This brief gathers one working arc for review. It covers the stack manifest becoming a runnable surface, the prototype scripts becoming one cub plugin, the full fleet example, and the homepage change that puts the ladder on display. The reviewer should judge three things: what a user arriving at the website can do, what the OCI catalog is worth to them, and whether the examples answer "what can I do here" well enough to carry blog posts.
What happened, in order
- The keystone platform became a stack manifest and the manifest became a command:
cub stack sandbox eks-inferencecertifies and renders the whole platform for free (#1687). The certify step caught a real duplication the deeper verdict had deduplicated silently, so the two layers check each other. - The manifest format was specified, bindings moved into it, and the composition verdict was armed as a regression gate (#1688). A second producer proved the format producer-neutral: the same verbs certify a platform composed from the catalog's certified renders (#1689).
- The fleet layer was written down as data: composition times placement times history, with a generated slice whose four attention tiles match the product's own queries (#1690).
- The CI-rendered catalog journey landed as a documented user journey: teams whose CI already renders charts hold the result as governed data with one command, canonically equal, reconciler untouched (#1691).
- The prototype scripts graduated to one real cub plugin named workshop, carrying the whole noun family:
cub config,cub app,cub stack,cub fleet. One install gives every noun, every free verb, and the content they run against: nine verified catalog renders, thirteen authored apps, eleven stacks, and a ten-cluster fleet manifest. - The full meridian fleet ran through the plugin: ten regional clusters, twenty components, 125 placements, built entirely through the governed verbs on a disposable self-hosted server. The build found a real capacity boundary, stopped with a named remediation, and resumes exactly where it stopped. The demo-aging step replays operations rather than faking state, so the attention tiles show real residue.
- The homepage changed to match: the hero terminal now shows the ladder instead of one packaging command, the four starting questions became six, and the vocabulary got one line with a home for each noun.
Review focus one: a user coming to the website
The homepage now answers "what can I do here" at three sizes. The reviewer should walk each row and confirm the page delivers what the row claims.
| Band | What the visitor can do | Where |
|---|---|---|
| Entry | See what a chart installs before installing, in the browser, no login | Check my config |
| Entry | Find a tested component and pull its verified render | Catalog, 112 components |
| Entry | Compare versions and preview a promotion | Promote my config |
| Entry | Render a reviewed chart to controller-native OCI and let their existing Flux or Argo CD reconcile it | starting question 5 |
| Entry | Certify and render a whole platform for free | the ladder terminal, cub stack sandbox |
| Mid | Land CI-rendered YAML as governed data, canonically equal | the CI-rendered catalog journey |
| Mid | Release by digest and watch either reconciler apply exactly that digest | the delivery receipts |
| Mid | Promote through an approval gate that refuses first | the operator ladder |
| Keystone | Build a platform and run apps through it | Kubara, recorded live |
| Keystone | Speak a platform as one certified stack, any producer | the stack manifest spec and receipts |
| Keystone | Generate and read a governed fleet from manifests | starting question 6 |
Known honest edges the reviewer should keep honest: the plugin verbs are labeled proposed and prototype on the page, the plugin installs from the public repository with cub plugin install confighub/cub-workshop, and the real-GPU rung remains the named unproven step.
Review focus two: the value of the OCI catalog
The catalog is the substrate every rung stands on, and the review should check the site says so where it matters.
- Every free verb works anonymously because the catalog publishes verified renders and digest-pinned bundles to public OCI. The plugin pulls bundles by digest and hash-verifies them against receipts before a single object parses.
- Delivery claims are exact because releases and bundles share the digest discipline: the receipts show Argo CD and Flux each applying exactly the published digest, including through a governed change.
- The certified-bundle receipts are what let a platform be rebuilt from evidence alone, which the replica track proved end to end.
- The commercial line is drawn plainly: public catalog and public journeys are free, while private catalogs, team access, and production responsibility are the paid tier.
Review focus three: blogs and examples
Every capability row above is also a post. The journeys already documented on the site each carry their receipts, so a writer cites rather than asserts: the CI-rendered catalog journey, the operator ladder with the gate that refused first, the Flux repeat, the stack sandbox for the keystone and for the catalog-composed platform, the fleet slice, the no-GPU proof, and the Kubara build. The reviewer should confirm each journey page stands alone: states what it proves, links its receipt, and names its limits.
What to check by hand
- Open the homepage and read the hero terminal as a newcomer. Does the ladder read as one story, and is the prototype labeling clear without deflating it?
- Click all six starting questions. Each should land on a page that delivers its promise within one screen.
- Follow the vocabulary line. Config, app, stack, and fleet should each resolve to a page that defines the noun and shows it running.
- Read the CI-rendered catalog journey as the strongest enterprise entry and check the honest twist is stated: text stays frozen, data moves with its reason.
- Run the install line from the page on a clean machine and confirm the four noun commands answer:
cub plugin install confighub/cub-workshop, thencub config check redis.