Produced with model assistance. Every page on this site, and every document in the dev packs area, was drafted with a large language model and reviewed by a person before publication. Nothing on this site is a compliance assessment, and no page claims conformity to any standard. What we do not say, and why · how every claim here is evidenced

store.sgit.ai / versions / v0.3.14

v0.3.14 — the buying flow, end to end, and one order across both designs

The v4 handback supplied the policy picker, the cart, the checkout and the post-purchase screens. All four are built at /next/ from the store's own data: /next/policies/, /next/cart/, /next/pay/ and /next/paid/, mirroring the routes the store that sells today already uses.

It is the live store's order, not a second one. The current store keeps an order in the browser under one key; /next/ reads and writes that record — same key, same schema, same line shape, same reference alphabet, same SKUs — so an order started in this design round is still there on the current store, and back. Two carts under two keys would be a bug that only appears for the one person who uses both. A check compares the two models field by field on every build.

Still no typing surface anywhere. Quantities are buttons, the behaviour filter is a <details>, and the picker's search holds no field at all: the query lives on the body element and characters are read off the keyboard. The cost is that a phone with no hardware keyboard cannot type there, and the dialog says so and points at the chips.

Two findings about the published catalogue, both on the board. The picker offers a filter by behaviour that narrows nothing: the 23-behaviour vocabulary is published and every shape's totals are published, but which behaviours make up a particular grant is not. And for five of the fifteen shapes, wanted plus not-asked overshoots the grant by exactly one — ten are exact. The panel's capability strip was being painted green and amber as though those numbers partitioned the grant; it is now one uncoloured cell per capability, and the five shapes say so.

The evidence state behind each shape is typed out in data/products.yml rather than derived, because deriving it got it wrong: chatgpt-web's summary calls it "the baseline every other shape is measured against", and a regex cannot read a preposition.

New rule: check_every_anchor_a_page_points_at_is_there. check_links has always held the file half of a link and never looked past the #. The site was clean when the rule was added, which is the only comfortable moment to add one.

Versionv0.3.14
Released2026-09-16
Built from commitde2d735b5a9b01f5dc13e7c436aaa7d519db991b
Sitestore.sgit.ai

Touched: assets · build.py · data · tools

← v0.3.15v0.3.13 →

Every release of this site is listed on the release history, and the same records are served as JSON so a script can read them without parsing a page.