The Founder Loop: how an audit becomes a 12-week build plan

The verdict is one frame of a thirteen-week film

You finished the audit. Your idea scored 76, Green zone, REFINE verdict. The dashboard told you the three weakest lenses and the two cleanest dealbreakers. You closed the tab.

That tab close is where most idea-validation tools fail. A score is a still frame. Founders operate in motion. By Friday your traction lens is two customer calls richer, your competition lens is one funding round older, and your business-model lens has a churn curve you didn't have on Monday. The 76 isn't wrong — it's just frozen.

The Founder Loop is what the score becomes once it has to walk.

Six surfaces, one cycle

The loop has six structured surfaces stacked above the finalized audit. Each surface is deterministic: every value can be derived from data already in the audit plus new evidence collected during the week. There is no "founder reflection" field. Vibes don't ship to production.

1. Kill criteria — the part most founders skip

Three to five quantitative shutdown triggers, committed before week one. Each criterion is a single sentence with three parts: the metric, the threshold, the deadline.

• "If demo-to-paid conversion is below 8% by week 6, kill."
• "If we can't close a single $5k ACV pilot by week 10, kill."
• "If average activation time stays above 22 minutes after three onboarding rewrites, kill."

The kill criterion is not a regret-management device. It is a pre-commitment to your future self that the project is allowed to die — written before you have a year of sunk cost arguing for resurrection. The Inite cohort data is unsubtle here: founders who pre-commit three or more kill criteria abandon dead ideas a median of 4.2 months earlier than founders who don't. That's a quarter of runway saved on each false start.

2. Weekly experiments — three hypotheses, run cleanly

Every Monday the loop proposes three experiments sampled from the audit's weakest lenses. The proposals carry a hypothesis, a success metric, and a maximum cost in days. You pick the three you'll actually run.

The point is not the experiments themselves. The point is that the experiment is named in advance with a falsifiable success condition. "Run a landing-page test" is not an experiment; "ship a landing page targeting commercial real-estate ops managers, measure CTR-to-demo over 7 days, success is >3% on ≥200 visits" is. The first formulation produces aesthetic conclusions ("the page didn't pop"). The second produces math.

3. Evidence log — every signal is a row

Every customer call, demo no-show, churn event, public competitor launch, fundraise announcement, regulatory filing — anything that should move the audit score — lands as a row in the evidence log. Each row has six fields: timestamp, category (problem-solution / traction / competition / etc), source (URL or attendee), sentiment (+1 / 0 / -1), magnitude (small / medium / large), notes.

By week 4 you have 15-25 rows. By week 12 you have 40-80. The next recompute is a math pass over those rows: it walks the log, attributes deltas to lenses, and produces a new composite score. Nothing in the recompute is a vibe — everything is signed math over recorded events.

4. Cohort mirror — the 47 audits like yours

The mirror queries the public-ledger ideas table for audits within 0.65 cosine similarity on the extraction vector. Typically 15-50 neighbours come back. For each neighbour the mirror shows: their stage, their verdict trajectory, the kill criteria they fired (or didn't), the channels they tested, the pivots they did, and where they are now.

This is not "competitive intelligence" — most of these are not in your market. The point is operator pattern recognition. If 23 of 47 audits like yours killed at week 6 over a specific dealbreaker, you want that signal weeks before you hit week 6. The cohort mirror is the loop's memory across founders.

5. Pivot detector — six weekly signals, deterministic firing

Pivots are bad when they are panic responses and good when they are data responses. The pivot detector watches six weekly signals and fires when any combination meets the trigger threshold:

• ICP drift — your last three closed customers don't fit the ICP you targeted.
• Channel collapse — your two leading acquisition channels lost ≥40% of their CTR over a 3-week trailing window.
• Pricing pressure — your average closed ACV trended down for 3 consecutive weeks.
• Kill-criterion proximity — you're within 25% of any pre-committed kill threshold.
• Cohort divergence — three or more cohort neighbours pivoted at this stage on the same lens.
• Founder velocity stall — weekly experiments shipped dropped below 1 for 2 consecutive weeks.

The fire doesn't mean "pivot." It means "the loop sees pivot-shaped data; surface this to the founder before the gut feel arrives." The distinction is the difference between a pre-data pivot, which is panic, and a post-data pivot, which is operating.

6. Channel playbook — twelve channels, ranked by your shape

The playbook scores 12 acquisition channels against four inputs: your stage (idea/mvp/seed/A+), your ICP (B2B / consumer / prosumer), your ACV band, and your observer profile (solo founder, operator with capital, etc). It returns a ranked list with a one-line rationale per channel: "Cold email — strong for $5k+ ACV B2B at seed; weak for consumer at idea stage."

The playbook also surfaces what cohort neighbours actually used and at what stage. If 31 of your 47 cohort audits tried cold outbound first and only 9 of them stuck with it past week 6, that's a stronger signal than any thought leader's tweet.

Why a loop, not a dashboard

A dashboard is a passive surface — the founder visits it, looks at the score, and leaves. A loop is an active surface — every week the loop arrives at the founder with three named experiments, six pivot signals, and a delta on yesterday's evidence log.

This architecture follows from a single constraint: founder attention is the scarcest input in any startup, and only a system that costs zero attention to maintain will compound across thirteen weeks. The Founder Loop is built so the operator can ignore the dashboard entirely and still operate against the score — the Monday digest arrives in the inbox; nobody has to remember to go fetch it.

The implementation is boring on purpose. Every surface is a stored view over the audit row plus the evidence log. The recompute is a function call. The pivot detector is six SQL queries with a threshold check. There is no machine learning in the loop. The machine learning is in the audit — the loop just disciplines the operator's twelve weeks of follow-through after the audit is done.

How to run it

The loop activates automatically when finalize_audit writes a verdict. The first three kill criteria are proposed from the dealbreakers pass — accept, edit, or replace them. The first three weekly experiments propose from the weakest lens. The pivot detector starts watching on week 1.

Every Monday morning the loop emits a digest: this week's experiments, last week's evidence deltas, any pivot-detector fires, the closest cohort neighbour's last move. Friday emits a recompute preview — what the score would be right now if we ran the math pass. The full recompute fires at week 12 unless you trigger one earlier.

Twelve weeks is enough time to learn the shape of the operator your audit said you could be. The loop is what makes the audit honest about whether you actually became them.