Stop studying viral videos. Your flops know more.

In World War II, the statistician Abraham Wald was asked where to add armor on bombers. The military had mapped the bullet holes on planes that returned: wings, tail, fuselage. The obvious plan was to armor those spots. Wald said the opposite — armor where the returning planes had no holes. Planes hit there never came back to be counted.

Every "study what's going viral" tool has the returning-planes problem, by construction.

The outward loop

The standard creator research loop looks like this: find outliers (videos massively overperforming their channel), decode why they worked, imitate the pattern. A whole shelf of tools automates it, and it feels productive — you're looking at data, after all.

But the dataset is only survivors. It cannot show you:

• The near-identical videos that died, posted the same week by accounts you'll never see.
• Whether the "pattern" you extracted is the cause or a coincidence of the winner you happened to study.
• Anything at all about your account — your audience's contract with you, your delivery, your baseline.

That last one is the killer. A hook that works at someone else's baseline can sit dead on yours, because their audience follows them for confrontation and yours follows you for competence. When you copy a winner and it flops, the outward loop has no explanation for you. It just serves the next winner.

Your flops are proprietary data

Here's what's strange about ignoring your own underperformers: they're the one dataset your competitors can't access, and they're the only posts that carry information about what suppresses reach on your specific account.

Your winners tell you what your ceiling looks like. Your flops tell you what keeps pulling you under your own median — and unlike a stranger's viral video, they come with everything attached: you know the hook you wrote, the topic you picked, the day you posted, how you opened, when you pitched. Attribution is actually possible.

The pattern is rarely visible one post at a time. That's why the flops feel random — "sometimes it hits, sometimes it doesn't." Line up the bottom ten posts of almost any stalled account and they stop looking random: six of them open with a wind-up, or five of them are the off-lane topic, or the self-promo moved into the first half around the same month reach fell.

One flop is bad luck. A cluster of flops with a shared attribute is a habit with a fingerprint.

The inward method

The alternative loop points at your own account and closes:

1. Baseline. Median views of your last ~30 posts. This is your 1.0x line — your account measured against itself, so follower count and niche stop mattering.
2. Multipliers. Every post scored against the baseline. 2.1x and 0.3x are both information.
3. Buckets. Group the multipliers by hook style, topic, format, duration, CTA. Skip any bucket with fewer than 3 posts — small samples lie.
4. The suppressor. The recurring attribute of your below-baseline cluster, named with the numbers that convict it.
5. One change, then a verdict. Change a single variable for a week, then re-measure against the same baseline. Above the noise band: validated. Below: failed, revert. Inside it, or too few posts: inconclusive — run it again.

That last verdict matters more than it seems. A method that only ever tells you "great insight, you're crushing it" is a horoscope. If a system can't say "inconclusive," its "validated" means nothing either. Honest verdicts are what make the loop compound: every week ends with a fact about your account that stays true.

What winner-studying is still good for

This isn't "never watch other creators." Studying winners is genuinely useful for vocabulary — knowing the formats, hook shapes, and structures that exist, the way a writer reads widely. It expands what you can try.

It just can't tell you what's broken. Vocabulary is not a diagnosis. The order matters: diagnose inward first, then raid the outward world for material to run through your own filter — one variable at a time, verified against your own baseline.

The bullet holes you can see are on planes that made it back. The ones deciding your reach are in your bottom ten posts, and nobody else is looking there.