Depth-probe strategy - adversarial, multi-persona, tier-aware

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

New to cub? Install the cub CLI first. You can pull and render public catalog packages without an account. Commands that save or change ConfigHub data require you to sign in.

Generated at: 2026-07-30T12:38:02.000Z UTC · source: committed helm-expt evidence for this rendered repository document.

_The adversarial-persona UX-test layer of the helm-expt test map._

Status: DRAFT (2026-06-01). Proves how users actually use the catalog in ConfigHub (not just that it installs) before broadening 20→100. Output = a usage-readiness scorecard that becomes the admission check for the 100.

Core move: persona-drivers run each archetype the way a real user would, at a given tier; multi-lens adversaries try to refute that it worked and that it beat the tier-appropriate alternative. A cell is PROVEN only if it survives every lens.

Tiers (must work free AND paid)

TierWhat it isUnlocksThe "better than…" bar
T0 - local/freeno/minimal account; render + local kind install; secret-free or out-of-band secretinstall · Helm-equivalence · default forkbetter than plain helm install locally
T1 - connected (free account)authenticated ConfigHub Spaces/Units/OCI/Argoderived variants · env promotion · gates/approvals · revisions/rollback · parameterized fillsbetter than Helm + manual review/GitOps
T2 - paid/managed/enterpriseConfigHub Server, managed overlays, ESO+Vault+rotation, fleet/promotion-wavesmanaged/customer overlays · paid secret rungs · fleet opsthe paid delta is worth it over T1 + DIY

The probe runs the tiers available in the environment (args.tiers, default ["T0","T1"]). T2 cells without a paid env are marked REQUIRES-PAID, not faked. The free path must be a complete, honest experience - never a teaser that breaks at a paywall (that's what adversary A4 hunts).

Matrix: archetypes × usage-dimensions × tiers

Archetyperole
nginxstateless web
postgresqlsecret-needing DB
cert-manager / external-secretsCRD operator
vault or lokiHA stateful
Usage dimensiontiers
install + Helm-equivalence (the floor)T0
derived variant → env promotion + gates/approvalsT1, T2
parameterized base + placeholder fillsT1
day-2 edit (image/replicas) → rollbackT1
governed secret fork (existing-secret + required-secret gate)T1; ESO+Vault rotation = T2

Persona drivers (Sonnet, rig-bound - run it as that user, at their tier)

Drivers discover the exact cub commands (via --help / the cub-mutate/cub-apply/create-initiative skills) - they must NOT invent flags.

Adversaries (Opus, read-only / no-rig - refute the claim)

Cell protocol & verdicts

Persona drives live (at tier) → emits receipts → A1–A4 attack independently → verdict: PROVEN (survives all) · FINDING (≥1 substantive refute) · WORSE-THAN-ALTERNATIVE (A2) · PAYWALL-SURPRISE (A4) · REQUIRES-PAID (T2 cell, no paid env).

Scorecard → admission check for the 100

Output: archetype × usage-dim × tier → verdict, plus a per-tier thesis verdict ("is ConfigHub better than the alternative at T0 / T1 / T2?"). Usage patterns that come back PROVEN at a tier become required cells in the 100-chart admission check at that tier - so we broaden only on usage that's proven and tier-honest, not just installable.

Capacity & free-tier execution

Runnable Shape

Use an agent workflow or equivalent runner to fan out by tier: drive the scenario, run the adversarial checks, then synthesize the scorecard. The runner should accept inputs equivalent to:

tiers: ["T0", "T1"]
archetypes: [...]