We tried to refute 20 growth tactics. 11 died. Here is the graveyard.

Every growth list you have read has the same structural flaw: it only contains survivors. Somebody collected tactics that appeared to work, wrote them up, and shipped. Nothing in the process tried to kill them.

So I added that step. Twenty candidate patterns went into an adversarial pass where the job was to refute each one against the evidence rather than confirm it. Eleven did not survive.

Here they are, with the score. "Refuted 3/3" means three independent re-checks each failed to sustain the claim.

1. Weekly, opaque, or retroactive usage caps as the number one cancellation driver. Refuted 3/3. Real cancellation comments over caps exist. The "dominant driver" framing did not hold.
2. Auto-listing public repos in your directory, then filing a GitHub issue to "claim" it. Refuted 2/3.
3. Outcome-and-refund guarantee with a hard deadline, pinned everywhere. Refuted 3/3.
4. Multi-hour "full courses" as evergreen top-of-funnel. Refuted 3/3.
5. Affiliate links with named creator promo codes. Refuted 3/3.
6. README as a conversion asset, with trust badges, testimonials, funnel numbers. Refuted 3/3.
7. Rewriting copy from feature terms to the one pain or fear removed. Refuted 3/3.
8. Reducing churn by embedding into the customer's systems and doing setup for them. Refuted 3/3.
9. Value-based or outcome pricing, and the free-audit to retainer ladder. Refuted 2/3.
10. App-store optimization: keyword in the name, distinctive first screenshot, review modal at the emotional peak. Refuted 3/3.
11. Launching paid from day zero, free trial and never freemium. Refuted 3/3.

The two that will annoy you

Number 7 is the one I expected to survive. "Stop describing features, describe the pain you remove" is close to universal advice. It reads as obviously true. It did not hold up, and I have had to sit with that.

Number 6 stings for a different reason. Most of us have spent an afternoon on a README hoping it would convert. The evidence says the README earns installs by being clear, not by being a landing page with badges on it.

What refuted does not mean

It does not mean the tactic never works for anyone. It means the evidence that looked like a pattern fell apart when someone actively tried to break it, so it is not a foundation to build distribution on. If you already do one of these and your own numbers say it works, your data beats this list. That is not a hedge, it is the correct epistemics: n=1 that you measured yourself outranks a pattern you read on the internet.

Why publishing this is the point

Eleven of twenty is a 55% failure rate among tactics that all sounded good enough to make a first draft. That number is the actual finding here. Not any individual entry.

It should change how you read the next growth thread you scroll past. Somebody wrote it, it sounded right, and nothing in the world checked it. Most of what circulates is confident-sounding and wrong, and there is no filter between you and it except the one you build.

The full graveyard, plus the nine patterns that did survive with a source URL on every claim, is an MIT Claude Code plugin at why-isnt-it-selling. The graveyard is in the free tier, because it is the part I would want before trusting anything else in there.