Validate

Research says what's likely. A week of real sign-ups, replies and pre-sales says what's true. validate prepares that week, and once the founder has numbers it re-scores the idea with them. It never publishes, sends, buys or signs up for anything: the founder runs the test.

${CLAUDE_PLUGIN_ROOT} is the plugin folder. Safety rules §0, source rules §3, the rubric and the fixed rules §5 of ${CLAUDE_PLUGIN_ROOT}/skills/kill/SKILL.md apply.

Kit mode (default)

1. Load the idea: ideas/*-<slug>/ with verdict.json, report-data.json (its test_plan, personas, customer voice and channels), and fix.json when it exists: then test the fixed idea (its price, its customer, its channel). No verdict → run /kill-my-idea:kill first.
2. One hypothesis, in one line: who pays how much for what. The test proves or kills it.
3. Targets (2–6), each with a metric, a goal with a number, by (a day), and moves: the sub-scores this evidence may change, decided now, before the results. At least one target is signups or presales: opinions alone don't validate anything. Use out_of for a rate ("8 replies out of 40 messages").
4. Landing page: name, headline (the outcome, not the feature), subhead, 3–5 bullets, the button, the price when the hypothesis has one, and one sourced proof if you have it. form_url stays empty unless the user gives the URL of their form tool (Tally, Formspree…).
5. Interview script: 5–12 questions about what they did the last time the problem came up, never "would you…" or "how much would you pay" (the script rejects those). Add the red flags that mean "no".
6. Outreach: 1–6 messages, each for a named place (a group, a forum, cold email…), with links_allowed from that place's rules. No link where links are banned.
7. Kill criteria with a number and a date.
8. Write validate.json (format below) and run python3 "${CLAUDE_PLUGIN_ROOT}/scripts/validate-kit.py" ideas/<folder>. Fix every error and rerun. It writes validate/landing.html (static, no script), validate/interview.md, validate/outreach.md, validate/tracker.csv, and the "7-day test kit" section of the report.
9. Answer in chat: the hypothesis, the targets, where the files are, and the first thing to do today. End with: "When you have the numbers: /kill-my-idea:validate --results."

Results mode (--results)

1. Get the real numbers from the user (or from validate/tracker.csv if they filled it in): one result per target, with how it was measured. Never invent or round up a result; a target with no result is 0.
2. A target is met when the result reaches the goal (as a rate when it has out_of, with at least as many tries as planned).
3. Moves, only to sub-scores listed in the target's moves: up on a met target, down on a missed one, 2 points at most per sub-score, from = the verdict's score. A met target may clear a kill shot it disproves (clears), for example no_distribution when cold outreach got replies. A missed signups or presales target should pull something down.
4. Re-score like fix does: copy the 21 sub-scores, apply the moves, set the kill shots, apply the fixed rules again, compute score_raw, score, verdict into results.rescored.
5. Run the script again. Answer: researched → validated, which targets were met, and the decision the kill criteria call for (build, change, or archive).

verdict.json never changes. The validated score sits next to it in the report.

validate.json

{
  "schema_version": "1.0", "slug": "<verdict.slug>", "generated": "YYYY-MM-DD",
  "based_on": {"verdict": "<copied>", "score": 0},
  "hypothesis": "<who pays how much for what>",
  "landing": {"name": "", "headline": "", "subhead": "", "bullets": ["", "", ""], "cta": "",
              "price": {"amount": 0, "currency": "EUR", "period": "month"},
              "proof": {"text": "", "source": "https://"}, "form_url": ""},
  "interview": {"who": "", "questions": ["", "", "", "", ""], "red_flags": [""]},
  "outreach": [{"channel": "", "where": "", "links_allowed": false, "text": ""}],
  "targets": [{"id": "T1", "metric": "signups | replies | interviews_pain | presales | visits | conversion | other",
               "label": "", "goal": 0, "out_of": null, "by": "Day 7", "moves": ["willingness_to_pay"]}],
  "kill_criteria": ["<with a number and a date>"],
  "results": null
}

results, once there are numbers:

{"date": "YYYY-MM-DD",
 "entries": [{"target": "T1", "actual": 0, "out_of": null, "evidence": "<how it was measured>"}],
 "moves": [{"sub": "", "from": 0, "to": 0, "target": "T1", "why": ""}],
 "clears": [{"kill_shot": "", "target": "T2"}],
 "rescored": {"criteria_detail": {}, "kill_shots": {}, "rules_applied": [], "score": 0, "score_raw": 0.0, "verdict": ""}}

Plain text in every string.