founder-os-earlyvangelist-qualifier

FounderOS earlyvangelist qualifier. Checks one interviewee against Steve Blank's five earlyvangelist criteria (has the problem, knows it, is looking for a fix, has built a workaround, has or can get b

SKILL.md

GitHub

Earlyvangelist qualifier

Earlyvangelist is Steve Blank's term, from The Four Steps to the Epiphany and The Startup Owner's Manual. Credit him and point the founder there. The five criteria roughly build on each other: the person has the problem, knows they have it, is actively looking for a fix, has already cobbled one together, and has or can get the money to pay for a real one.

Laya screens. It never decides. The stage 3 gate wants every earlyvangelist backed by interview evidence, checked by Claude and then Sean.

When to run

  • On one saved interview (an interviews row), after its Mom Test audit. Criteria met in a conversation full of hypotheticals mean little.
  • Run it for every interview in the founder's target segment, not only the ones the founder liked.

Laya pass

python3 "$FOUNDER_OS_DIR/scripts/audit.py" earlyvangelist \
  --founder-id <founder_id> --interview-id <interview_id> --transcript \
  --founder "<founder's speaker label>" --lang <en|pt|...> --record

--transcript sends only the interviewee's words; the founder's pitch describes the problem too and would score as if the interviewee had said it. Drop it for prose notes, and warn the founder that notes in their own words make this weaker.

The output has criteria (per criterion value, p, trust, and on long text an excerpt of the passage that carried the answer), met, and escalate, which is true at escalate_at (4) or more.

Claude pass

For each criterion, find the evidence in the interview: a quote and the turn it came from. A criterion counts only with a past-tense, specific quote behind it, whatever Laya said. "I'd pay for that" is not budget; "I pay a bookkeeper 300 a month to chase these" is.

Then:

  • 4 or 5 confirmed: escalate. Tell the founder this looks like an earlyvangelist, list the quote per criterion, and say that Sean signs off before it counts at the gate. Suggest the next ask that would prove it: a paid pilot, a pre-order or a deposit.
  • 3 or fewer: not an earlyvangelist yet. Name the missing criteria and the question for a follow-up that would settle each one.
  • Where you and Laya disagree, say so. That disagreement is the most useful label in the set.

Without Laya

If mode is claude-only, criteria_to_check lists the five questions. Answer each from the interview yourself, with a quote and turn, and apply the same four-or-more rule. Say in one line that Laya was unavailable. Write no labels.

Label capture

Show the founder the five answers and ask which are wrong. Record every one they accept or reject:

python3 "$FOUNDER_OS_DIR/scripts/labels.py" accept --founder-id <id> --by <founder_id> <label_id> ...
python3 "$FOUNDER_OS_DIR/scripts/labels.py" reject --founder-id <id> --by <founder_id> <label_id>

When Sean reviews an escalation, record his calls with --by sean. Do not accept tags nobody looked at. The cross-founder export is operator-only.

Ledger writes

  • Labels, one per criterion, from audit.py --record; corrections from labels.py.
  • One audits row on the interview: record_audit.py --table interviews --id <interview_id> --auditor claude --check earlyvangelist --verdict flag|fail --findings "<criteria met, with quotes and turns>". flag means escalated to Sean; fail means fewer than four.
  • After Sean confirms, his own audits row with --auditor sean --verdict pass. Sean writes it himself: --auditor sean is operator-only (FOUNDER_OS_OPERATOR=1), and no skill sets that.
  • Never set interviews.earlyvangelist from this skill. The ledger has no call to change an interview after it is written, and the gate reads the audits.