Leading the landing with what Systemix concretely is — an AI-native design system built on the learning loop (rationale in MDX, skills that update around your workflow, lives in Claude) — plus a named consultancy ladder (DS build sprint, design partner) converts the founder better than the abstract loop-memory pitch, which is a burden to understand.

No evidence ever pulled — the hero copy was extended/succeeded by landing-rebrand-hifi-2026-07 on the same seam before this bet could be measured.

Why this hypothesis

v7 made the abstract loop the pitch and demoted the design system to an optional substrate — and the landing became a burden to understand. The founder's read: the design system is not the substrate, it is the flagship use case of the loop, and the consultancy upsell. There are many AI-slopped design systems; Systemix is the AI-native one — MDX files as the decision layer, a learning loop for the design rationale behind every component, custom skills that update around the user's own workflow, living in Claude. The loop serves both verticals (design system + experiments); the hero should lead with the concrete one people recognize.

Given

A pre-PMF founder (or anyone who owns a design system) lands on Systemix from a feed and gives the hero five seconds. Today they meet "Your shipping memory" — an abstraction they must decode before they know what Systemix is for.

The AI workflow

The repositioning bet → Hermes frames the hypothesis (ICP · JTBD · metric) → write the variant copy (hero + the services ladder) → ship at 100% through the existing landing-hero variant seam → measure hero_cta_click per visitor, before vs after → a human decides → the learning is written back. Ship-and-compare, not a split: at ~6 visitors/30 days a real A/B cannot resolve, so the honest read is directional.

Measurement notes

  • hero_cta_click previously fired only on the GitHub link; the ship extends it to every hero CTA (cta: github | install). The before-window baseline for the install CTA is approximated by the pre-existing install_command_copied event.
  • book_a_call stays experiment #1's site-wide signal; the new design-partner CTA fires it with location: services-design-partner, so the ladder's effect is visible without contaminating #1's metric.
  • The full-page body re-anchor (problem/effect/loop/CTA/metadata repositioned to the design-system framing) ships in the same PR and evidence window — the variant is the repositioned page as a whole; the hero metric still owns the read.

Conclusion

The design-system-first hero engages a meaningfully higher share of visitors than the loop-abstraction hero, and the consultancy ladder produces its first design-partner inquiry.

Decision criteria

  • promote: hero CTA clicks per visitor rise meaningfully over the prior window → the design-system framing becomes the landing's spine (extend it to Problem/Effect/Loop in a follow-up pass).
  • iterate: engagement flat but design-partner inquiries arrive → the offer works, the hero words don't; tighten the H1/body, keep the ladder.
  • kill: no lift and no inquiries → the category framing isn't the hook; revisit which vertical leads (experiments-first?) or the ICP.

Changelog

  • 2026-07-06 — Pre-launch pivot extends this bet (not a new experiment): variant_b's hero grafts a "+ growth engine · wire the signals as a service" emphasis onto the anti-slop spine; /for/<persona> pages, a ValueProps + SignalsWall pass, and the /contract/memory feed layer on. Same hero metric (hero_cta_click), same seam — evidence read after this date reflects the extended page under the same bet.