Concepts

Five record types and one rule of thumb: score what you saw, not what you hope.

  1. Signal
    1. Evidence rubric
  2. Cluster
  3. Idea
    1. Kill criteria
    2. WIP limit
  4. Experiment
  5. Interview
  6. Founder files
  7. Ranking

Signal

One observed pain, with its source: who does what painful thing, how often, what they do about it today and whether they already pay for it. A signal lives in signals/YYYY-MM-DD-<slug>.md.

  • ref is a URL that was opened or seen in a search result, an interviews/ path, or an archive/ path for an older idea list. Never a constructed or guessed URL.
  • pain is one paraphrased sentence of at most 200 characters. The body holds at most one verbatim quote of 15 words or fewer, with its link.
  • Businesses and public roles only: no names, emails or profiles of private individuals.
  • Status: new → clustered → promoted (it fed an idea) or dismissed.

Evidence rubric

Score Meaning
1 One anonymous complaint, or a legacy idea with no observed demand
2 Several independent complaints from the same kind of source
3 Complaints plus a paid workaround: a job post, a freelancer gig, a badly reviewed tool people still pay for
4 A direct conversation confirming the pain and its cost in hours or money
5 A stated price they would pay, or proof they already pay for a worse alternative

When unsure between two scores, take the lower one. Online scouting alone rarely gets past 3; scores 4 and 5 come from conversations and pre-sales.

Cluster

A short slug shared by signals that describe the same pain in the same kind of segment. Clustering is a judgment, written into each signal’s cluster field; the /scout-online digest suggests clusters, and proposes shaping one into an idea once it holds three or more signals with evidence 3 or more.

Idea

A sellable offer built on observed signals, in ideas/<slug>/idea.md. Other files in the folder are free-form research notes.

  • At least one signal. No signal, no idea. When an idea is written, its signals become promoted.
  • Two versions. offer is the service you can sell by hand this month; product_path is the software it becomes. shape is service, product or hybrid; service and hybrid need a product_path.
  • Five scores, 1 to 5: score_pain, score_reach (judged against founder/channels.md), score_willingness to pay, score_speed to the first paying customer, score_fit (judged against founder/profile.md). score_total must equal their sum.
  • Status: shaped → validating → go, or parked, killed, building-elsewhere. Closed ideas keep their file and say why.

Kill criteria

An idea in validating needs kill_criteria: a list of conditions, each carrying a date (by convention at the start), under which the idea dies. For example: 2026-11-07 fewer than 3 of 20 firms reply. Writing them before the experiment starts keeps the decision honest when the results come in.

WIP limit

At most three ideas are validating at once. A fourth is an error, not a warning.

Experiment

The cheapest test of an idea’s riskiest assumption, in experiments/exp-NNNN-<slug>.md.

  • method: outreach, interviews, landing-page, pre-sale, concierge, fake-door or other.
  • hypothesis and a success_metric (numeric where possible: “3 of 10 confirm a paid workaround”).
  • budget_eur at most 100 and a deadline at most 14 days after created. Breaking either is a warning, so it is allowed as a conscious choice, written in the body.
  • The body ends with ## Log: one dated line per contact or result, newest last, written by /log-experiment.
  • done requires result (with numbers) and decision: continue, pivot or kill.

Interview

A Mom Test note in interviews/YYYY-MM-DD-<slug>.md: past behaviour and money spent, not opinions about a pitch. It records the channel (in-person, call, chat, email) and whether the person agreed to notes being kept, holds a first name and role at most, and can promote new signals. A direct conversation that confirms the pain and its cost is what earns evidence 4.

Founder files

founder/profile.md, constraints.md and channels.md describe what you can build, what you will not do (the vetoes) and where you can reach people. They filter ideas and inform three scores; they are never a source of ideas. An idea that breaks a veto is killed with the veto named.

Ranking

Ideas are compared by speed to the first paying customer (score_speed, first_customer_days), then by total score. That comparison is yours to make; INDEX.md lists ideas by status, then by score_total. A solo founder runs out of time before running out of market.