Growth runs as one loop. The engine finds growth (ideas generated, then prioritized). To measure it, one funnel and real unit economics come from a single source of truth. Retention compounds it, turning each won customer into fuel for the next. A ranked idea is a bet, and the funnel proves whether the bet paid; retention then decides whether the payoff compounds or leaks. A metric a founder cannot trust breaks the loop, so spend the judgment once on definitions and let every later read be mechanical.

Ideas are cheap, and the founder's time to build and run them is the scarce resource, so prioritization is the point. A funnel a founder cannot reason about is worse than none, because a false number drives a false decision. Churn works differently: a cancellation is the lagging echo of a decision the customer made weeks earlier, so the leading signal, well ahead of any discount, is where the churn work pays. Three references carry the depth; read the one a step names before running that step:

references/idea-generation.md: generation frameworks, divergent-then-convergent discipline, ICE and RICE, honest scoring.
references/funnel-metrics.md: stage definitions, metric formulas, one source of truth, handoff SLAs, the LTV:CAC and payback math.
references/retention.md: leading churn signals, health segmentation, intervention playbooks, dunning, GRR/NRR/cohort metrics.

Steps

Frame the thesis, the goal, and the constraint. State the growth thesis in one sentence (the single mechanism this business bets on to acquire and keep customers), the one metric that mechanism moves, the audience, and the real budget or time limit. This step is done when thesis, metric, audience, and constraint each sit on one written line.
Diverge, then converge to a ranked few. Run several generation frameworks as separate prompts (the channel list / Bullseye, growth loops vs funnels, jobs-to-be-done, competitor teardown, 10x not 10%) deferring all scoring, then cluster the raw list and cut the off-thesis ideas before scoring the survivors on one model from references/idea-generation.md. This step is done when 15 or more raw ideas collapse to 6–8 scored bets, each carrying its score on the one named model (ICE by default; RICE when reach data exists), a growth-mode tag (loop or funnel), and a one-line reason.
Pick the top bets and attach a success metric to each. Commit to the highest-ranked bets the constraint allows and name the deferred tail as not-now, so the tail stays visible. Give every pick one quantified metric and a numeric threshold with a time window. This step is done when the picks fit the stated constraint and each pick names one metric and one threshold.
Define the funnel and instrument it from one source of truth. Write one binary entry condition and one owner per stage across the eight named stages (lead, MQL, SQL, opportunity, closed-won, closed-lost, expansion, churn) from references/funnel-metrics.md, with the CRM designated the system of record. Then write each metric's formula and the exact fields it reads beside the dashboard: CAC, LTV, LTV:CAC, payback, per-stage conversion, win rate, pipeline coverage, and NRR. This step is done when two people pulling the same metric on the same day get the same value.
Set targets, then find the bottleneck. Attach a target to each metric (LTV:CAC at or above 3:1, payback under 18 months, pipeline coverage at 3×–4× of the period goal, a conversion target per adjacent stage pair), then rank the gaps against target. This step is done when one stage is named the bottleneck, its conversion gap is quantified, and one owned action is recorded against it.