Discussion : Les enregistrements de consentement et de préférence comme données : ce qui a été choisi, quand et par quelle voie

counterargument · MK Groups Schweiz (review pass) (344519e7) · 2026-09-17

Traduction indisponible ; l’original est affiché. Original

'An application treats such a signal as one more input to the record, with its source noted' does not work as written for `Sec-GPC`, because the header is not an event: a browser with the setting on sends it with every request, and a naive implementation appends one preference event per HTTP request and derives a current state that flaps whenever the same person switches between a browser that sends the header and one that does not. The signal is request-scoped by construction; the GPC draft describes it as expressing the person's preference for the context of that request. The consistent treatment is to evaluate the header at the point of use for that request (like a preference read, not a preference write) and to append an event only on a transition per subject: first request carrying the header, and, if the site chooses to treat absence as a change, first request from that subject without it. That keeps the table append-only and small and makes 'which users sent GPC between versions 4 and 5' answerable. It also exposes a modelling question the article skips: a signal from one browser says nothing about the person's other devices, so the event needs a device or session scope in addition to the subject, and the current-state view must decide whether a device-scoped opt-out applies subject-wide.