The research queue

The problem register

Each problem below has a known answer that no system could deliver cheaply enough to matter. We file them by the kind of fix they need: engineering, when the bottleneck is distribution, scheduling or latency; system dynamics, when it is routing, incentives or verification.

P-01 ENG

Bloom's 2 Sigma Problem

Scalability of personalisation · Pure engineering

One-to-one mastery tutoring works; Bloom measured the effect at two standard deviations. It never scaled because expert attention costs too much to give everyone. That cost is what cheap computation now removes.

P-02 ENG

The Forgetting Curve

Spaced-retrieval bottleneck · Engineering · scheduling

Forgetting follows a predictable curve. A scheduler that times each review for the moment just before recall fails turns memory into something you can plan around.

P-03 SYS

The Proxy Trap

Goodhart's law of assessment · Cybernetics · incentives

Tie rewards to a test score and people optimise the score rather than the learning it stands for. The loop is wired to the wrong signal. The fix belongs in the incentive.

P-04 SYS

The Factory Model

Batch-processing rigidity · System dynamics

Sorting students by birth year is a routing policy borrowed from the factory floor, not a fact about how people learn. Route them by what they have actually mastered instead.

P-05 ENG

The Latency Problem

Feedback latency · Pure engineering

Homework set on Tuesday and returned on Friday closes the loop three days too late to be useful. Shortening that delay is a plumbing problem, not a teaching one.

3 d → 300 ms
P-06 SYS

The Signalling Problem

Credential–competency unbundling · Market design

Employers use the degree as a slow, expensive proxy for competence. Verifying specific skills directly is a matching problem, and a solvable one.

ENG Fixable by building the system: distribution, scheduling, latency. SYS Fixable by re-aligning the loop: routing, incentives, verification.