A Claims Library, And A Privacy Signal The Site Finally Honours

The email account was the kind of broken that does not announce itself, because nothing had been sent. Every draft in it rendered raw anchor markup straight into the customer facing body, alongside an internal build note that was never meant to leave the workspace. Twenty four of the forty five templates could not even be previewed: a template marker was arriving at the platform as syntax it tried to execute, and it gave up rather than render. All 270 images in the account still carried a wordmark the brand had retired.

The finding that mattered most was quieter. The abandoned cart flow had three buttons pointing at a URL that returned 404. Because the flow was still a draft, no customer had ever hit it. It would have fired the instant anyone switched the flow on, which is exactly when nobody is watching the links.

So the first half of this work is unglamorous repair. We fixed the rendering defect at its source rather than template by template, added a way to place images that the platform would not mangle, and re-pushed all forty five templates with a read back control, which is to say we asked the platform what it now held rather than trusting our own write. Previewable templates went from twenty one to thirty nine. The remaining six are blocked on something outside the templates: the account has no coupon pools, so a coupon code marker has nothing to resolve against. We wrote that down rather than working around it. We also recorded the thing that did not work, so nobody tries it again: encoding the braces does not help, because the platform decodes them before it parses.

The second half is governance. Marketing copy across the site, the product pages, the flows and the marketplace listings had grown up separately, and a claim that is fine on one product is not automatically fine on another. So we built a claims library that says what may be said about the product, mapped it across eleven surfaces, and then applied it. That is not a document handed over at the end. It is the thing copy gets checked against.

The privacy work is the piece we would point at first. The site's own policy promised to honour the browser's Global Privacy Control signal, and the site did not honour it. We shipped it properly: the signal now denies all four consent categories before the tag manager loads, it outranks whatever the visitor once clicked in the banner, it suppresses the banner, and an ungated opt out link sits on every page and routes to a control that works.

Then we measured it with a control, because a privacy feature that is not measured is a claim. With the signal on: no cookies set. With consent granted instead: three. Two different readings from the same page proves the check can tell the difference, which is the only reason the first number means anything.

The policy rewrite that goes with all this was deliberately left as a draft with its open questions marked, and recommended for a lawyer to read before anything is published. We also did not bump its date, because the live text tells readers the date signals a content change.