Difficulty Calibration — Map of Content

Most products pick a difficulty once, in a room, for an average user who does not exist. This skill owns the alternative — putting each user at the edge of what they can currently do, and knowing whether you actually are.

Three ideas run through everything below. Difficulty is a property of the pairing of item and user, not of the item. The target is a band with an upper bound, not a floor. And the user's own satisfaction signal points the wrong way, because comfortable repetition feels excellent and teaches nothing.

This skill does not own whether a single screen is asking too much at once. "This is too hard" arrives here by default, but an item can be perfectly targeted to the user's ability and still fail because five things compete for attention simultaneously — that is [[../duo-attention-budget/SKILL]].

This skill is a graph: scan the descriptions, follow only the [[wikilinks]] you need.

Building the model

• [[references/model-the-item-and-the-user-jointly]] — estimate content difficulty and user ability in one system, because difficulty is a property of the pair and half the signal is useless alone.
• [[references/read-your-difficulty-weights-as-a-content-brief]] — the fitted per-item parameters separate hard-because-unfamiliar from hard-because-structural, which is a redesign queue, not just a scheduling input.

Choosing the target

• [[references/the-narrow-band-between-bored-and-lost]] — specify an interval, not a threshold; control the familiar-to-new ratio rather than session volume.
• [[references/train-above-the-real-load]] — behavior attenuates under real conditions, so practice at exactly the target level lands below it. Over-constrain on purpose, and pay for it knowingly.