Browse Docs
Catalog
Config
Stacks
Operate
Docs

Documentation Map

A repository document, rendered for the site. View source markdown.

New to cub? Install the cub CLI first. Public catalog packages pull and render anonymously, and you sign in only once a command saves or changes ConfigHub data.

Generated at: 2026-09-04T11:23:58.207Z UTC · source: committed helm-expt evidence for this rendered repository document.

EXPERIMENTAL

The public website is the starting point for people who want to use the project. This directory holds the detailed guides and the evidence behind that site.

Human Route

Use these pages in order for the shortest introduction.

PagePurpose
Try RedisInspect one checked package without a ConfigHub account, server, or Kubernetes cluster.
ExamplesStart from Helm, AICR, OCI, or YAML, then explore promotions, fleets, policy, and Apps.
CatalogChoose a tested starting configuration for a public Helm chart.
DeploymentChoose files, OCI, ConfigHub, Argo CD, Flux, or direct Kubernetes delivery.
DocsFind instructions for the current step.
ConfigHubSave, change, approve, promote, and roll out reviewed configuration.

The rest of this file is a reference index. You do not need to read it from top to bottom. Most Markdown files in the repository are generated catalog, recipe, package, test, or run records.

Each section groups documents by the question a reader arrives with rather than by where the file happens to live, so the same document can appear under two questions when it genuinely answers both. It never appears twice under one.

User Guide Index

The manual user documentation, grouped by what a reader is trying to do. user/README.md keeps the serial reading order for anyone who would rather read it straight through.

Start here

The shortest path from arriving to running something.

DocumentPurpose
../AGENTS.mdWorking agreement for coding agents: who owns which surfaces, the gates every change runs, and the traps already paid for.
../CATALOG.mdGenerated chart catalog: charts first, recommended variants underneath.
../README.mdMain user introduction: why the repo exists, what is proven, and how to try it.
../WEBSITE_UX_TEST.mdLocal static-site UX test runbook for walking a fresh user through the public website.
choose-your-path.mdRoute picker for Helm, AICR, OCI, or YAML; local, anonymous, or managed use; and the official ConfigHub tutorial handoff.
expected-results-and-clusters.mdPractical guide for what users should see after each step, when they need a Kubernetes cluster, when to use kind or cub cluster up, and when npm verifiers are optional.
first-run-walkthrough.mdCaptured real serverless try-out (render → kubectl apply → running pod) with the honest namespace rough edge; tested-UX companion to try-now.
offering.mdPublic offering overview: why visible Helm stages matter, what is free to try, and where managed ConfigHub workflows begin.
try-now.mdWalkthrough capture and reference links behind the short public package exercise; the Helm-parity comparison now lives on Try Redis.
tutorial-sequence.mdDetailed repository proof sequence for Redis, variants, overlays, GitOps, and bulk operations; not the ConfigHub product tutorial.
user/README.mdSerial reading order for the manual user docs.
what-you-get.mdCompact product model: what ConfigHub adds around Helm output, what is proven, and what remains product work.
why-this-exists.mdSkeptical entry point: why this is more than a one-shot upload or GitOps import, and what the catalog/proof path adds.
Answer the ten questions Helm users askThe known path for a Helm user: ten questions, the pain point under each one, the local command that answers it, and the check you run on your agent.
Answer the questions Timoni users askThe known path for a Timoni user: build the exact objects, check what they need, and see which lifecycle work has to travel with them, using the retained Redis and Flux All-In-One entries.
Compose a platform with an appComplete local cub and assistant Guide: save, move, edit and refuse an incompatible API in the pinned Kubara/Argo/app selection.
Test whether a Helm value changes the outputControlled local renders with a misspelled key, corrected key, transformation and retained comparisons.
Review a chart upgradePinned candidate comparison, unexpected replacement and a review packet that names missing promotion evidence.
Find hook and CRD workRender a hook without executing it, retain a served-API refusal, and complete the lifecycle checklist.
Add a field and restore the sourceLocal cub and assistant Guide: preserve source bytes, inspect a label addition, detect an object replacement and restore an exact copy.
Adapt and review a configurationComplete local cub and assistant Guide: retain an exact replica diff and detect an unexpected second change.
Match a GPU workload to supplied factsComplete local cub and assistant Guide: retain candidate, mismatch and unknown results against illustrative Node inputs.

Worked examples

Complete journeys someone has already run end to end.

DocumentPurpose
AI change review exampleA real AICR training object, an unsafe proposed change, the corrected candidate, and an honest receipt showing which checks and live steps did or did not run.
AICR CPU starter examplePull the public AICR configuration anonymously, reproduce and hash-check the seven-Application CPU starter, write a local OCI, then follow its ConfigHub and bounded kind proofs. No GPU, cloud account, or NGC key is needed for the first run.
AICR GPU-node snapshot comparisonCompare two observed GPU-node snapshots, bind each node to its intended networking profile, keep the exact review as files or OCI, and optionally retain it in ConfigHub without treating the review as deployable configuration.
AICR platform variant parityOne AI-proposed StorageClass change accepted only after object identities, affected Applications, and the exact changed field pass; one over-broad request refused with a receipt and no candidate.
AICR H100 training exampleThe retained EKS + H100 + Kubeflow entry: committed bundles and receipts, the digest-bound platform index, OCI round trip, ConfigHub upload, and the development-to-staging promotion.
AICR v0.20.0 H100 training entryPull and inspect the newest retained EKS + H100 + Kubeflow source package and 17 exact Argo CD Applications, then see what changed from v0.19.0 and which ConfigHub, delivery, and runtime steps remain.
AICR KServe NIM inference exampleThe retained NIM-on-KServe entry: the Apache-2.0 nim-deploy KServe subtree at an exact commit, the digest-bound index with licensing-boundary refusal rules, and the first model profile described as data.
AICR catalog overviewThe three AICR entry classes (training, inference, CPU starter), what exists against what is planned, the retained-versions discipline while upstream moves, and the config-plane-only boundary.
AICR-native NIM inference exampleThe platform-level inference entry generated by AICR itself with platform: nim, held next to the KServe entry, with the training-to-inference component diff and a direct comparison of the two inference sources.
AICR h100-gke-cos-training-slurm exampleMirrored from NVIDIA AICR's recipe catalog: 18 components resolved into 21 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR h100-eks-ubuntu-training-slurm exampleMirrored from NVIDIA AICR's recipe catalog: 17 components resolved into 19 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR h100-aks-ubuntu-training-slurm exampleMirrored from NVIDIA AICR's recipe catalog: 16 components resolved into 20 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR gb300-eks-ubuntu-training-slurm exampleMirrored from NVIDIA AICR's recipe catalog: 17 components resolved into 20 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR gb200-eks-ubuntu-training-slurm exampleMirrored from NVIDIA AICR's recipe catalog: 17 components resolved into 20 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR vr200-rke2-ubuntu-training exampleMirrored from NVIDIA AICR's recipe catalog: 14 components resolved into 15 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR vr200-rke2-ubuntu-inference-dynamo exampleMirrored from NVIDIA AICR's recipe catalog: 18 components resolved into 21 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR vr200-rke2-ubuntu-inference exampleMirrored from NVIDIA AICR's recipe catalog: 16 components resolved into 19 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR rtx-pro-6000-lke-ubuntu-training exampleMirrored from NVIDIA AICR's recipe catalog: 11 components resolved into 12 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR rtx-pro-6000-lke-ubuntu-inference exampleMirrored from NVIDIA AICR's recipe catalog: 13 components resolved into 16 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR rtx-pro-6000-lke-training exampleMirrored from NVIDIA AICR's recipe catalog: 11 components resolved into 12 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR rtx-pro-6000-lke-inference exampleMirrored from NVIDIA AICR's recipe catalog: 13 components resolved into 16 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR rtx-pro-6000-eks-ubuntu-training-kubeflow exampleMirrored from NVIDIA AICR's recipe catalog: 15 components resolved into 17 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR rtx-pro-6000-eks-ubuntu-training exampleMirrored from NVIDIA AICR's recipe catalog: 14 components resolved into 15 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR rtx-pro-6000-eks-ubuntu-inference-nim exampleMirrored from NVIDIA AICR's recipe catalog: 17 components resolved into 20 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR rtx-pro-6000-eks-ubuntu-inference-dynamo exampleMirrored from NVIDIA AICR's recipe catalog: 18 components resolved into 21 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR rtx-pro-6000-eks-ubuntu-inference exampleMirrored from NVIDIA AICR's recipe catalog: 16 components resolved into 19 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR rtx-pro-6000-eks-training exampleMirrored from NVIDIA AICR's recipe catalog: 14 components resolved into 15 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR rtx-pro-6000-eks-inference exampleMirrored from NVIDIA AICR's recipe catalog: 16 components resolved into 19 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR rtx-pro-6000-any exampleMirrored from NVIDIA AICR's recipe catalog: 11 components resolved into 12 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR rke2-training exampleMirrored from NVIDIA AICR's recipe catalog: 11 components resolved into 12 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR rke2-inference exampleMirrored from NVIDIA AICR's recipe catalog: 13 components resolved into 16 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR rke2 exampleMirrored from NVIDIA AICR's recipe catalog: 11 components resolved into 12 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR oke-ol-training exampleMirrored from NVIDIA AICR's recipe catalog: 11 components resolved into 12 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR oke-ol-inference exampleMirrored from NVIDIA AICR's recipe catalog: 13 components resolved into 16 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR oke-ol exampleMirrored from NVIDIA AICR's recipe catalog: 11 components resolved into 12 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR ocp-training exampleMirrored from NVIDIA AICR's recipe catalog: 10 components resolved into 14 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR ocp-inference-nim exampleMirrored from NVIDIA AICR's recipe catalog: 11 components resolved into 16 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR ocp-inference exampleMirrored from NVIDIA AICR's recipe catalog: 10 components resolved into 14 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR ocp exampleMirrored from NVIDIA AICR's recipe catalog: 10 components resolved into 14 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR lke-training exampleMirrored from NVIDIA AICR's recipe catalog: 11 components resolved into 12 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR lke-inference exampleMirrored from NVIDIA AICR's recipe catalog: 13 components resolved into 16 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR lke exampleMirrored from NVIDIA AICR's recipe catalog: 11 components resolved into 12 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR l40s-oke-training exampleMirrored from NVIDIA AICR's recipe catalog: 12 components resolved into 14 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR l40s-oke-inference exampleMirrored from NVIDIA AICR's recipe catalog: 13 components resolved into 16 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR l40s-any exampleMirrored from NVIDIA AICR's recipe catalog: 11 components resolved into 12 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR l40-any exampleMirrored from NVIDIA AICR's recipe catalog: 11 components resolved into 12 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR kind-inference exampleMirrored from NVIDIA AICR's recipe catalog: 14 components resolved into 17 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR kind exampleMirrored from NVIDIA AICR's recipe catalog: 12 components resolved into 13 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR h200-eks-training exampleMirrored from NVIDIA AICR's recipe catalog: 14 components resolved into 15 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR h200-eks-inference exampleMirrored from NVIDIA AICR's recipe catalog: 16 components resolved into 19 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR h200-any exampleMirrored from NVIDIA AICR's recipe catalog: 11 components resolved into 12 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR h100-kind-training-slurm exampleMirrored from NVIDIA AICR's recipe catalog: 16 components resolved into 18 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR h100-kind-training-kubeflow exampleMirrored from NVIDIA AICR's recipe catalog: 13 components resolved into 15 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR h100-kind-training exampleMirrored from NVIDIA AICR's recipe catalog: 12 components resolved into 13 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR h100-kind-inference-dynamo exampleMirrored from NVIDIA AICR's recipe catalog: 16 components resolved into 19 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR h100-kind-inference exampleMirrored from NVIDIA AICR's recipe catalog: 14 components resolved into 17 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR h100-gke-cos-training-kubeflow exampleMirrored from NVIDIA AICR's recipe catalog: 15 components resolved into 18 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR h100-gke-cos-training exampleMirrored from NVIDIA AICR's recipe catalog: 14 components resolved into 16 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR h100-gke-cos-inference-dynamo exampleMirrored from NVIDIA AICR's recipe catalog: 17 components resolved into 21 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR h100-gke-cos-inference exampleMirrored from NVIDIA AICR's recipe catalog: 16 components resolved into 20 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR h100-eks-ubuntu-training-kubeflow exampleMirrored from NVIDIA AICR's recipe catalog: 15 components resolved into 17 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR h100-eks-ubuntu-training exampleMirrored from NVIDIA AICR's recipe catalog: 14 components resolved into 15 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR h100-eks-ubuntu-inference-nim exampleMirrored from NVIDIA AICR's recipe catalog: 17 components resolved into 20 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR h100-eks-ubuntu-inference-dynamo exampleMirrored from NVIDIA AICR's recipe catalog: 18 components resolved into 21 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR h100-eks-ubuntu-inference exampleMirrored from NVIDIA AICR's recipe catalog: 16 components resolved into 19 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR h100-eks-training exampleMirrored from NVIDIA AICR's recipe catalog: 14 components resolved into 15 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR h100-bcm-ubuntu-training exampleMirrored from NVIDIA AICR's recipe catalog: 12 components resolved into 13 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR h100-bcm-training exampleMirrored from NVIDIA AICR's recipe catalog: 12 components resolved into 13 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR h100-any exampleMirrored from NVIDIA AICR's recipe catalog: 11 components resolved into 12 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR h100-aks-ubuntu-training-kubeflow exampleMirrored from NVIDIA AICR's recipe catalog: 14 components resolved into 18 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR h100-aks-ubuntu-training exampleMirrored from NVIDIA AICR's recipe catalog: 13 components resolved into 16 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR h100-aks-ubuntu-inference-dynamo exampleMirrored from NVIDIA AICR's recipe catalog: 17 components resolved into 22 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR h100-aks-ubuntu-inference exampleMirrored from NVIDIA AICR's recipe catalog: 15 components resolved into 20 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR h100-aks-training exampleMirrored from NVIDIA AICR's recipe catalog: 13 components resolved into 16 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR h100-aks-inference exampleMirrored from NVIDIA AICR's recipe catalog: 15 components resolved into 20 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR gke-cos-training exampleMirrored from NVIDIA AICR's recipe catalog: 12 components resolved into 14 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR gke-cos-inference exampleMirrored from NVIDIA AICR's recipe catalog: 14 components resolved into 18 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR gke-cos exampleMirrored from NVIDIA AICR's recipe catalog: 12 components resolved into 14 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR gb300-generic-ubuntu-training exampleMirrored from NVIDIA AICR's recipe catalog: 13 components resolved into 15 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR gb300-eks-ubuntu-training-kubeflow exampleMirrored from NVIDIA AICR's recipe catalog: 15 components resolved into 18 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR gb300-eks-ubuntu-training exampleMirrored from NVIDIA AICR's recipe catalog: 14 components resolved into 16 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR gb300-eks-ubuntu-inference-dynamo exampleMirrored from NVIDIA AICR's recipe catalog: 18 components resolved into 22 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR gb300-eks-ubuntu-inference exampleMirrored from NVIDIA AICR's recipe catalog: 16 components resolved into 20 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR gb300-eks-training exampleMirrored from NVIDIA AICR's recipe catalog: 14 components resolved into 16 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR gb300-eks-inference exampleMirrored from NVIDIA AICR's recipe catalog: 16 components resolved into 20 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR gb300-any exampleMirrored from NVIDIA AICR's recipe catalog: 11 components resolved into 12 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR gb200-oke-ubuntu-training-kubeflow exampleMirrored from NVIDIA AICR's recipe catalog: 13 components resolved into 17 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR gb200-oke-ubuntu-training exampleMirrored from NVIDIA AICR's recipe catalog: 12 components resolved into 15 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR gb200-oke-ubuntu-inference-dynamo exampleMirrored from NVIDIA AICR's recipe catalog: 15 components resolved into 18 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR gb200-oke-ubuntu-inference exampleMirrored from NVIDIA AICR's recipe catalog: 13 components resolved into 16 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR gb200-oke-training exampleMirrored from NVIDIA AICR's recipe catalog: 12 components resolved into 15 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR gb200-oke-inference exampleMirrored from NVIDIA AICR's recipe catalog: 13 components resolved into 16 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR gb200-eks-ubuntu-training-kubeflow exampleMirrored from NVIDIA AICR's recipe catalog: 15 components resolved into 18 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR gb200-eks-ubuntu-training exampleMirrored from NVIDIA AICR's recipe catalog: 14 components resolved into 16 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR gb200-eks-ubuntu-inference-dynamo exampleMirrored from NVIDIA AICR's recipe catalog: 18 components resolved into 22 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR gb200-eks-ubuntu-inference exampleMirrored from NVIDIA AICR's recipe catalog: 16 components resolved into 20 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR gb200-eks-training exampleMirrored from NVIDIA AICR's recipe catalog: 14 components resolved into 16 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR gb200-eks-inference exampleMirrored from NVIDIA AICR's recipe catalog: 16 components resolved into 20 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR gb200-any exampleMirrored from NVIDIA AICR's recipe catalog: 11 components resolved into 12 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR eks-training exampleMirrored from NVIDIA AICR's recipe catalog: 13 components resolved into 14 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR eks exampleMirrored from NVIDIA AICR's recipe catalog: 13 components resolved into 14 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR bcm-training exampleMirrored from NVIDIA AICR's recipe catalog: 12 components resolved into 13 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR bcm-inference exampleMirrored from NVIDIA AICR's recipe catalog: 14 components resolved into 17 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR bcm exampleMirrored from NVIDIA AICR's recipe catalog: 12 components resolved into 13 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR b200-gke-cos-training-kubeflow exampleMirrored from NVIDIA AICR's recipe catalog: 14 components resolved into 17 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR b200-gke-cos-training exampleMirrored from NVIDIA AICR's recipe catalog: 13 components resolved into 15 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR b200-gke-cos-inference-dynamo exampleMirrored from NVIDIA AICR's recipe catalog: 17 components resolved into 21 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR b200-gke-cos-inference exampleMirrored from NVIDIA AICR's recipe catalog: 15 components resolved into 19 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR b200-any exampleMirrored from NVIDIA AICR's recipe catalog: 11 components resolved into 12 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR aks-training exampleMirrored from NVIDIA AICR's recipe catalog: 12 components resolved into 15 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR aks-inference exampleMirrored from NVIDIA AICR's recipe catalog: 14 components resolved into 19 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR aks exampleMirrored from NVIDIA AICR's recipe catalog: 12 components resolved into 15 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR a100-oke-ubuntu-training-kubeflow exampleMirrored from NVIDIA AICR's recipe catalog: 12 components resolved into 14 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR a100-oke-ubuntu-training exampleMirrored from NVIDIA AICR's recipe catalog: 11 components resolved into 12 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR a100-oke-training exampleMirrored from NVIDIA AICR's recipe catalog: 11 components resolved into 12 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR a100-gke-cos-training-kubeflow exampleMirrored from NVIDIA AICR's recipe catalog: 14 components resolved into 17 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR a100-gke-cos-training exampleMirrored from NVIDIA AICR's recipe catalog: 13 components resolved into 15 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR a100-eks-ubuntu-training-kubeflow exampleMirrored from NVIDIA AICR's recipe catalog: 15 components resolved into 17 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR a100-eks-ubuntu-training exampleMirrored from NVIDIA AICR's recipe catalog: 14 components resolved into 15 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR a100-eks-training exampleMirrored from NVIDIA AICR's recipe catalog: 14 components resolved into 15 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR a100-any exampleMirrored from NVIDIA AICR's recipe catalog: 11 components resolved into 12 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR a100-aks-ubuntu-training-kubeflow exampleMirrored from NVIDIA AICR's recipe catalog: 14 components resolved into 18 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR a100-aks-ubuntu-training exampleMirrored from NVIDIA AICR's recipe catalog: 13 components resolved into 16 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR a100-aks-training exampleMirrored from NVIDIA AICR's recipe catalog: 13 components resolved into 16 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR h100-eks-inference exampleMirrored from NVIDIA AICR's recipe catalog: 16 components resolved into 19 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
AICR eks-inference exampleMirrored from NVIDIA AICR's recipe catalog: 15 components resolved into 18 rendered Argo CD Applications, retained and digest-pinned, generated by scripts/generate-aicr-from-overlay.mjs.
Current Kubara + ConfigHub technical mini-IDPComplete v0.13.0 maintainer and release path: catalogs, config, overlays, hub/spoke topology, component-first OCI, governed definitions and instances, matrix, wiring, platform delivery, and two applications.
c3agent fleet configurationA bounded AI-agent infrastructure example: compact settings become exact objects, companion OCI records, ConfigHub variants, a promoted release, and Argo delivery while the private runtime remains disabled.
Configuration decision chainWorked NGINX case from AI-written values through six accepted fixes, one narrow exception, a retained ConfigHub decision Unit, development-to-staging promotion, and two Argo CD test deployments.
Website and command-line contractGenerated Helm and literal-YAML examples for the same three user jobs, one source-neutral WorkshopResult, exact object identity, ConfigHub continuation commands, and one bounded live retention-and-promotion proof.
Pull-request reportsSource-neutral Markdown and JSON reports generated from WorkshopResult, including exact object identity, findings, lifecycle requirements, omitted checks, and bounded CI exit behavior.
Historical Kubara v0.12.0 proofHistorical. Read-only compatibility evidence for the original one-cluster generation, 77-object bootstrap render, lifecycle route, OCI delivery, and dated live receipt.
Hooks and CRDs examplePlain-English Kube Prometheus Stack install order, chart-specific route records, Argo CD and Flux choices, live evidence, and the remaining manual boundary.
Kubara six-step adoption tutorialOne linear path from Kubara component selection and generation through Git, OCI, the user-selected ConfigHub organization, and application delivery with Argo CD.
Kubara with ConfigHubBuyer overview: what remains Kubara, what becomes better, the measured evidence, honest boundaries, and the path to a future dedicated repository.
RBAC review exampleOne exact permission correction: find unnecessary Secret access, require approval, publish the reviewed objects as OCI, and let Argo CD check the result on an isolated cluster.
Sveltos Kyverno fleet exampleTwo-wave example: ConfigHub approves a pilot and one selector expansion at different OCI digests; Argo CD and Sveltos deliver Kyverno to one cluster, then two, and Sveltos restores drift on each target.
The Sveltos fleet chaptersOne page tying the fleet chapters together, frozen at five here; the maintained story lives in confighub/sveltos-confighub, where all chapters are recorded live and a sixth, the held cluster, has joined them.
aicr-catalog-brief.mdBrief for the AICR catalog: three entry classes, the OCI digest spine ported from the Kubara importer, retained versions, and the config-plane-only proof boundary.
aicr-composition-model.mdMaintained reference on how AICR composes a platform (overlays, mixins, the component registry, deployment order) and where its command surface overlaps this catalog's own concepts, plus what the study settled about our two sourcing choices.
how-the-catalog-is-built.mdWhy the records are shaped this way: retained versions and what retention buys, the matrix taxonomy, and what a passing verifier does and does not mean.
aicr-evidence-and-our-receipts.mdHow AICR's own evidence machinery compares with this project's receipts, which tool is authoritative for which question, and the three public claims that were corrected once upstream's evidence surface was read.
aicr-track-conclusion.mdThe closing record for the AICR track: the five entries that exist, what they prove, what they refuse to claim, the four lessons that outlived them, and the work deliberately left for whoever picks it up.
aicr-nim-track-backlog.mdThe next fifty tasks in the AICR and NIM track, grouped by theme, with the ones that change shared Catalog and ConfigHub Workshop surfaces marked and the four gating tasks named.
stacks-platforms-apps-taxonomy.mdThe settled taxonomy for stacks, platforms, and apps: the menu of four ways to get a stack, altitude and planes, the certify loop, two-layer adapt, apps operated in ConfigHub, platform as an outcome; the spec the Stacks and Apps pages are rebuilt from.
aicr-pilot-variants-brief.mdBrief for applying the Pilot variant model to AICR platform shapes: what parity means when the unit of change is a platform, the blast-radius level the entries' receipts already record, and the control-point map each entry needs.
aicr-signature-verification.mdMaintained decision for verifying upstream AICR signatures: pinned cosign in a container, a committed sigstore trust root, signed timestamps for Rekor v2 era entries, offline execution, and what the check does not cover.
installer-package-signing.mdHow every published installer package is signed at its immutable OCI manifest digest, how users verify the catalog publisher, and what a valid signature does not say about cluster suitability.
aicr-version-refresh-brief.mdBrief for a second retained AICR version: what upstream changed since v0.14.0, why signature verification is the rung that justifies the refresh, what the derived starter costs, and the probe that decides whether to build it.
chart-hooks-what-happens.mdPractical answer to "if my chart has hooks, what happens with the ConfigHub/installer version?": routes, phases, who runs each, and that a known route is not an automatically executed one.
demo/redis/demo-script.mdRunnable Redis walkthrough.
demo/redis/function-scan-lane.mdRedis ConfigHub function scan lane.
demo/redis/safe-ops-lane.mdRedis safe operation lane.
demo/redis/ux-acceptance.mdRedis demo acceptance criteria.
kubara-git-to-confighub-blog-brief.mdFuture public article brief using the same six-step adoption journey, buyer wording, proof boundaries, and GUI evidence order as the website.
nim-ngc-license-read.mdCited read of the NVIDIA NIM and NGC licensing surfaces with the config-plane verdict for the inference entry: retain the Apache-2.0 scaffolding, never touch the gated runtime artifacts, keys stay target facts.
pathway-route-hooks-transparently.mdTest pathway (Helm-user + SecOps × hook/lifecycle): render parity proves the non-hook objects, the hook runs as an explicit non-automatic route with a receipt, shown side-by-side vs. what Helm/Argo/Flux do silently. Anchored on fluent/fluent-bit; live proof via npm run hook-test:proof.
sveltos-fleet-brief.mdBrief for the Sveltos fleet example in the Kubara style: a reference fleet we operate, governed changes with receipts, and a DIY path for readers with their own clusters. Chapters one and two are recorded live; the kyverno, canary, patching, and bulk-operation chapters are proven offline and await their live runs.
remote-url-oci-probe.mdThe live probe of the Sveltos remote fetch path: which OCI artifact shapes a ClusterProfile deploys today, which two fail and why, and the Secret type and TLS the fetcher requires. Mirrored from the Sveltos repository.
chapter-three-gateway-rework.mdThe specification for moving the Sveltos chapters onto the ConfigHub OCI gateway: what each runner becomes, what it deletes, and the two constraints the gateway imposes. Mirrored from the Sveltos repository.

The catalog and its contracts

What the catalog promises, and the briefs behind those promises.

DocumentPurpose
../data/master-catalog-matrix/matrix.htmlBrowser view of the full chart/version/base matrix: user route, strongest evidence, core lanes, production scope, hooks, quirks, hard gaps, and next action.
catalog-consumer-contract-brief.mdBrief adopting the catalog consumer contract from integrator feedback: fetched-bytes digests, published paths, a change feed, retention and schema-version commitments, and the CORS and object-inventory keep-clauses.
catalog-entry-contract-brief.mdBrief proposing one contract for every catalog entry: the three-state rule generalised across every era, plus a permission axis so a forbidden obligation is not read as a gap, a limits axis so each artifact says what it does not prove, and the measured cost of adopting it today.
catalog-refresh-plan.mdPlan for turning the weekly catalog refresh on: why its schedule is off, the never-delete-published-versions retention rule, the two preconditions (a current catalog, and every npm lane with a recorded role), and the gaps still open in the lane.
certified-bundle-track-conclusion.mdThe closing record for the certified bundle track: what it proves, what it refuses to claim, the five lessons that became doctrine, and what it deliberately leaves.
certified-bundle-track-backlog.mdThe next fifty tasks in the certified bundle track, grouped by theme: deciding lanes where evidence sits, extending routes, putting the evidence in front of a reader, finishing the bundle, upstream drift and retention, and the consumers.
chart-successors-brief.mdBrief for replacing the six single-publisher chart sources: the measured breakage, the live-verified successor per component, the redis engine question, and the operator shape tax.
config-catalog-demo-program.mdExecution plan and acceptance criteria for the shared records, AICR, literal OCI uploads, promotions, fleet paths, and Apps.
config-catalog-demonstrations.mdGenerated, plain-English status of the Helm, AICR, cub installer, OCI delivery, promotion, Kubara, and Sveltos pathways and the five planned ConfigHub Apps.
config-catalog-doctrine.mdCanonical business purpose and user journey, including how the Catalog, local advisory checks, promotion checks, and ConfigHub validation find and prevent misconfiguration; also defines the source-neutral model for materializing, flattening, routing, protecting, retaining, and delivering configuration.
Configuration decision contractSource-neutral contract for deciding every finding as a fix, rejection, or scoped exception while keeping local advice, ConfigHub validation, promotion, and delivery evidence separate.
../data/catalog-shared-checks/summary.mdReleased cub check results for every exact maintained Helm base, with scanner and bundle identity, exact input digests, stable controls, and the deliberately partial mapping to chart-specific Catalog reviews.
flattening-alignment.mdSource-neutral decision guide for retaining exact objects, retaining them with recorded setup, or processing the source late.

Working with charts

Charts as they actually behave, including when they misbehave.

DocumentPurpose
ai-assisted-helm-changes.mdPlain-English guide to safe AI-assisted Helm changes: propose, diff, gate, approve, deliver, and observe.
broken-chart-triage.mdPractical triage path for broken charts, values files, target prerequisites, lifecycle routes, image pulls, runtime failures, and model gaps.
creating-variants.mdSimple guide to base variants, derived ConfigHub variants, AI assistant tasks, and bulk creation.
cub-variant-command-surface.mdCurrent cub variant command surface and how Space/Unit metadata maps to derived variants.
derived-variant-walkthrough.mdCaptured real cub variant create read back from its receipt: faithful 7-unit clone, no Helm re-render, gates applied, honest intended-state-only scope.
helm-pain-points.mdUser-facing map from common Helm pain points to current proof, handoffs, and per-chart reports.
helm-user-pain-evidence.mdPublic Helm, chart, GitOps, and operator reports behind the ten configuration questions used by Check my config.
helm-presets-and-values.mdHow public presets map to repo base variants, why the catalog does not claim every values combination, and how AI helps maintain chart-specific choices.
helm-render-intents.mdThe Helm-specific source-and-intent record, with examples and generated coverage states for render inputs, captured output, lifecycle routes, target prerequisites, runner evidence, freshness, and incomplete contracts.
helm-to-cub-migration.mdHelm-fluent dev's cheat-sheet: how Helm habits (--set, -f values.yaml, --set image.tag, --set-string, release names) map to cub's declared-input / named-base model, and why the declared model catches the typo footgun Helm absorbs.
helm-upgrade-crash-example.mdUser-facing day-2 example: how an opaque Helm upgrade becomes staged, reviewed, rehearsed, gated, and observed.
existing-helm-release-diagnostic.mdRead-only commands for capturing an existing Helm release, comparing a candidate render, finding upgrade and rollback hazards, and understanding what the files cannot prove.
serious-chart-proof.mdConcise kube-prometheus-stack proof path for the serious chart example.
serious-charts.mdThe serious-chart packets: why kube-prometheus-stack, cert-manager, and external-secrets are the hard cases and how to read their status.
variants-after-upload.mdThe step-by-step walkthrough after cub installer upload: why and how to run cub variant create (labels, namespace, gates), preview with cub variant promote --dry-run -o mutations, promote, and where delivery and rollback pick up. Commands verified against cub --help.

Packages and delivery

Getting a package, looking inside it, and shipping it.

DocumentPurpose
change-routing-before-oci.mdShort guide for choosing a preset/base variant, derived ConfigHub variant, or delivery prerequisite before OCI handoff.
day2-upgrade-rollback.mdDay-2 explainer: how an opaque helm upgrade becomes a staged/reviewed/rehearsed/observed change, and how versioned-Unit rollback works (with honest irreversible-migration caveats).
ci-render-check.mdTurn a WorkshopResult into a bounded Markdown or JSON CI report locally, then optionally post it in GitHub, GitLab, Jenkins, or another review system.
nim-coverage.mdNIM on this catalog in one page: the retained KServe shapes with one readable example, the NGC licensing boundary, and a plain list of what is not covered.
approval-story.mdThe recorded dev-to-production cycle: preview, revision-bound approval, promotion, one-target rollback - with the approver-identity gap named honestly.
check-and-promote-with-ai.mdBrowser-local Check and Promote flows, source-aware field attribution, Catalog lifecycle context, exact target results, AI prompts, and the optional ConfigHub handoff.
anonymous-browser-workshop.mdThe bounded no-sign-in browser path: inspect rendered YAML, compare exact objects, run static checks, download one complete result for local AI or CI, and understand which work remains local.
test-candidates-before-promotion.mdWorked NGINX example: test three exact configurations against one fixed workload and destination requirement, select the smallest passing candidate, then promote and deliver that same object set.
example-rendered-diff.mdOne real rendered diff, redis 25.5.3 to 27.0.0, computed from the two committed renders: same 14 objects, 85 changed lines, a dropped ACL block, and checksum rotations that restart pods.
day2-upgrade-story.mdThe day-2 upgrade path for catalog configuration and flattened bundles: value-model diffs first, control points and immutable fields checked, verdicts re-read per version, bases recreated in the safe order.
inspect-oci-package.mdOne command for identifying an OCI package, resolving its digest, listing exact Kubernetes objects and lifecycle clues, and checking the permanent public NGINX example.
installer-oci-packages.mdPublic installer package OCI refs, how users pull them, and how package OCI differs from ConfigHub delivery OCI.
installer-package-signing.mdVerify the publisher and exact manifest digest of a Catalog installer package with cosign.
prometheus-overlay-promotion-example.mdWorked Prometheus example for a values overlay and a ConfigHub-only promotion variant.
ci-rendered-catalog-journey.mdFor teams whose CI already renders their chart catalog into YAML in git: land those exact files as governed data with variant upload, receipted canonically equal, then the four things text cannot do, a fleet query, a change with its reason in history, a staging variant with a protected choice, and a release by digest, with the commercial boundary for private catalogs at CI scale stated plainly.
real-human-trial-protocol.mdThe repeatable twenty-minute trial of the public site: who to ask, the unbriefed invitation to send, what one trial yields through the Site trial feedback issue template, how to read stalls and distrust findings, and the boundaries, no analytics, no task score, no names in the record.
serverless-mode.mdA local path that needs neither ConfigHub Server nor a ConfigHub account. It distinguishes those two choices, compares Helm install with cub render plus apply, and shows how reviewed files can become OCI for GitOps.
transform-oci-package.mdChange one field in a literal Kubernetes OCI, preserve the input records, build a new local OCI, and inspect the separately published permanent example.

What is proven, and what is not

The honesty surface: evidence, gaps, and refusals.

DocumentPurpose
../data/outcome-evidence-contract/summary.mdGenerated user-outcome contract: question, status, evidence, verifier command, scope, and next action.
../tests/README.mdTest directory map: npm verification scripts and portable runtime tests.
../tests/npm-scripts.mdWhat each npm script family checks, why it exists, and when to run it.
current-proof-status.mdShort guide to current proof status and the generated summaries that are authoritative.
hard-questions.mdSkeptical user route: hooks, upgrades, overlays, false-green sync, free versus managed, and how to challenge the model.
known-gaps-we-surface.mdUser-facing guide to current watch findings that the project deliberately surfaces instead of hiding.
question-intake-operation.mdHow maintainers turn a public Helm, AICR, OCI, or YAML question into a reproduced Catalog answer, named refusal, or evidence decision.
outcomes-and-tests.mdUser-facing outcome and test map, with links to the front-door CSVs.
verification-lanes.mdPlain-English definition of each verification lane and the commands that check them.
verification.mdLanding page for npm proof commands, user-side checks, committed evidence, fresh live lanes, and render-record-route.
what-we-refuse-to-claim.mdTrust boundary: why watchlist rows and strict witness blocks remain visible.

Design notes

Where a design is written down before it is built.

DocumentPurpose
agent-operated-cross-format-catalog.mdPlan for an installable ConfigHub Workshop agent skill, task evaluations, a source-neutral Catalog, and the first Timoni source entry.
config-workshop-ai-drift-growth-strategy.mdDecision-ready positioning for the Workshop, its AI-assisted question path, the four drift stages, a thirty-day demand test, and the ranked growth backlog.
config-workshop-simulation-findings.mdCanonical record of the deterministic and qualitative website journey checks, the August 23 three-click results, the remaining managed-operation journey, their limits, and the human tests still required.
custom-stacks-and-apps.mdProposal for custom stacks and the cub-noun vocabulary. cub app names the workload spine, cub stack installs a certified composition by name, the platform is the governed keystone, and cub server is self-hosted ConfigHub. The certify step is the moat, and it reuses the certified bundle engine.
cub-noun-vocabulary.mdThe cub noun and verb table, using the get-started tutorial's own commands. A config is checked and deployed for free, then uploaded into ConfigHub, released to go live, and promoted across environments. A stack adds certify and sandbox like eks-inference. upload (cub variant upload) is where public config chains into your private org, and it is the account line; upload keeps install out of the table so it never collides with cub installer. deploy consumes an existing OCI, while release produces ConfigHub's governed one.
site-ia-phase-2.mdThe second-pass execution plan for the website's structure: the three decisions of 6 September 2026 (the site is what a user reads and runs, five buttons Catalog · Config · Stacks · Operate · Docs, what Config must explain), the target page map, who defines each term, ten steps each naming its page, sources, satellites and gates, the new writing ranked, and the open questions.
site-ia-phase-3.mdThe third-pass plan for the content and wayfinding on the pages a reader lands on: the purpose (earn trust, prompt action), the code-block doctrine (comment above the command, no wrapped lines), the left and right column redesign, the top pages hand-crafted by the maintainer with the site session implementing, and the home page fully specified from the maintainer's notes as the first worked page.
workshop-ux-readiness.mdPinned UX preparation protocol, six mission briefs, scoring, and complete Guide portfolio gap accounting.
workshop-execution-plan.mdSingle active Catalog and Workshop execution plan: all 42 backend tasks, eight journey acceptance blocks, evidence ledger, priorities and dependencies.
workshop-frictionless-entry-plan.mdThe umbrella plan for a frictionless entry point for individual developers and for ConfigHub Workshop: the leadership thread and the mission, config as data as the position, the Workshop as catalog plus tools plus a frictionless doorway into ConfigHub (the Moat around the Castle), the ladder framings we have looked at listed as options rather than chosen, the engine behind the door, the pull-and-make flywheel, the cohort at each band, a proposed sequence, what is already proven, and the open decisions with their options. Two companion documents carry the detail.
workshop-stories-entry-mid-keystone.mdCompanion to the Workshop plan: the stories behind each band, written in full and anonymised. For entry, spine, and keystone, the doctrine's worked examples (the assistant does the easy part, the gate does the safe part), the six personas' success signals, the app proofs, and the three real wins of the first cohort, each mapped to the gap it named and the ladder options it exemplifies, with a reference table and sources.
workshop-ai-api-plan.mdCompanion to the Workshop plan: the API for AI use cases. The premise that the user is AI-first and the agent is both maker and buyer, the nine things an agent needs to do and where each lives, the machine surface that exists today, the concrete gaps, the static-or-SaaS split, every marketplace motion mapped to a ConfigHub primitive, the AI-first flywheel, Dan's ladder of progressive disclosure, a machine-card format per listing, and the delivery options listed as options with a cost and risk table and a proposed sequence.
workshop-context-and-handoff.mdHandoff package for the catalog backend session to take the Workshop strategy on: what to take on and the working agreements, pointers to the three plan docs, the mission and leadership context, a decision log of the framing developed and corrected turn by turn, the entry-point and first-step-to-value inventory with what is really missing, the open decisions the maintainer holds, and what has shipped since the review (site work, the plan amendments, and the AICR NVIDIA-catalog mirror in flight).
site-information-architecture.mdThe plan that maps every one of the site's 45 pages to the three things the definition names, a verified catalog, platforms and stacks on demand, and operating with ConfigHub, with a disposition for each: keep, merge, retire as a redirect. Five nav items, one hub per section with a Run it block that links the plugin at stable paths, the leftover prototypes to remove from this repository, and the order it lands in with a definition of done.
ladder-on-spine.mdA three-lens site review found the value ladder is not on the site's spine: the site jumps check to promote and skips upload, release, and the composition rung. The plan puts the ladder on the deployment page, reframes the ConfigHub page as the account spine with the chaining story, gives the stack and inference rungs an honest home, and keeps the composition certify gate marked roadmap. All of it from shipped verbs.
composition-certification.mdProposal for the composition certify step a cub stack verb needs: the eight-check composition verdict, the three producers that already enforce slices of it at three strengths, the composition digest they already compute, a staged plan that proves the verdict on eks-inference and a Kubara platform before hardening it on a broken one, and the boundary where the assistant composes while the engine certifies.
roadmap-2026-09.mdThe work queue going into September 2026, written so any operator can continue from a cold start. Orders the replica track by issue, lists the delivery and proof follow-ups, the vocabulary alignment, the human-gated proofs with their owner steps, the engineering builds and their owners, and the parked migration PRs that stay with the review loop.
website-ladder-review-2026-09-01.mdReview brief gathering one arc for the review loop: the stack manifest becoming a runnable surface, the prototypes becoming the workshop plugin with all four nouns, the ten-cluster fleet generated through the governed verbs up to a named capacity boundary, and the homepage change that puts the ladder on display. Focuses the reviewer on what an arriving user can do, what the OCI catalog is worth, and whether the journeys stand alone well enough to carry posts, with a by-hand checklist.
oci-design-center.mdThe design center stated once: every result is an OCI image, every stack an index of images, every release a flattened image, one receipt linking them, and no choice without a digest. Fixes the artifact shape the workshop plugin publishes and verifies, the digest rule, what the Catalog holds, the migration path for shipped examples, and the open items including the product half.
stack-manifest-spec.mdThe stack manifest specified as the neutral artifact every producer emits and the shared verbs consume: three component forms, bundle by digest with receipt verification, committed render, and first-class authored YAML, plus planes and order, the declared bindings the single-owner check and the generic upload read, and the armed regression gate that refuses a new finding, a regressed check, or an unrecorded digest change.
eks-inf-replica-plan.mdStaged plan to manufacture the hand-built EKS inference platform from certified Catalog parts and match the original or name its departures. Stage A.1 is done: a derived closure map shows all eight components resolve to retained supply, three rebuilding from certified eks-inference chart variants and five selecting literal bundles by digest. Later stages rebuild and compare objects, judge the whole composition, rebuild the ConfigHub organization with generic tooling, and accept the inference workload on simulated GPU capacity, with the real-GPU run kept as the final rung.
demand-to-verbs.mdThe demand map grounding the site's journey and the proposed cub verbs in written users and questions: six personas, the ten research-counted practical questions, the five ConfigHub Apps and three stack families that answer them, the who-asks-what-answers-which-verb chain, the measured entry-spine-keystone funnel, and the build order the counts dictate.
promotion-workshop-test-program.mdAdversarial test program for a public promotion review: exact candidates, destination differences, flattening refusals, lifecycle routes, stale evidence, partial fan-out, and the ConfigHub handoff.
what-config-workshop-is.mdConfigHub Workshop in three parts: a verified catalog of tested configuration in one OCI form, where every image works in ConfigHub, updates reach you without losing your changes, and each image is verified, certified, and signed; platforms and stacks on demand for people and their assistants, checked before anything runs; and a ConfigHub plugin to operate apps, platforms, and stacks correctly. Each part ends with what you can do and what problem it solves.
confighub-data-model.mdHow Helm, AICR, generators, OCI, YAML, and ConfigHub fit one processing model: materialize exact objects, decide whether to flatten, attach lifecycle routes, retain variants, publish, reconcile, and observe.
cub-scout-diff-design.mdUser-facing design for one field-level desired-vs-live differ that can serve dry-run and drift across Argo, Flux, or cub-direct delivery.
generative-gitops-fit.mdUser-facing boundary between the broader Generative GitOps thesis and what helm-expt proves today.
model-and-vocabulary.mdOne page for the source-and-intent role, the five main terms (recipe, rendering, render record, base variant, derived variant), the F1-F4 stages, and where each artifact lives today.
configuration-question-workflow.mdThe decision-led workflow for checking an unknown configuration, comparing it with known or current config, retaining the reviewed result, and turning accepted public cases into Catalog entries.
reverse-reconcile-design.mdUser-facing design for the reverse live-to-desired frontier: authority policy, bounds check, fixture receipt, and the missing cub write-back capability.
runtime-drift-boundaries.mdWhat local files, kubectl, Argo CD or Flux, and ConfigHub can each say about live state, pruning, conflicts, convergence, and workload health.

Understanding the harness

What the machinery is and why it holds together.

DocumentPurpose
chain-of-proof.mdWhich tool proves each boundary, plus the generated source-to-observation record for Helm, AICR, cub installer, Kubara, Sveltos, and literal Kubernetes configuration.
choosing-commands.mdGuide to the distinct roles of cub helm template, cub helm install, cub installer, ConfigHub upload and variant commands, and repo verifiers.
how-it-works.mdThe "how it works" hub: the four-move mental model (render → route → deliver → observe) plus customize/promote, day-1, day-2, secrets, and free-vs-managed - an organized index into the mechanism docs.
how-the-harness-works.mdShort technical explanation of the harness lifecycle and where user value is created.
introduction-to-the-harness.mdDetailed recipe-generation workflow and the table for where Helm pieces belong.
why-this-does-not-collapse.mdSkeptic-facing explanation of how hooks, quirks, config volume, and blocked rows are routed.

Adopting and operating it

Bringing existing work in, and running it afterwards.

DocumentPurpose
adopting-existing-apps.mdHow existing Argo, Flux, KRM, rendered-manifest, and live-resource apps enter the ConfigHub model.
app-to-live-walkthrough.mdThe full chain from plain YAML into a base Space, staging and production variants, release OCI, Argo CD delivery, and staged promotion. It links to the smaller four-object fixture when import is the only question.
cub-deployment-path.mdHow deployment works on the cub path: source package → ConfigHub Units → release OCI → Argo or Flux, with direct local apply kept as a portability test; also covers OCI pull credentials and application Secrets.
custom-overlays.mdPlain-English ExternalDNS example for wrapper charts, platform values, customer overlay values, and target facts.
extension-slots.mdGeneral guide for raw manifests, tpl snippets, sidecars, config blocks, and other NGINX-like extension slots.
gitops-adopter-guide.mdFor Argo/Flux teams: what changes (the source becomes one OCI bundle) and what doesn't (your controller); Argo proven, Flux/cub-direct in progress; vs raw Helm-through-Argo; hooks as explicit routes.
image-registry-migration.mdMove a digest-pinned image to another registry, promote it to three environments, keep one environment pending, and verify the current ConfigHub records with a live or offline command.
large-config-operations.mdUser-facing guide for watching large ConfigHub upload/apply/GitOps paths without collapsing them into a vague hang.
maintenance-sla.mdHow catalog entries are refreshed, patched, and supported.
product-support-tiers.mdWhich Helm scenarios fit the public catalog, managed imports, or commercial support.
production-support-decisions.mdHow a review-ready chart becomes production-supported for one target scope.
target-prerequisites-before-rerun.mdThe action packets for a non-green row's target prerequisite: create-namespace / stage-secret / install-crds / provide-external-service / provide-storage-or-topology / operator-review, with required inputs and the rerun command.
target-prerequisites.mdHard-chart guide: CRDs, Secrets, lifecycle checks, and target facts that must be staged or observed beyond YAML parity.

Reading the evidence

Where to look, and how to check it yourself.

DocumentPurpose
live-parity.mdUser-facing guide to pass, watch, blocked, and rerun rows in the live parity lanes.
reading-the-matrix.mdHow to read the master catalog matrix: the lanes, the G/P/K shorthands, the cell states (pass/watch/blocked/n-a/blank), and where to look when a row is not green.
top100-readiness.mdUser-facing guide to the top-100 corpus buckets and generated readiness data.
top100-status.mdPlain-English top-100 answers: what works today, what needs prerequisites or review, and how it differs from plain Helm.
verify-it-yourself.mdReader verification commands for corpus checks, rendered installs, parity receipts, and cub-scout receipts.
why-synced-is-not-working.mdRuntime false-green explanation: why sync/object presence does not prove workload health.

Specific charts, scale, and security

Notes that answer a narrower question than the guides above.

DocumentPurpose
../data/chart-use-guide/summary.mdGenerated chart-use guide: one short answer per top-100 chart for use now, promotion review, base-variant work, or limitation decision.
../data/claims-register/summary.mdGenerated claim-to-evidence register for checking whether a public claim is backed, partial, planned, or refused.
../data/confighub-example-guides/summary.mdGenerated plain-English guides for each public chart preset config: what was rendered, why it is the right starting point, how to repeat it, and what prerequisites or lifecycle work remain visible.
../data/cub-adoption-caveats/summary.mdPer-chart adoption caveats: for each of the 100 charts, which "worse-than / more-confusing-than plain Helm" rough edges apply (baked shared password, CRD-first ordering) with the exact remediation, plus the three universal cub-direct caveats - the per-chart companion to the adoption audit. CSV + colored HTML alongside.
enterprise-parity-contract.mdReference contract for private enterprise estates that combine first-party Helm charts, values-only repositories, and Argo ApplicationSet fan-out.
hook-lifecycle-strategy.mdHow Helm hooks are inventoried, classified, translated, or blocked.
nginx-configuration-files.mdNGINX-specific guide for config-file extension slots, base variants, derived variants, and future nginx -t checks.
prometheus-high-fanout.mdKube-prometheus-stack example showing how one base choice changes many objects and target prerequisites.
remote-images-and-supported-bases.mdWhat to do when a catalog base is watch-grade because an upstream image disappeared: refresh the chart/base, override the image, pin/mirror a digest, route a lifecycle image, or watch/refuse.
security-end-to-end.mdThe "no silent privileged step" story end to end: app secrets (generated/existing/target-fact), delivery creds (copied not printed), RBAC visible in the recipe, and the scan + F/G adversarial lanes.

Maintained Orientation Layer

The project knowledge index is a short maintained orientation layer for humans and agents. It summarizes the current model and routes to authoritative evidence. It is not a second source of truth; generated data, receipts, recipes, current issues, and verifiers remain authoritative.

The delivery-pattern survey maps seven Flux and Argo CD pattern families onto the catalog OCI shapes, with source links, assumptions and route limitations. The d2 family awaits the maintainer's layout list.

Agent And Operator Docs

These files are for agents and maintainers operating the repo. Keep public website pages and docs/user pages human-readable; put command routing, recovery steps, and repo-specific AX notes here.

FileRole
agent/README.mdCanonical start page for agents and maintainers working in this repo.
agent/tasks.mdTask-to-command recipes for common repo operations.
agent/recovery.mdRecovery guide for stale generated files, broken links, command-surface failures, and live-lane failures.
agent/verification.mdAgent-focused map from claims to the narrowest useful verifier.
agent/catalog.mdRead-only catalog guide for inspecting chart evidence without redesigning the catalog.
agent/human-agent-doctrine.mdRules for keeping human-facing pages and agent/operator pages consistent without copying the same text.
agent/terms.mdCompact glossary for repo-specific product, proof, catalog, and lane terms.

Generated And Reference Markdown

LocationRole
recipes/<repo>/<chart>/<version>/README.mdGenerated recipe summary for one chart/version.
recipes/<repo>/<chart>/<version>/CATALOG.mdGenerated per-chart catalog page with supported variants, revisions, receipts, and package links.
recipes/<repo>/<chart>/<version>/weirdness-and-mitigations.mdChart-specific notes for top-20 proof entries.
config-catalog/package-extras/<repo>/<chart>/<version>/README.mdMaintained source for package-local lifecycle instructions copied into the public installer package.
packages/<repo>/<chart>/<version>/README.mdGenerated cub installer package usage note. This is package reference, not a primary user guide.
packages/<repo>/<chart>/<version>/records/README.mdGenerated index for the source, render-intent, object, and lifecycle records carried inside one public installer package. This is package reference, not a primary user guide.
packages/<repo>/<chart>/<version>/prerequisites/<action>/README.mdPackage-local instructions for one prerequisite or lifecycle action, including when it runs, what it creates, and how its receipt is recorded.
docs/demo/<chart>/confighub-proof.mdPer-chart ConfigHub proof summary. Corpus evidence.
docs/demo/<chart>/confighub-proof-transcript.mdPer-chart command transcript and observed outputs. Corpus evidence.
data/*/summary.mdGenerated status, review, or matrix summary.
data/README.mdGenerated data index and start-here guide for CSVs.
data/csv-index.csvGenerated machine-readable index of every CSV under data/.
data/status-dashboard/summary.mdGenerated front-door status dashboard for top100, proof lanes, hooks, quirks, GitOps, and live parity.
data/outcome-evidence-contract/summary.mdGenerated front-door map from user-visible outcomes to evidence, verifier commands, limits, and next action.
data/chart-use-guide/summary.mdGenerated front-door guide for whether one top-100 chart can be used now, promoted after review, improved with a better base, or held for a limitation decision.
data/confighub-example-guides/summary.mdGenerated per-preset guides for the chart journey into ConfigHub, including the render intent, render variant, package OCI ref, scripts, prerequisites, and evidence boundary.
data/master-catalog-matrix/matrix.htmlGenerated browser front door for chart/version/base status, user route, strongest evidence, production scope, and visible gaps.
data/useful-base-design-queue/summary.mdGenerated front-door queue for proof-grade top-100 charts whose current base is too default-shaped to recommend.
data/claims-register/summary.mdGenerated claim-to-evidence register used to keep public claims scoped and reviewable.
data/blast-radius-accuracy/summary.mdGenerated blast-radius accuracy seed: predicted affected objects compared with actual committed rerender diffs.
data/top100-readiness/next80-queues.mdGenerated next80 operating queue: promotion review, user-shaped variant work, and limitation review.
data/outcome-coverage/*.csvFront-door CSVs for chart, base, derived variant, and feature outcomes.
data/latest-top20-refresh/candidates/README.mdGenerated summary for latest-version candidate proofs.
runs/**/*.mdGenerated execution output from Pilot or other run harnesses. These are receipts/log summaries, not narrative docs.

Do not hand-edit generated Markdown unless the generator is also updated.

Manual Docs By Role

Core Model And Harness

FileRole
chart-recipe-manifest-flow.mdCanonical object model: chart, recipe, variant, revision, rendered objects, receipts.
seven-stage-helm-lifecycle.mdDoctrine for render parity and for routing hooks, CRDs, target facts, generated values, overlays, GitOps, and observations.
offering.mdPublic overview of the Helm catalog offering and the free-to-managed adoption path.
generative-gitops-fit.mdUser-facing map from generated config and AI/GitOps expectations to current helm-expt evidence and limits.
try-now.mdWalkthrough capture behind the short public package exercise; the Redis Helm-parity comparison now lives on Try Redis.
choose-your-path.mdRoute picker across input formats and local, anonymous, or ConfigHub-managed use.
tutorial-sequence.mdDetailed proof flow with commands and expected results; not the ConfigHub product tutorial.
current-proof-status.mdUser-facing entry point for current proof status.
hard-questions.mdSkeptical user route for the hard questions a Helm reviewer asks before trusting the catalog.
what-we-refuse-to-claim.mdUser-facing trust boundary for strict witness blocks, watchlist rows, and normalization rules.
../data/claims-register/summary.mdGenerated register mapping public claims to evidence, limits, and scoped verification commands.
why-synced-is-not-working.mdUser-facing explanation of object-set false greens and runtime/prerequisite checks.
target-prerequisites.mdUser-facing explanation of target facts and live prerequisite checks for hard charts such as cert-manager, VPA, and OpenTelemetry Operator.
why-this-does-not-collapse.mdUser-facing explanation of why hooks, quirks, and config volume are handled as routed facts instead of hidden risk.
verification.mdUser-facing landing page for proof commands, committed evidence, fresh live lanes, and render-record-route.
verify-it-yourself.mdUser-facing commands for independently checking repo, render, parity, and cub-scout receipt evidence.
production-support-decisions.mdUser-facing path from production-review-ready to a target-scoped production support decision.
chain-of-proof.mdUser-facing proof-boundary map across helm-expt, cub installer, ConfigHub, GitOps, and live observations, linked to the generated OCI evidence-chain records.
top100-readiness.mdUser-facing guide to public catalog rows, promotion candidates, default-only rows, and limitation decisions in the top-100 corpus.
outcomes-and-tests.mdUser-facing map from promised outcomes to tests, CSVs, and commands.
../data/outcome-evidence-contract/summary.mdGenerated product-facing contract for current outcomes, evidence, scopes, and next actions.
../data/useful-base-design-queue/summary.mdGenerated queue of proposed useful base shapes for top-100 charts that are not yet good catalog offers.
live-parity.mdUser-facing explanation of strict live parity status, non-pass rows, and rerun rules.
large-config-operations.mdUser-facing guide for large charts and 100+ Unit operations: what to watch, what to capture, and how to classify slow or progressing stages.
helm-pain-points.mdUser-facing map from common Helm pain points to helm-expt, ConfigHub, installer, and live-observation answers.
helm-upgrade-crash-example.mdUser-facing upgrade-crash example linking old/new render, blast radius, live rehearsal, ConfigHub gates, and scoped support decisions.
verification-lanes.mdUser-facing explanation of proof lanes and command checks.
how-the-harness-works.mdLifecycle-stage explanation of the harness.
introduction-to-the-harness.mdDetailed import workflow, recipe decisions, and hook policy.
choosing-commands.mdUser-facing guide to source, preparation, ConfigHub, delivery, and proof commands.
creating-variants.mdSimple user guide for base variants, derived ConfigHub variants, AI assistant tasks, and bulk creation.
cub-variant-command-surface.mdCurrent cub variant syntax for create, promote, and upload, including labels, annotations, targets, namespaces, and gates.
change-routing-before-oci.mdUser-facing routing guide for presets/base variants, derived variants, and delivery prerequisites before OCI handoff.
adopting-existing-apps.mdUser-facing adoption path for existing Argo, Flux, KRM, rendered-manifest, and live-resource apps.
custom-overlays.mdExternalDNS managed overlay example with wrapper chart, platform values, customer values, target facts, and ConfigHub variant routing.
prometheus-overlay-promotion-example.mdConcrete Prometheus example showing when an overlay becomes an installer base and when a change is ConfigHub-only.
prometheus-high-fanout.mdKube-prometheus-stack example for base-variant fanout and prerequisite tracking.
extension-slots.mdGeneral extension-slot routing for raw manifests, tpl snippets, sidecars, config blocks, and add-on slots.
nginx-configuration-files.mdNGINX-specific config-file routing: Helm extension slots, derived ConfigHub variants, and future syntax checks.
product-support-tiers.mdFree, public, managed, and commercial support boundaries.
maintenance-sla.mdCatalog maintenance and update expectations.

Tutorial UX Proposal Companions

These are product sketches linked from tutorial-sequence.md. They are not extra first-pass requirements; they show how each tutorial's CLI steps could become a simpler human-facing Creator-style flow backed by formal contracts, AX/FX, and current ConfigHub primitives.

FileRole
ux-proposal-redis-quick-start-tutorial.mdProposal for turning the Redis quick-start render/upload path into an intent-first base creation flow.
ux-proposal-redis-secret-modes-tutorial.mdProposal for presenting generated Secret versus existing Secret as a base-variant choice with target fact checks.
ux-proposal-prometheus-base-variant-tutorial.mdProposal for presenting Prometheus server-only as a base-variant creation flow.
ux-proposal-prometheus-promotion-tutorial.mdProposal for presenting Prometheus promotion as a derived ConfigHub Creator flow.
ux-proposal-externaldns-custom-overlay-tutorial.mdProposal for separating ExternalDNS render-time overlay choices from customer Creator choices.
ux-proposal-gitops-runtime-proof-tutorial.mdProposal for presenting GitOps and runtime proof as publish-and-prove intent.
ux-proposal-bulk-scan-patch-tutorial.mdProposal for presenting scan, patch, review, and approve as a single bulk hardening flow.

Installer And Proof Contracts

FileRole
artifact-verifier-spec.mdWhat the artifact verifier must check.
redis-worked-example.mdThe catalog's reference chart in one place: the proof, installer-package, local end-to-end, scan and variant lanes that every later chart's generated spec follows.
certified-bundle-spec.mdThe shared bundle-plus-receipt spec every producer emits against: bundle shape, receipt fields, quirk dispositions, and the flattening-safety verdict lanes.
deciding-a-flattening-lane.mdHow a lane is decided from a witness: six rules, each drawn from a draft the chart source proved wrong, plus the regeneration order and what a decided lane does not mean.
seven-stage-helm-lifecycle.mdSeven-stage lifecycle, render parity boundary, hook routing, and support claims.
proof-kit-migration.mdHow repeated chart proof scripts are migrated to the shared proof-kit generator/verifier.
catalog-record-lookup.mdExact Catalog record lookup with digest mismatch handling and complete evidence preservation.
config-disruption-review.mdLocal before/after manifest review contract, static risk categories, input limits and machine-readable results.
customization-algorithm.mdReference algorithm for values files, overlays, wrapper charts, and post-render variants.
catalog-doctrine.mdCatalog doctrine: chart → recipes → placeholdered base variants → derived ConfigHub variants.
customization-decision-tree.mdDesign-level routing tree for customization and support outcomes.
complete-corresponding-model.mdCompleteness contract a supported chart must satisfy.
per-chart-recipes.mdMethod for recommending per-chart recipe + variant counts.
top100-user-readiness.mdMethodology for the generated top-100 user-readiness view: sources, bucket rules, and honest limits.
master-catalog-matrix.mdDoctrine for the master matrix: HTML for human/product decisions, CSV for machine/spreadsheet use, Markdown for compact orientation.
residue-families.mdWhat a non-green matrix row's residue category means (remote-image, render-input, capability-profile, hook/lifecycle, target-prerequisite, runtime, controller-health, model gap): who fixes it, the next action, and where the rows live.
matrix-completion-audit.mdHow the matrix completion audit triages every non-green/not-yet-run cell into needs-run / needs-target-or-prereq-fix / needs-modeling / already-decided, and how to use it to finish the matrix faster.
variant-promotion-closeout.mdHow the promotion-closeout queue makes the promotion column actionable: ready-to-run vs watch-grade vs blocked by the ConfigHub changeset bug, with owner classes; promotion is a ConfigHub server value.
what-hook-support-means.mdThe hook disposition vocabulary: what observed/routed/refused/per-target/recipe-needed claim and refuse to claim.
kube-prometheus-stack-serious-chart-review.mdReviewer-facing map of the serious chart: what is proved, partial, and not yet claimable.
helm-quirk-support-matrix.mdHow each Helm quirk class (hooks, CRDs, lookup, capabilities, secrets, slots, RBAC, webhooks, storage) is handled across the seven lifecycle stages, with honest status per quirk.
fork-vocabulary.mdShared vocabulary for base vs derived variants and how names map.
helm-import-contract.mdHistorical. Contract for graduating a direct Helm render into maintained cub installer recipes.
capability-profile-catalog.mdNamed Kubernetes capability profiles used during render.
generated-fact-receipts.mdGenerated secrets, certs, random values, and time-value receipt specification.
secret-lifecycle.mdHow rendered Secrets, target Secret facts, and Kubernetes lifecycle Secret state are classified and checked.
observation-freshness-slo.mdFreshness states for live observations in a workerless ConfigHub model.
upgrade-rollback-receipts.mdUpgrade and rollback receipt shape.
hook-lifecycle-strategy.mdHow Helm hooks are inventoried, classified, translated, or blocked.
direct-cub-helm-model.mdPlugin-specific reference for the separate cub-helm command surface and how it differs from durable installer recipes.

Variants, Promotion, And Operations

FileRole
variant-creation-artifact.mdReference design for guided variant creation over existing ConfigHub capabilities.
cub-variant-command-surface.mdUser-facing command surface for current cub variant create, promote, and upload operations.
redis-variant-creation-plan.yamlConcrete Redis variant creation blueprint.
variant-creator-verification.mdHow Creator previews, checks, receipts, UX, agent, and fleet paths are verified.
derived-variant-live-proof.mdFirst live cub variant create execution receipts for derived ConfigHub variants.
lane-test-doctrine.mdCore corpus lane-test doctrine for every chart-recipe-variant row.
two-cluster-parity-harness.mdStrict Helm-vs-installer live parity contract using two vanilla kind clusters.
enterprise-parity-contract.mdCustomer-safe extension of the parity model for private Helm, values-only, and Argo ApplicationSet estates.
variant-promotion-worked-example.mdWorked promotion examples for Redis and managed overlays.
variant-promotion-model.mdConsolidated model for ConfigHub server-side variant promotion: UX, AX, fleet flow, current top-20 evidence, and product gaps.
promotion-diff-classes.mdThe four classes every promotion diff entry falls into (inherited, overridden, upstream-added, no-op), the presentation each deserves, and the committed BYO NGINX promotion walked through them.
confighub-promotion-mapping.mdMapping between Helm-derived bases, ConfigHub variants, and promotion concepts.
kubara-customized-overlays.mdManaged wrapper chart plus customer overlay test case.

Generated proof data for this section lives in:

LocationRole
data/variant-goldens/redis-prod-us-east/Generated Redis example: redis/default to prod-us-east with preview, checks, and receipts.
data/managed-overlay-goldens/external-dns-customer-acme-prod/Generated ExternalDNS managed overlay example: wrapper chart plus platform/customer overlay values and route classification.
data/lane-test-matrix/Generated lane-test coverage matrix for every chart-recipe-variant row.
runs/derived-variant-execution/Live ConfigHub intended-state receipts for derived variant creation.

Catalog, Scale, And Refresh

FileRole
helm-community-persona-prd.mdProduct requirements for Helm-user personas, free/paid boundaries, and the shift from rendering to day-1/day-2 variant operations.
helm-community-persona-plan.mdExecution plan for making the public repo and catalog valuable to different Helm user personas.
helm-community-persona-reference.mdReference matrix for personas, free/paid capabilities, day-0/day-1/day-2 value, and variant routing decisions.
user-journey-test-pathways-brief.mdBrief for the user-journey test pathways (persona × decisions × chart-class) and the hook-method validation work.
user-journey-test-pathways-plan.mdVerified research map + build plan for the test pathways: the pathway matrix, phase sequencing, live-run policy, and corrections to the brief.
chart-claim-integrity-audit-2026-06-22.mdAdversarial audit of the generated chart pages for false claims (page contradicts its cited receipt): the verified findings, the two systemic root causes, the clarity issues, and the permanent chart-claim-integrity gate that prevents them.
catalog-promotion-review.mdRules for deciding whether a proven chart becomes catalog-supported.
catalog-promotion-next-candidates.mdCandidate charts for the next promotion wave.
serverless-verified-install-plan.mdPlanning model for the no-login verified-install wedge: public catalog package resolution, local apply, in-cluster receipt, and where ConfigHub Server begins.
verified-install-commercial-model.mdCommercial model for verified installs, factory scans, image digest inventory, signed artifacts, refresh SLAs, private catalogs, and fleet security queries.
robust-sceptic-plan.mdSceptic-facing attack taxonomy and adversarial test plan, including claims register, blast-radius accuracy, torture fixtures, environment matrix, and external reproduction.
corpus-rationalization-plan.mdNamed redundancy map across the data views (lane truth, hook family, readiness family) with merge/retire queue and the rules that stop view re-accretion.
maintenance-strategy.mdMaintenance NOTE: the free-tier daily-refresh SLA for public data and tests, perpetual append-only retention of all public data and its changes, and the free/paid boundary backed by a larger daily-updated private corpus.
where-does-my-hook-go.mdProblem analysis and solution proposal generalizing the hook disposition model (observed/routed/per-target/refused) to every Helm behavior that does not survive a config-only render: named routes, the default-plus-legible-off-ramp requirement for humans and agents, a phased plan, and the #684 review notes.
hook-route-execution-plan.mdScope for executing hook lifecycle routes (closing automatic:false): what each route class needs, GitOps-native emission vs product-direct cub execution, the execution receipt, and the phasing so hooks are automated and audited rather than run by hand.
top20-full-proof-target.mdDefinition and status of the top-20 full proof milestone.
../data/top50-completion/summary.mdGenerated fifty-task completion plan: current status, evidence, verification command, and the next step for every agreed programme outcome.
top500-matrix-refresh-review.mdHow the top-500 analysis should be regenerated and interpreted.
latest-top20-refresh-plan.mdLatest-version refresh plan for the supported top-20 charts.
upgrade-story-plan.mdUpgrade proof lane: old/new rendered sets, lifecycle review, live before/after observation, and scoped support decision.
legacy-patch-review.mdReview lane for older chart versions and patch support.
../data/attack-plan-workdown/summary.mdGenerated workdown for import, gaps, variants, production, runtime/GitOps, latest-version candidates, and image digests.
../data/refresh-survival/summary.mdGenerated refresh-survival surface for current supported versions, upstream update candidates, and the kube-prometheus-stack upgrade seed.
../data/hook-lifecycle/summary.mdGenerated hook lifecycle queue and required receipt paths.
known-adversarial-charts.mdPublic chart set used to exercise difficult Helm behaviors.
quirk-coverage.mdWhich Helm quirks we track, scan-but-don't-surface, or don't track yet - the honest taxonomy audit.
verification-properties.mdThe catalog's acceptance contract: properties every proven chart must satisfy.
next-20-tasks.mdDistilled near-term catalog task queue.

Redis Proof Specs

The seven separate specifications that used to sit here are now one document, redis-worked-example.md, listed under Installer And Proof Contracts above. Redis is the reference chart, so its lanes are the ones written out in full; every later chart is generated from a declarative spec through scripts/lib/proof-kit.mjs.

Demo Docs

LocationRole
demo/redis/demo-script.mdRedis walkthrough script.
demo/redis/function-scan-lane.mdRedis ConfigHub function scan lane.
demo/redis/safe-ops-lane.mdRedis safe operation lane.
demo/redis/ux-acceptance.mdRedis UX acceptance criteria.

Tests

LocationRole
../tests/README.mdEntry point for the repo's test material.
../tests/npm-scripts.mdNPM script guide: what, why, how, mutability, and recommended checks by change type.
../tests/strategy.mdLong-term catalog testing strategy.
../tests/runbook.mdReproducible per-chart runtime test procedure.
../tests/findings.mdCurrent runtime findings and guardrails.
../tests/adversarial-strategy.mdAdversarial usage-test methodology.

Operating Skills

These docs turn repeated hard-chart work into reusable operating knowledge for humans and agents. They are not product proof tables, and they do not create a dependency on Pilot or any other external demo runner.

FileRole
skills/README.mdEntry point for helm-expt operating skills and common rules.
skills/live-parity.mdHow to run and classify live Helm-vs-ConfigHub parity, GitOps/OCI, kind, and cub-scout evidence.
skills/large-app-evidence-funnel.mdHow to break a 100+ Unit app into rendered, uploaded, pulled, synced, converged, and controller-health stages.
skills/target-facts-and-lifecycle.mdHow to route existing Secrets, CRDs, APIService readiness, topology, storage, and generated runtime state.
skills/hook-and-secret-lifecycle.mdHow to classify Helm hooks and Secret lifecycle rows without claiming execution from render evidence.
skills/serious-chart-playbooks.mdStart points for kube-prometheus-stack, Consul, cert-manager, External Secrets, Loki, and Argo Workflows.

Planning, Review, And Communications

These files support project planning, review, and public explanation. They are not the primary user path.

FileRole
roadmap.mdCanonical roadmap index: active workstreams, release guardrails, issue trackers, and which planning files are authoritative versus historical.
nvidia-fleet-catalog-requirements.mdRequirements baseline for CRD-defined fleet reconciliation, vetted signed digests, build/install boundaries, fleet classes, interruption-safe upgrades, separate confidence evidence, and the unmeasured 100–1,000 scale target.
../data/top50-completion/summary.mdCanonical generated completion view for the agreed Top 50. Source: config-catalog/top50.yaml; verifier: npm run top50:completion:verify.
agreed-execution-plan.mdConsolidated doctrine and historical execution record. Use current handover, status dashboard, and next-task docs for current counts.
current-handover.mdCurrent pickup snapshot, including the base-variant versus derived ConfigHub variant vocabulary.
large-machine-handover.mdStrict handover for a Codex instance that can run live Kubernetes and GitOps proof lanes.
large-machine-roadmap.mdOutcome-driven roadmap for live parity, target-bound derived variants, chart quirks, and docs organization.
next-execution-plan.mdCurrent execution plan and near-term backlog.
backend-seven-day-plan.mdHistorical 42-task backend checklist and dated evidence record; current execution is in workshop-execution-plan.md.
next-execution-plan-2026-06-helm-proof.mdHistorical. Helm-proof execution programme, retained for traceability after the current plan moved to the cross-format check, retain, promotion, release, and observation journey.
serverless-verified-install-plan.mdProduct planning for cub installer as a no-login verified-install transcript and upgrade path into ConfigHub.
verified-install-commercial-model.mdPaid-tier planning for scan receipts, image digest inventory, signed artifacts, private catalogs, and fleet-wide security operations.
robust-sceptic-plan.mdAttack model and adversarial test queue for public claims, lifecycle gaps, blast-radius prediction, and external reproduction.
issue-backlog.mdIssue index and execution order.
independent-review-brief.mdBrief for an independent product/technical review.
review-prompts.mdLonger review prompts for structured critique.
pilot-adversarial-testing.mdPlan for adversarial testing with external automation.
pilot-demo-proof-plan.mdThe demo-proof plan: the head-to-head benchmark design (bare model+helm versus Pilot+substrate), the agent-drivable variant generator, and the switch-effect maps, with the honest-run protocol.
outside-user-test.mdFifteen-minute outside-user protocol for testing whether the public site and docs explain the product.
blog-posts.mdPublic writing plan.
dedicated-website-plan.mdStandalone website plan.
house-voice.mdThe house voice for all tech copy: the Fowler/Jobs/Dickens hybrid, ten rules, ready-to-apply before/after rewrites of the front-door site surfaces (Get Started, How It Works, chart-page intro), and how to check it (Flesch 60–70).
onboarding-and-entry-paths.mdDoctrine for one canonical ConfigHub tutorial, short Helm/AICR/OCI/YAML entry paths, distinct local/anonymous/managed choices, and STE limits for the Try page.
free-path-pitch.mdThe free-path value proposition: lead with a "look before you install" pre-flight check, keep parity as a quiet assurance, and separate ship-now copy from future security and AI builds.
server-account-pitch.mdThe account value proposition: your versions, your apps, your releases, plus the shipped-versus-frontier wording that ties the homepage variant and Apps boxes together.
house-layout.mdThe house layout (companion to house-voice): seven principles drawn from hall.kvick.dev - narrow prose, hairline sections, terminal-card commands, two-column hero, light card grid, one accent, restraint - plus the terminal-card pattern and where to apply it.
agent-experience-audit.mdAudit for the agent/operator docs: website pages stay for users, agent instructions live in repo Markdown, and catalog changes are out of scope for the first pass.
agent-experience-worklog.mdRunning log of observed agent friction and the fixes made.
get-started-rewrite-brief.mdBrief for Codex to rewrite the generated Get Started page (try.html) install-first around parity of outcomes: the full new copy, the recipe-vs-package explanation of --pull, verified commands, and generator notes.
persona-ux-rerun-2026-06-22.mdRerun of the public-site persona UX audit after the adoption-lens site updates, including page metrics, eight persona findings, and ranked follow-up work.
persona-ux-audit-2026-06-22.mdTen-persona public-site audit after the homepage and guide restructuring, with verified findings and the next ranked UX fixes.
fuzz-corpus-tests-roadmap.mdRoadmap for the non-website work (corpus, tests, fuzz, migration-UX): the test-our-tool principle, the persona taxonomy (F/G/cub-fuzz/Helm-migrant), the offline-first PR sequence, and the deferred live work.
helm-vs-cub-adoption-audit.mdAdoption audit for places where cub is worse than or more confusing than plain Helm on the common journey, with each gap marked solved, managed, partial, or unmanaged.
per-chart-fact-sheet-spec.mdDesign spec for the per-chart website page: section list, the exact data source per field, and a solid/partial/needs-more-testing status for each, plus the "level of support vs evidence depth" rule.
archive/README.mdHistorical. Planning snapshots retained for traceability.

Naming Rules

  • User entry points should have plain names: README.md, CATALOG.md, how-the-harness-works.md.
  • Keep the primary user docs small. Generated proof files are allowed to be numerous because they are evidence, not a required reading path.
  • Dated files are historical handoff snapshots. New standing plans should use undated names.
  • A doc a newer page or plan has replaced keeps its place and its URL. Mark it Historical. at the start of its description here instead of moving it.
  • Generated files should stay near the artifacts they summarize.
  • Recipe-level docs belong under recipes/.
  • Executable package docs belong under packages/.
  • Proof summaries and matrix outputs belong under data/.
  • Product guidance and implementation plans belong under docs/.

Generated from the committed markdown file docs/README.md. The source file is the authoritative version.