store.sgit.ai / versions / v0.3.3

v0.3.3 — you see a vault on the page you land on, and the rule was narrowed a second time to allow it

When you have sold a vault, you should just see the vault. The difficulty is that at the moment somebody pays there is nothing of THEIRS to show: at the vault level it is built for them within a day or two, at the two upper levels after they send their details. A page that showed your vault at that moment would be lying at the one moment a buyer is paying most attention to it.

So what is embedded is a published vault of exactly this kind of work — an agent's grant, its mandate and the eight-capability delta between them — said to be an example in its first sentence. For a buyer who has never held one that is the more useful of the two things anyway: a vault is hard to picture from a description and takes about a minute to read. The pack level has nothing to show and does not get the component at all, which is the difference between a narrow exception and a wide one.

THE RULE WAS NARROWED, NOT LOOSENED, FOR THE SECOND TIME. What stayed absolute: every page that SELLS anything opens no connection, and there is now a check that refuses an embed on one by name rather than only by omission from an allowlist. Five deliberate breaks were run.

AND THE NARROWING MADE A PUBLISHED SENTENCE FALSE, WHICH IS WHAT CAUGHT IT. Both embedding pages had to say the one place on this site that opens a connection, and a second kind of page made that untrue. The check requiring it failed on the release that made it false, before either page shipped saying it. The sentence is now true of both.

Chasing that down found the same imprecision in seven more places, and it had been wrong since the FIRST narrowing rather than this one: the review pages told a reader typing into a reason box “no page here opens a network connection”, on the very pages that embed a vault. Also the lab prototypes, the catalogue page, two claims in the ledger, and two data files. Each now says what is true, and check_the_network_claim_is_qualified fails the release if the comfortable version comes back. A quoted historical claim is exempt, because a site that could not quote its own corrections could not explain them.

ONE REAL HOLE FOUND BY A BREAK THAT DID NOT FIRE. The read-key check matched the prefixed form and the colon-joined form, and the embed carries its key and its vault id in two separate attributes — so a key sitting in live markup matched neither, which is the exact shape a careless copy would produce because it is the shape this site's own pages carry. It is caught now.

And OS-4 is blocked for a named reason rather than a vague one: their manifest exists, with bytes and a sha256 for all fifteen shapes, and it lives inside a JavaScript const in their HTML rather than at a JSON address. CORS is confirmed; the shape is the gap. The brief on /boundary/ now says exactly that.

Touched: build.py · content · data · tools

← v0.3.4v0.3.2 →

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.

Which agent do you run? → How buying works → Every claim, with its state →