Breaching Questions

A breaching question finds leftover unfeelable holes by writing an unanswerable question on each. The hunt runs late, after ordinary retrieval has stopped turning anything up, and it pairs a sweep across the subject’s headings with that unanswerable question.

By this late stage the leftover holes are few, and the ones left hide. A known hole shows up as difficulty recalling or applying — the edges can be felt. An unknown hole is a perspective that was never there, so answers come confidently from what is known while an entire angle is missing. Ordinary self-testing rarely catches that second kind, because confidence tracks what can be retrieved, not what was never encoded. Nothing isolates that as a trialled finding; it is the reason the method exists.

Finding the soft spots

Ordinary self-testing is the wrong instrument for a hole that never announces itself. The sweep starts from the map of the subject already held: the concepts, facts, and details that will not come quickly or confidently. When that reflection stalls, the published syllabus headings, exam-spec topics, or chapter titles work as the spine — an exam built from a published spec can only ask what the spec covers.

Under each heading, every related keyword that can be reached gets listed, and the relations get mapped as they come. Difficulty forming that map is itself the signal of a known hole. Then a scan for the term that belongs and was not listed. The last move is the one the method is named for. A word sitting under one heading and missing from the next gets asked of the next heading anyway. That forced fit is what brings up holes that never announced themselves.

A term listed under control loops — feedback — asked of market regulation,
a heading that never held it. Either it applies and that angle was never
looked at (a hole, now visible), or it does not and the reason can be
stated (itself a check).

The question you cannot answer

Those surfaced holes, once listed, are only a list, and a list is not the method. The name is literal: the question breaches the plateau on one piece of knowledge that ordinary retrieval work — repeat explanation cycles, flashcards, revision — has already stopped moving.

Related soft spots group into one question; scraps that will not group take one each. The question is written by the learner and aimed above the hardest thing the assessment could plausibly ask — no real curveball could be that curved. It must be unanswerable at the moment of writing. A question that can be answered as it is written is a revision question.

The question is not written to be answered.

The value is the depth of the dig, which usually runs out of scope, and the soft spot comes out strong. A maximally hard question draws on several concepts at once, so answering one covers neighbouring material without aiming at it. Time scales with importance: on the order of thirty minutes to two hours for a central concept, fifteen to twenty minutes for a minor detail, then stop whether or not it is answered.

The same term that cross-applied becomes the question. Feedback, asked of market regulation at a height no real paper would reach, is a question the writer cannot answer as it is written. That is the check.

The method is advanced. It asks for cognitive-load tolerance and the ability to map relationships non-linearly in the head. It rests on the relational depth built through Higher Order Learning; without that network the keyword lists stay flat and there is nothing to cross-apply.

The sweep costs an evening per subject; each question costs fifteen minutes to two hours to chase. It spends time the late stage has, not time a full schedule has. Run early it produces flat keyword lists and nothing to force. Near an exam it can eat hours going out of scope while broad coverage is still weak. Nothing isolates its advantage over ordinary practice. Two headings’ worth of cross-application turning up nothing new means the holes left are not the kind this finds: the next move is timed practice. Difficulty mapping one topic is the diagnostic, and the sweep continues. Inability to hold any map at all means the network is not there yet, and the network comes first. Past papers adjust for predictable styles. Concept mapping draws the network already held, rather than forcing terms onto headings that never held them. Those are neighbours, not this method.

It sits in the last stretch, once Spaced Interleaved Retrieval has already cleared the obvious holes and Revision — the ordinary test, target, teach pass — has stopped turning anything up. The layer immediately before is practice papers, in roughly the last two to three weeks; the signal to switch is practice questions no longer surfacing anything new. What the late stage is buying is a small number of much less predictable marks, not coverage. The protocol is this system’s own teaching model. Its parts are plausible — confidence tracks retrieval, forced comparison surfaces relations — and no trial isolates it.

Sources

• Kruger, J., & Dunning, D. (1999). Unskilled and unaware of it. Journal of Personality and Social Psychology. Confidence can survive the absence of the knowledge that would reveal the miss.
• Koriat, A. Metacognitive monitoring. Confidence tracks what gets retrieved, not what was never encoded. Neither paper isolates this protocol.