What shipped,
and what hasn't.
Written from the repository record rather than from memory. The last section is the honest half — what this product does not claim yet, and why.
Shipped
- This siteOverwatch, wired to the same sources the pages render
The assistant answers from a knowledge file generated at build time out of the same pricing, pillar and rubric sources the pages use — so it cannot quote a price the site does not show. A gate fails the build if the two drift apart.
- This siteV3 Fire Control — the marketing site, rebuilt
Sixteen sections, fifteen routes, one design system. Every number on the page is read from the same source the product uses, and anything that could not be traced to a source was removed rather than estimated.
- AppFeed endpoints and webhooks, verified against a real store
The canonical record serves in four formats and reconciles against the database two independent ways. Product-update webhooks re-enrich on change, with a loop guard so our own write does not wake us forever.
- AppThe write layer, and a contract enforced in code
Enriched fields land in one dedicated metafield namespace. Every other mutation is refused at the API client before the request is issued — an allowlist, not a policy. Revert is proven byte-for-byte on real products.
- CommercialPriced by catalogue, never by pillar
All four disciplines are in every tier. A tier decides how large a catalogue is read and how often — it never decides which pillars a merchant gets.
- MeasurementThe measurement audit became a standing gate
Three checks were found measuring something other than what they claimed — one scored every store on earth 0% on a tenth of its SEO pillar. The audit that found them now runs on every rubric change, because we expect to find more.
- FoundationCrawler policy, read by name
robots.txt is parsed against 10 named AI agents rather than pattern-matched, and a blocked agent is scored as a hard exclusion instead of a ranking penalty — because that is what it is.
- FoundationThe scanner and the published rubric
A public scanner that reads any Shopify storefront and scores it against 4 pillars, with every check and weight published at /rubric rather than described. Rubric v1.0, versioned so a score always says which rules produced it.
In build
- In buildBefore/after shot reports
The write pipeline is real and the scoring is real. What does not exist yet is a store scanned, enriched and then re-scanned — so there is no measured "after". The section is built and deliberately renders nothing until there is one.
- In buildMore attachments
New modules ship continuously and switch on from the console — no re-install, no migration. Each is tagged to the pillar it serves.
- In buildStorefront JSON-LD block
The hosted feed works today. The theme extension that publishes the same record on your product pages needs one toggle in your theme editor — we do not request permission to edit themes, which is why it is your switch and not ours.
- In buildFull French and Spanish
Navigation, hero and footer are translated. The remaining section copy is moving into the message catalogue, and machine-assisted entries are marked for review rather than presented as final.
Not claimed yet
- Not claimedA published benchmark
No zero-touch rate or industry median is published, because the only catalogues measured end to end so far are demo data. A number about sample data wearing this product’s name is worse than no number.
- Not claimedAgent traffic figures
Nothing here reports agent requests served or referred revenue. Those are not measured yet, so they are not shown — the hero counters and the command centre suppress themselves rather than estimate.
- Not claimedA refund or trial offer
There is none, deliberately. If one is ever introduced it will be written fresh, not revived from a retired draft.
Every ruling behind these lines is dated and kept, including the ones that reversed an earlier decision. If a claim on this site ever looks unsupported, ask us for the entry.
