UNOFFICIAL/EXPERIMENTAL. Structured re-run of the persona UX strategy, scoped to the 8 guide pages reachable from the top menu (Upgrade→/private/ excluded as commercial, per the strategy's scope rule). 480 graded tests = 8 pages × 10 personas × 6 checks, each scored P=1 / p=0.5 / F=0 by a per-page audit that read the page's actual content. Colored matrix: matrix.html · raw data: tests.csv.
Headline: 52% overall - the guides orient but don't deliver
The clearest signal is per-check. The guides are strong at getting you oriented and weak at letting you act:
| Check | Score |
|---|---|
| C1 CoreQ - answers this persona's core need | 76% |
| C2 Action - runnable command/action on the page | 33% |
| C3 Output - expected output/result shown | 16% |
| C4 Prereq - prerequisites clear (account/cluster/tools) | 31% |
| C5 Next - next step clear | 71% |
| C6 Scan - findable fast (not buried/jargon-walled) | 87% |
Orient (scan 87%, core-question 76%, next-step 71%) ≫ deliver (command 33%, prerequisites 31%, expected-output 16%). A visitor can find the right guide and knows where to go next, but the guide rarely shows a runnable command, its prerequisites, or what they'll see - they have to leave for a doc to actually do anything. Expected-output (16%) is the single weakest dimension of the whole site.
By guide page (best → worst)
| Guide | Score | /5 |
|---|---|---|
| Get Started | 68% | 3.4/5 |
| FAQ | 65% | 3.3/5 |
| Docs | 60% | 3.0/5 |
| Ops | 50% | 2.5/5 |
| Home | 48% | 2.4/5 |
| Helm Catalog | 43% | 2.2/5 |
| Variants | 43% | 2.1/5 |
| Apps | 42% | 2.1/5 |
Get Started is the model page (real commands + "you should see" blocks). Apps, Variants, Helm Catalog are orientation pages with almost no runnable action. Docs/FAQ route well but answer in prose+links.
By persona (best → worst served)
| Persona | Score |
|---|---|
| Skeptic | 70% |
| LowHelm | 61% |
| HelmExpert | 61% |
| Novice | 57% |
| Maintainer | 57% |
| GitOps | 54% |
| ExistingApp | 50% |
| SRE | 43% |
| Security | 38% |
| AILead | 31% |
The site is built for the Skeptic (70% - gaps, refusals, and proof are first-class) and underserves the doers: the AI-curious lead (31%) is nearly invisible across the guides (no AI content on Get Started, Apps, or Ops), and Security (38%) and SRE (43%) have no menu guide that lets them act on their core task.
Worst-scoring cells
| Guide × persona | Score | Gap |
|---|---|---|
| Get Started × AILead | 0.5/6 | page never mentions AI/agent at all |
| Helm Catalog × Security | 0.5/6 | no secrets/OCI provenance content at all |
| Home × ExistingApp | 1/6 | no read-only adopt-running-stack path |
| Home × Security | 1/6 | secrets/OCI provenance barely mentioned |
| Variants × Security | 1/6 | no secrets, no OCI provenance |
| Apps × AILead | 1/6 | no AI-change or safety story at all |
| Ops × AILead | 1/6 | no AI-change or safety surface at all |
| Home × AILead | 1.5/6 | 'AI with guardrails' prose only, dead-end card |
Ranked fixes (the output)
- [technical voice] Add an expected-output block to every command on every guide - the weakest dimension site-wide (C3 16%). "You should see …" under each
cub/npm/kubectlline. - [technical voice] Put a runnable command on the orientation guides (Apps, Variants, Catalog) - at least one copy-paste action + result, instead of prose that routes to a doc (C2 33%).
- [technical voice] State prerequisites inline per guide - account / cluster /
cubinstall / inputs, near the first command (C4 31%; only Get Started + FAQ do this). - [marketing voice] Give the AI lead, Security, and Existing-app personas a home - AI content on Home/Get Started (not just a dead-end card), a security/provenance section, and a read-only "adopt my stack" command on Apps.
- [marketing voice] Complete the Ops guide - it names rollback/observe/patch but ships none; add those cards so SRE can act.
Re-run after Codex lands changes and diff the matrix.