The ID.me correction, and Home rebuilt for browsing

Verified against ten live store pages

Casper 5%, Lowe's 10% and Samsung 30% match what Claude Cowork and ChatGPT found independently, which is the strongest confirmation available without a person standing at a register.

The fixtures could never have caught it

They were hand-written markup with no banner in them, so they tested a page ID.me has never served. They are now trimmed from real store pages, keep ID.me's own class names, and every one of them carries the 40% sentence — so a regression that reads it fails in the suite rather than in a member's hands.

Both new publish tests were confirmed to fail with the fix removed. A test that has never been seen to fail is a test that has not been checked.

A store checked and carrying nothing is now a published fact

When a store had no military offer the pipeline emitted nothing at all, which threw away the only record that the page had been read. A brand swept that morning looked unexamined.

It also broke the change feed. When a published route stopped being emitted, "Nike dropped from ID.me Shop" pointed at a route no longer in the dataset, and the app drew a row with no brand name, no logo and nothing to tap. 63 negatives now carry their sweep date, following the convention the GovX sweep already set.

"Worth a look" — the shelf

Home used to run out after the exchange cards. With nothing in the change feed and no watched categories, the lower half of the screen was empty and the only way in was to already know what you wanted.

The new shelf draws from the 68 brands with a verified discount, takes one pick per category so nothing sits next to the shelf it came from, and rotates daily. The arrow re-rolls it, because poking at it is the point.

Picks are weighted toward the bigger discount rather than sorted by it. Sorting shows the same six brands until the data moves. A flat shuffle put three −20% cards in the first four, which is the sameness the shelf exists to break. Weighted sampling leads with something worth seeing and is still different tomorrow.

The category row: eight, not twenty-four

All of them in one scrolling row read as a wall to get through, and several tiles opened on nothing — Insurance carries four brands and Memberships three, and not one of the seven has an offer. A tile is now sized by the brands behind it a member can actually use, and the rest are one tap away under "All 24".

An ID.me offer now opens the brand's store page

Tapping a −20% Columbia card opened shop.id.me and left the member to go hunting for the discount the card had just shown them. ID.me is 36 of the 42 routes that state a real amount, so that was most of what the new shelf puts in front of someone.

The store path had been in the registry all along — it is how the sweep fetches the page it reads the offer from — but it was marked pipeline-only, so the app could not see it and could not rebuild it. Published now, on 99 of 162 brands.

Two changes worth your eye, because I made the call myself

The discount pill is gone from plain offer cards. The card read "−20% for military" next to a "−20%" pill — the same fact twice, spending the one piece of colour the row has. Deadlines and "New" marks still earn theirs. This changes offer cards everywhere, not only on Home, so it is the change most likely to be one you want reversed.

I tried the lead find as a hero card and took it back out. The hero layout sets a 17pt caps headline against a full-bleed illustration because it was drawn for the narrow carousel card. At the width of the screen it turns the phone into a billboard and the type into a caption. If you want hierarchy on that shelf it needs a fourth density, not a stretched hero.

Nine of twenty-four categories have no priced offer at all

The highest-leverage gathering, in order

• Amounts for the 26 GovX-listed brands. We know they are carried and not what they give. Filling these in raises priced offers from 42 to 68 — a 62% increase — with no new brands to find. It needs either the GovX SPA or an extraction pass; I did not start one, since it spends API budget you did not authorise.
• The nine empty categories. Auto, Baby & Kids, Pets, Gifts, Memberships, Insurance, Moving, Software, Storage. These are the tiles a member taps and finds nothing behind. Insurance and Memberships in particular look like categories that were created for brands whose offers we then could not verify.
• Brand accents. Nearly every brand falls back to the house copper, so every eyebrow on the shelf is the same orange. Accents would do more for how the shelf reads than any further layout work.

Q1  Should the pill come back?

I removed it as a duplicate. It was also the only colour on a standard card, and the shelf is flatter without it. The alternative is to keep the pill and change the headline to lead with the brand name instead of the number.

Q2  Do the GovX-listed brands belong on the shelf?

Right now a find must state a number, so the 26 "Via GOVX" brands never appear as finds — though they do count toward a category tile's depth. That inconsistency is deliberate and worth a second opinion.

Q3  What should Home say when there is genuinely no news?

The change feed is empty today and the section simply does not draw. That is honest and it is also silent. With the shelf there, the page no longer looks broken — so this may need nothing at all.

Q4  Footwear and Running read as the same drawing

They are not — illus-sneaker and illus-running-shoe are separate objects. At 21pt inside a 56pt circle they are indistinguishable. Either one of them wants a more distinct silhouette, or Running does not need its own tile.