The experiment
An experiment is the unit of work in Systemix. It's a single MDX file in experiments/ that starts as an open hypothesis and ends as a written decision with evidence.
The shape
---
type: experiment
id: landing-live-loop-2026-06
section: landing
status: open # open | running | decided
# Who this is for and why
icp: daily-shipping founder
jtbd: convert a stranger into a consultancy conversation, earlier than a feature pitch
# The bet
hypothesis: seeing a live, dogfooded loop converts the founder to a consultancy call
given: |
You land on Systemix. You see one real experiment — the product's own landing bet —
rendered as a step graph with a live verdict. You understand the loop from watching
it run on itself.
conclusion: book-a-call rate lifts within 30 days
# The AI workflow (optional — renders as a step graph on the experiment page)
workflow:
steps:
- id: start
kind: input
label: Founder lands on /
- id: loop
kind: agent
label: Sees the live experiment
- id: cta
kind: output
label: Books a call
edges:
- from: start
to: loop
- from: loop
to: cta
# Measurement
metric: book-a-call-rate
posthog-event: book_a_call
# Evidence (filled by /measure)
evidence-posthog: ~
evidence-social: ~
# Decision (filled by /close-experiment)
result: ~
decision: ~
confidence: ~
---
Rationale and notes go here in markdown.Lifecycle
| Status | What it means |
|---|---|
open | Authored, not yet running. Variants may still be in progress. |
running | Live. PostHog is tracking the posthog-event. |
decided | Closed. Result, decision, and confidence are written. Learning is in LEARNINGS.md. |
Key fields
icp — the specific person this bet targets. Not "users" — a role, a context, a moment.
hypothesis — the assertion you're testing. One sentence.
given — the prompt or setup that puts the ICP in the situation. What do they see? What do they do?
conclusion — the win state. What would have to be true for this to be "decided: ship"?
workflow — optional. Defines the AI workflow as steps and edges. When present, the experiment page renders it as a step graph.
metric / posthog-event — what you're measuring and the event name in PostHog.
evidence-posthog — filled by /measure from PostHog data.
evidence-social — qualitative evidence (quotes, interviews, replays) logged manually.
result / decision / confidence — filled by /close-experiment. The decision is written to LEARNINGS.md.