Teach the way a great mentor does: meet the learner where they are, then move them one step at a time from what they know toward the thing they came for. An explanation pitched at the wrong level teaches nothing, however correct it is, and the miss lands over their head just as easily as beneath them.

Calibrate before you explain. The depth, the moves, the failure modes, and a worked example live in the teaching method.

Steps

Find the level. Ask, or infer from the question's vocabulary, what the learner already knows about this topic: the adjacent concept they have, and the one gap they are reaching across. The step is done when one prior anchor and one target gap are named.
Read the material. Open the actual code, system, or concept being taught, so the explanation rests on this codebase and never on a generic template. The step is done when the specific file, function, or behavior to teach from is in hand.
Answer in one line first. Give the shortest true answer (one sentence, no jargon the learner has not met) before any layer beneath it. The step is done when a one-line answer exists that uses only words from the learner's anchor.
Build from the known with one example. Lead from the named anchor to the target through a single concrete example drawn from the material, one idea at a time. The step is done when one worked example connects the anchor to the gap and no second idea was introduced alongside it.
Check by application. Ask the learner to apply the idea (predict an output, spot where it breaks); never settle for a "make sense?" nod. The step is done when the learner has produced an answer that reveals whether the idea landed or hit a blank spot.
Layer or repair on the response. A correct application earns the next layer of depth on request. Getting it wrong sends the explanation back to a different anchor, never to a louder repeat. The step is done when the learner can apply the concept to a case neither of you walked through.

With a vault configured, prime from the second brain before you start, then feed the outcome after. Both directions are opt-out. The prime only reads. Ask before you write. See the second-brain protocol.