Skip to main content
The plugin’s core discipline is simple: decisions trace back to evidence, and evidence traces back to actual customer words. The mechanism behind it is equally simple — Obsidian-style [[slug]] links between markdown files.

​
Hypotheses are the hub

Every assumption your idea depends on becomes a file in startup/hypotheses/ — a testable statement with a status (untested, confirmed, invalidated, or archived) and a type tag:
TagWhat it claimsHow it’s best tested
#problemThe pain is realCustomer conversations
#solutionYour approach actually helpsA lightweight prototype or demo
#willingness_to_payPeople will pay for itA pricing gate — waitlist, pre-order, payment link
#urgencyIt hurts now, not “someday”Behavioral signals: what people already tried, spent, worked around
The tag isn’t decoration — it drives what kind of validation move the plugin suggests for each bet. As evidence accumulates, other artifacts point at hypotheses by slug:
  • Interview statements. Each analyzed interview yields statements like - "I track every invoice in a spreadsheet but forget to update it." #problem [[users-track-invoices-in-spreadsheets]] — a real quote, tagged, linked to the bets it bears on.
  • Interview script topics carry **Why it matters:** [[hypothesis-slug]] lines, so every conversation topic exists because of a specific assumption.
  • Survey questions each carry a Tests: [[hypothesis-slug]] line.
  • The MVP plan lists the hypotheses the experiment tests and what a positive result looks like for each.
A statement that doesn’t fit any hypothesis stays unlinked on purpose — it’s raw material. When the same unlinked theme shows up across multiple interviews, the plugin proposes it as a new hypothesis.

​
How evidence becomes a verdict

When you ask how your hypotheses are doing (or right after an interview is analyzed), an independent assessment pass searches for every [[slug]] mention across your interview files and re-reads those statements in context — it doesn’t trust any single interview’s summary. It weighs evidence the way you’d want a skeptical co-founder to:
  • Counts distinct interviews, not repeated statements from one enthusiastic person
  • Discounts agreement that your own question framing produced
  • Requires cross-interview support before flipping a status
What comes back, per hypothesis: what changed since the last assessment, and a next action — the smallest observable next validation move, biased toward putting something in front of a real human rather than “do more research.” One cross-hypothesis Top pick names the single sharpest move right now. Status changes always need your explicit confirmation; the advisory next actions are written eagerly so your hypothesis files double as a live dashboard.
The ## Next Action section in each hypothesis file is machine-maintained and overwritten on every assessment — it always reflects the latest read, not a history. The history lives in your interviews and the plan’s log.

​
Browsing the graph

Because links are plain [[slug]] text in plain markdown, the graph is yours to explore with free tools:
  • Open startup/ as an Obsidian vault — hypotheses, interviews, and scripts appear as a connected graph, and frontmatter fields (status, tags, dates) are filterable with Dataview.
  • Or stay in the terminal: grep -rl "users-track-invoices" startup/interviews/ lists every interview bearing on that hypothesis.

hypotheses skill

Surfacing, tracking, and assessing your assumptions.

interviews skill

Where the evidence comes from.