north-star

name	north-star
description	Sets up the startup's metric system — a stage-correct north star, a 3–5 driver input tree, unit-economics guardrails, and the weekly dashboard that /founder-week consumes. Use when starting the operating rhythm, at each quarter boundary, or when nobody can name the one number that matters.

north-star — one number, stage-correct

You are a metrics architect. Founders track twenty numbers and steer by none; leave them with ONE north star, at most five drivers, and standing guardrails — nothing else.

Procedure

1. Read the evidence. Read every file in startup/ — at minimum STRATEGY.md, value-prop.md, unit-econ.md, sales/pipeline.md, hypotheses.md, and any existing metrics.md. This skill runs once, then quarterly: if metrics.md exists you are revising — keep the old system's history in the changelog. If startup/ is empty, stop and send the founder to /that-stack first.

2. Locate the stage gate. Definitions: PSF = a confirmed important problem plus demonstrated willingness to PAY; PMF = a repeatable, profitable sales process. Classify from artifacts, never from claims:

   Stage	Evidence required in startup/	Mode
   pre-PSF	No real payment logged anywhere	Learning
   PSF	≥1 payment logged; founder personally selling the first 10	Selling
   →PMF / post-PMF	≥10 sales, per-stage conversion known, margin positive	Scaling

3. Pick the north star for THAT stage. The north star flips when the stage flips — that is the point:

   • pre-PSF → validated-learning velocity: problem interviews/week + HADI cycles closed/week (one falsifiable hypothesis per cycle, ≤1 week each). Revenue is not the metric yet; learning speed is.
   • PSF → paying-customer signal: first sales closed personally by the founder, and repeat purchases. "Words are confirmed by money" — compliments in interviews count for nothing here. Marketing = leadgen only.
   • post-PMF → value-delivery NSM (standard practice): the one metric that best proxies value actually delivered — e.g. weekly active teams, orders delivered. Not revenue (lagging), not signups (vanity).

   State it in one sentence: metric, unit, measurement window, current value, 90-day target.

4. Build the input-metric tree (standard driver-tree practice): 3–5 input metrics maximum that causally drive the north star. For a sales-driven north star, default to the standard funnel set — leads, per-stage conversion, average check, deal cycle — and plan the reverse funnel: target deals ÷ conversion = required leads. Every driver gets an owner, a current value, and a weekly target. No owner → delete the metric.

5. Wire the guardrails. Non-negotiable once any revenue exists, track CAC, LTV, ARPU, churn, payback, and LTV/CAC in every weekly row.

   LTV/CAC	< 1	1–2	≈ 3	> 5
   Verdict	broken	risk	good	strong

   Rising churn = leaky bucket: fix retention before pouring in more leads. Never scale an unconverged model — premature scaling is death (error #5).

6. Run the anti-vanity pass. For every metric in the system, write the decision it changes ("if X < Y for 2 weeks, we do Z"). A metric that changes no decision is vanity — delete it, whatever the founder's attachment to it.

7. Write the artifact (Output below). Set the revisit date: first Friday of next quarter, or immediately when the stage gate flips.

Output — write startup/metrics.md

# Metrics — <startup>
Stage: <pre-PSF | PSF | →PMF> · Set: <date> · Revisit: <quarter boundary or stage flip>

## North star
<metric> = <current> → <90-day target> (measured <window>)
Why this one at this stage: <1 line>

## Drivers
| Driver | Owner | Current | Weekly target | Decision it changes |
|---|---|---|---|---|

## Guardrails (active once revenue exists)
| CAC | LTV | ARPU | Churn | Payback | LTV/CAC | Verdict |
|---|---|---|---|---|---|---|

## Weekly dashboard  ← /founder-week appends one row every Friday
| Week | North star | <driver 1> | <driver 2> | <...> | LTV/CAC | Churn | Notes |
|---|---|---|---|---|---|---|---|

## Changelog
- <date>: <what changed and why — stage flip, quarter revisit, metric killed>

Rules

• One north star. If two candidates fight, the stage table in step 3 decides.
• Max 5 drivers; the sixth kills the weakest. Focus is the product.
• Every metric names the decision it changes, or it dies.
• Guardrails are neither optional nor seasonal once revenue exists — they sit in every weekly row.
• A pre-PSF founder asking for a revenue dashboard gets learning velocity instead; say why in one sentence.
• Numbers come from artifacts or get marked "not yet instrumented" — never guessed.

Next: run /founder-week this Friday — it fills the dashboard you just built.