receipts

The rest of the pack makes guesses: a board of model opinions, a panel of model buyers, numbers from estimates. That is useful research. It is not proof, and without this skill it starts to look like proof. This skill keeps the line between the two, and it is the only place a guess can become a fact: when a real person does something.

What the user typed after /receipts: $ARGUMENTS

Read ${CLAUDE_SKILL_DIR}/rules.md first, every time. Every other Shark Tank Proof skill follows it too.

Tool in this folder (if the path is not expanded, use the base directory Claude Code printed for this skill):

python3 "${CLAUDE_SKILL_DIR}/ledger.py" <init | add | check | summary | gate | scan>

Below, LEDGER means exactly that. State lives in founder/evidence/.

The folder

founder/
  START-HERE.md        the map (copy templates/START-HERE.md)
  CONTEXT.md           the stage map: which skill reads, writes and hands on what (made by init)
  me.md                who is behind it (copy templates/me.md)
  idea.md ...          what the other skills write
  work/                one file per project, pilot or client (copy templates/work-record.md)
  evidence/
    claims.csv         the ledger; claims.md and summary.json are written from it
    evidence-log.md    one row per original file
    people.md          real people only: who, what they said, may we name them
    calls/             notes from real conversations
    <originals>        screenshots, emails, receipts, recordings, signed pages
  opportunities/       one file per customer, funder or internal ask
  outputs/             every pitch, dated, written by /handshake
  handoff/             labelled results passed between stages: 04_litmus.json, 06_burn-test.json

Four jobs

Ask which one if it is not clear. Do one at a time.

1. Start or update the folder

• LEDGER init, then copy templates/START-HERE.md and templates/me.md into founder/ if missing. Say what you made.
• Interview the founder with ${CLAUDE_SKILL_DIR}/interview.md. One question at a time.
• Read any files they give you. File the originals in founder/evidence/, add one row each to evidence-log.md, then add one claim per fact with LEDGER add. Keep their words.
• For each project, pilot or client they describe, write one founder/work/<name>.md from templates/work-record.md: the problem before, who did what, where it broke, what is still unproven, and one suggested sentence for pitches. It lists ledger ids; the facts stay in the ledger.
• If other Shark Tank Proof skills already ran, sweep their outputs and log what they found with the right status: the panel's buy rate and board scores are SIMULATED; the CFO's margin and ramp are PLANNED; a supplier quote or a public price page is MEASURED with the link as its evidence.

2. Log real contact

Anything a real person did: a call, an email, a deposit, a signed pilot, a refusal. Follow "Logging the call" in ${CLAUDE_SKILL_DIR}/conversations.md. A yes is MEASURED, kind demand, with --commitment naming what they did (paid, deposit, signed, booked, intro) and the count as the number; the ledger refuses a buyer with no commitment, no count or no evidence. The evidence row must name the original that fits the commitment: a receipt, invoice, payment or order export for paid and deposit, a signed page, contract or LOI for signed, a booking or calendar invite for booked, the email or message for intro. A signup or a waitlist is not a commitment: log it as kind other. A buyer claim that says signups, waitlist, followers or free trial is refused even with a commitment. A no is logged too: MEASURED, kind demand, number 0 of 1, with the reason in their words. If the founder has not talked to anyone yet, give them that file and the five questions, and help them pick the first five people.

3. Audit

LEDGER check      # broken labels, missing evidence ids, stale claims
LEDGER summary    # writes founder/evidence/claims.md and summary.json

Then tell the user, in this order:

1. The hard truth: how much of the pack rests on real evidence. For example: "31 claims. 2 measured, 0 from real buyers. Nothing about demand is real yet."
2. Claims that are stated as fact somewhere in founder/ but are PLANNED or SIMULATED in the ledger (read board.md, offer.md, pitch.md, marketing.md, one-pager.md). Name the file and the line, and the honest wording.
3. The three strongest real claims.
4. The three biggest gaps.

4. Find the gaps for one receiver

For a chosen customer, funder or internal ask (founder/opportunities/), compare what they need to see (${CLAUDE_SKILL_DIR}/../handshake/receivers.md) with what LEDGER gate allows. List each missing piece as something to go and get: a measured number, a quote with permission, a pilot, a partner, a registration. For each, the cheapest way to get it this week and who could help.

When you list the gaps (job 3 or 4), put first the gap that blocks the first sale: where the customers are, what they buy first, at what price, or what words make them buy (Ready, Fire, Aim, rule 11 in rules.md).

Rules

• Never upgrade a status without a new original in evidence/. A claim moves from PLANNED to MEASURED only when the thing was counted, paid or signed.
• Never log a simulated buyer, a board member or your own reasoning as THEY SAID IT. Only real people say things.
• Say who did what. A workshop is not a contract; a mention is not a client.
• PRIVATE stays in the folder. Ask before you log anything about a third party as PUBLIC.
• Date every claim. Check anything older than 90 days again before it goes in a pitch.

Output

founder/evidence/ (ledger, log, people, calls, claims.md, summary.json), founder/START-HERE.md, founder/me.md. End with: what you changed, the hard truth in one line, the gaps, and the one real-world action this week that would add the most MEASURED evidence.