_172622598_646 private injects code from tags.creativecdn.com · runs page-level globals__cvt_WK2T3 gallery injects code from gum.criteo.com, dynamic.criteo.com · runs page-level globals__cvt_172622598_122 private runs page-level globals__cvt_172622598_165 private injects code from app.crossengage.io, x.xng.rocks · runs page-level globals__cvt_172622598_45 private runs page-level globals__cvt_172622598_570 private injects code from cdn.taboola.com__cvt_172622598_85 private injects code from unpkg.com__cvt_5RM3Q gallery injects code from connect.facebook.net · runs page-level globals__cvt_PLW64 gallery injects code from cdn.taboola.comDetected on
Sites observed loading this container. A container used across many hosts is a shared template; one on a single host is a bespoke build. Follow a site to see every container it loads.
What this audit describes
Containers change. These findings describe one specific artifact, and you can re-fetch it to check this reading.
| Container | GTM-N345CDJX |
|---|---|
| Published version | 112 — the identity of this configuration. It changes only when someone publishes. |
| Read at | 2026-08-13 11:45:39 UTC (a month ago) |
| Source | https://www.googletagmanager.com/gtm.js?id=GTM-N345CDJX |
| Body checksum | 1a29d4c08d3047793b2def4dcc83edb1844594d51fc099f70459c38c01695a52Evidence of the exact bytes we read, not an identity: Google varies the response per request, so re-fetching an unchanged container yields a different checksum. Compare the published version instead. |
Re-audit
This reading is from a month ago. A container can be republished at any time, and nothing here updates on its own — a fresh read is the only way to know whether it still says this. Running it again replaces this page in place, so the link keeps working.
Put this container under version control. Tagora exports it to a repository you own, so every change arrives as a pull request. How it works »