How the matching works

Two stages: hard limits, then weights

A constraint is treated as hard only if breaking it makes the format impossible this month, and no amount of willingness fixes it. Not wanting to appear on camera rules out every format that needs a face — that is a question of consent, not effort, and it is the one limit that is never relaxed. Having £0 rules out formats needing real kit. Having three hours a week rules out formats with a twelve-hour floor.

Editing skill is deliberately not hard for a one-level gap, because that is learnable in six weeks. A two-level gap is hard again: you would have to become a different person before your first upload. Platform choice is only hard if you say it is — otherwise a checkbox would quietly eliminate most of the directory without you understanding why.

What the weights are

Everything that survives the hard limits is scored on eight axes, out of 100:

• goal22
• time18
• energy17
• platform14
• editing11
• presence8
• cadence6
• budget4

Most curves have two sides

Having more capacity than a format needs is not a perfect match — it is a mismatch in the other direction. If you have thirty hours a week, a sixty-second vertical serial should not win. If you are completely at home on camera, a faceless compilation channel is a bad recommendation even though you could do it. A model where cheaper and easier always scores higher is exactly how three low-effort formats end up winning for everybody.

There is also a crowdedness penalty, worth up to 20%. Ease of entry and saturation are strongly correlated, and a model that ignores that will always conclude the easiest format is the best one to start.

Why nothing scores 100%

The scale is capped at 97%. No format is a perfect fit for a person, and a scale that tops out below the maximum makes an 82% read as information rather than a disappointment. The calibration constants are not chosen by feel — a script scores the entire answer space and fails the build if the median top match drifts outside 76–88%, or if more than 5% of results come out above 95%.

How we know it isn't a random-number generator

The obvious failure of a quiz like this is that the same three formats win for everyone, and it is invisible from the inside because every individual result looks reasonable. So every build enumerates the full answer space — 720,000 combinations — and checks the distribution of winners. If any of the 30formats can never reach anyone's top five, or can never be anyone's number one, the build fails.

It found a real problem the first time it ran: one format was winning 22% of the space purely because it had the fewest requirements. That led to three rounds of correction to the underlying data rather than to the thresholds.

What it can't do

It does not know your topic, your existing audience, whether you are good on camera, or whether you will enjoy the thing it suggests. It matches constraints to demands. That is genuinely useful for ruling formats out, which is most of the value — and it is why the results page shows you what was excluded and why.