Main idea

Filtering harshly at the moment you discover a source looks clean: before saving an article or a video, you would ask whether it really deserves to enter the system. In practice, the state you are in at that instant doesn't allow that judgment. There is the thrill of the find, and a source seems essential mostly because it answers a question you have only just asked yourself.

Yet you shouldn't keep everything, and the criterion that holds isn't "is this interesting?" — almost anything is, to someone curious. It is "does this look genuinely promising for my projects, my thinking, or my acquisition of knowledge?". The word promises little: it asks for a plausible attachment, not a justification.

A capture is therefore not a validation. It is a holding pattern.

Why it matters

This avoids the two symmetrical failures of an intelligence system. A filter that is too demanding at intake loses sources that would have mattered, and it costs a full arbitration at every encounter. No filter at all produces a stock you will never open again.

It also moves the load: the serious sorting doesn't have to be paid for at the moment of discovery, where it is done badly, but later, where it is decidable.

Nuances and limits

"Promising" remains a judgment, and it needs calibrating: too loose, and it becomes "interesting" again. You observe the setting downstream, in the rate of captures thrown out without regret.

And the criterion presupposes identifiable projects. Someone exploring a field with no destination has nothing to attach the promise to.

Open questions

• What proportion of captures eventually thrown out indicates a well-set filter rather than an absent one?