feat: measure clicks through to the ticket shop

[thevvalverde]

What

Adds a select_ticket_link GA4 event fired when a visitor follows any link to the Pretix ticket shop, so we can see which pages and which traffic sources actually lead people to tickets.

Why not purchase or begin_checkout

The purchase happens on Pretix and we never learn whether it completed, so a purchase event would be a claim we cannot support. begin_checkout was rejected for the same reason at one remove: it flows into GA4's ecommerce funnel reports and would invent a funnel with a permanently missing bottom. A custom event keeps the reports honest. There is a test pinning this.

To count it as a conversion, mark it as a key event in the GA4 UI — no code change needed.

How

One delegated listener on the document rather than a handler per link. Ticket links appear in the header, the home hero, the four ticket cards, navigation.json and the FAQ Markdown; matching on the shop host means a link added later, including one written in content, is measured with no extra work.

• PRETIX_HOST is a setting so there is a single thing to change, and so a staging shop still reports.
• The hit goes out with transport_type: 'beacon', because the click navigates away and a normal request would be cancelled in flight.
• Traffic source is deliberately not attached: GA4 resolves source/medium/campaign for the session, so the event inherits the attribution of the visit it belongs to. Only source_page and the destination — which GA4 cannot know — are passed.

Privacy

No change to the consent posture. The listener returns early when the tag was never loaded, so a visitor who declines is not measured and no request reaches Google — the promise the privacy guide already makes. The guide is updated to state what is now recorded and its limit.

Counting clicks from visitors who decline was considered and dropped: GA4's Measurement Protocol requires a client_id, and a random one per click manufactures a fake user each time, corrupting user and session counts. A server-side counter was the alternative, but base.py states plainly that the site has no database, and this did not seem worth reversing that for. Consequence worth knowing when reading the numbers: ticket-click counts undercount by roughly the decline rate, so they are sound for comparing channels rather than as an absolute measure of interest.

Testing

• 23 tests pass (6 new).
• The new assertions were mutation-checked — swapping the event name for begin_checkout turns two of them red.
• The consent gate itself was exercised against the real rendered script in a headless DOM: an accepted visitor fires exactly one event, a declined visitor fires nothing and never requests gtag.js, non-ticket links are ignored, a click on an element nested inside the link still resolves, and a lookalike host (evil-pretix.evolutio.pt.attacker.com) is correctly rejected.
• test_the_home_page_actually_links_to_that_host is the test that earns its keep: the tracker matches on the host, so a ticket link edited to point elsewhere would silently stop being measured — this turns that into a red test.

Follow-up in the GA4 UI (not code)

Custom dimensions are not retroactive, so this is worth doing right after deploy or the parameters read (not set) forever. Under Admin → Custom definitions, register link_url, link_text and source_page as event-scoped dimensions.

🤖 Generated with Claude Code

https://claude.ai/code/session_011i8AYdEkHBURzkFmyGpQmy