agentsclimarketplace

Experiments

Skill ajstars1/startup-coach/experiments

Agent-based startup coaching skills for Claude Code — every claim labeled FACT/ASSUMPTION/UNKNOWN. Zero-hallucination protocol, 10 skills, one master orchestrator.

Install
npx -y skills add ajstars1/startup-coach --skill experiments

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Design and grade validation experiments — price tests, concierge/manual MVPs, landing/smoke tests, channel tests, pre-sales — as test cards with explicit hypotheses, metrics, and pass/fail bars. Use when an assumption can't be validated by interviews alone (especially willingness to pay — only refusable prices produce truth), when choosing an MVP, or when a founder reports experiment results.

SKILL.md

3.8 KB, as published. Nobody here has run it

Experiments — Test Cards & Evidence Grading

Credits: David J. Bland & Alexander Osterwalder — Testing Business Ideas (strategyzer.com); MVP/concierge practice — Eric Ries (The Lean Startup); hypothesis discipline — Steve Blank.

Why this skill exists: interviews produce E1 stories and E3 commitments, but some beliefs — above all willingness to pay — only yield truth at E4: a refusable price, accepted or declined. Surveys and "would you pay?" answers are E0–E1 regardless of how many you collect.

The Test Card (one per experiment — template in TEMPLATES.md)

  1. Belief under test — one ledger item (A⟨n⟩), phrased falsifiably.
  2. Test — the cheapest setup that produces a real behavior (not an opinion).
  3. Metric — what you count, defined before you run.
  4. Pass/fail bar — a number committed in advance (an educated guess beats no bar; moving the bar after results is self-deception).
  5. Evidence grade the test can produce — design for the grade you need: pricing needs E4-capable tests; channel questions need E2+.
  6. Cost & duration cap — cheap and fast beats rigorous and never-run.

The core test menu (pick the cheapest that can move the belief)

  • Concierge / manual service — deliver the service by hand, no software; real transactions (E4) at near-zero build cost. Usually the strongest first test for service businesses; instrument every delivery (time, distance, cost-to-serve, order value).
  • Price test — quote a real, named, refusable price to real prospects at the moment of intent. Track accept/negotiate/decline by level. Visible surcharges beat bundled costs — you learn nothing from a price nobody can refuse. (E4)
  • Pre-sale / deposit / LOI — money or signed intent before the thing exists. (E4/E3)
  • Smoke test / landing page / forwardable offer card — a real call-to-action into a channel; measure who acts, not who likes. (E2–E3)
  • Channel test — same offer, 2+ channels, counts kept per channel; instrument referral chains (who forwarded to whom). CAC only exists after this. (E2)
  • Wizard-of-Oz / prototype session — watch real users attempt the real task; count completion and drop-off step, not opinions. (E2)
  • Shadowing / observation — watch the job being done today; workarounds found in the field are E2 gold.

Grading results (when the founder reports back)

  1. Compare against the pre-committed bar — pass, fail, or invalid test (didn't run as designed; a broken pipe is not a demand verdict — check delivery before interpreting silence).
  2. Write the ledger delta: grade moved (e.g., A7: E1 → E4-fail at ₹150, E4-pass at ₹100), items killed, new unknowns.
  3. Watch for: bar moved after the fact · counting compliments as conversions · n=1 celebrated as a pattern · concierge costs ignored (a test that loses money on every order proved demand, not a business — route to /business-model-canvas unit economics).
  4. Persist to discovery/experiments/⟨date⟩-⟨name⟩.md.

Output

Test cards for the 1–3 riskiest ledger items (never a backlog of ten — the queue is the enemy of the loop), each with belief ID, setup, metric, pre-committed bar, target grade, cost cap — plus, for reported results, the graded ledger delta and the next cheapest test.

Fill-in templates: TEMPLATES.md in this skill's folder.

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.