Learning improves when the system can observe the attempt, model the state and return the next signal fast enough.
Fig. 1 — a stylised two-sigma benchmark for one-to-one mastery tutoring. Bloom, 1984.
Mereth Labs takes education's oldest bottlenecks and treats them as systems you can actually measure — slow feedback, memory left to chance, incentives aimed at the wrong target, skill nobody can verify. It starts as research and is built to leave as software.
Est. 2026Bengaluru6 problems in queueMethods: measurement, retrieval modelling, control loops, mechanism design
The thesis
The evidence is old. The delivery stack is new.
Bloom's 1984 "2 sigma" paper made the benchmark vivid: one-to-one tutoring with mastery learning can shift achievement far beyond ordinary classroom instruction. The exact effect depends on domain, tutor quality and implementation, but the mechanism is not mysterious: frequent diagnosis, targeted feedback and progression after mastery.
Many education bottlenecks have the same structure. The learner has a changing hidden state. The system observes that state poorly, acts too late, or rewards the wrong signal. Computation matters because it lowers the cost of observing, estimating and acting on that state for each learner.
The claim must be measurable.Each project starts by naming the state variable, the signal, the intervention policy and the evaluation metric.
The product must preserve the science.If a system cannot be instrumented, audited and compared against a baseline, it is not ready to scale.
Technical approach
Every problem becomes a loop.
We turn a learning question into a loop you can measure, test and tighten — without pretending education is only a software problem.
Observe
Instrument the attempt
Capture the learner's action, timing, confidence, error type and context. A score alone is too coarse; the trace is where the useful signal lives.
Infer
Estimate the state
Maintain explicit beliefs about mastery, recall probability, misconception class and uncertainty. The model should say what it knows and what it does not.
Act
Choose the next signal
Pick the next move from the estimated state: what feedback to give, when to schedule a review, how hard the next task should be, which way to route. The goal is a loop you can steer, not personalisation for its own sake.
Validate
Measure the outcome
Track learning gain, retention, transfer, calibration, latency and robustness to gaming. A fluent interface is not evidence of learning.
The research queue
Six problems as systems.
Each is filed by the kind of fix it needs: engineering, when the bottleneck is measurement, scheduling or latency; system dynamics, when the bottleneck is routing, incentives or verification.
ENG Fixable by building the system: distribution, scheduling, latency.SYS Fixable by re-aligning the loop: routing, incentives, verification.
Where it goes
Research that can survive contact with use.
A paper that never leaves the page helps nobody. Everything we study is built to ship as a product you can inspect — assumptions, measurements, outcomes and all.
For learners
Consumer tools
Software that gives a learner tighter feedback, better retrieval timing and clearer mastery signals than a fixed classroom clock can provide.
Retrieval scheduled from observed recall, not generic review plans
Feedback returned while the attempt is still actionable
Progression based on mastery evidence, not seat time
For industry
Institutional infrastructure
Tools for schools, training teams and employers that need stronger evidence than grades, attendance or coarse credentials.
Assessments designed against Goodhart pressure
Portable skill records with explicit validity evidence
Routing policies that adapt without hiding their criteria
Mereth Labs is at day one.
If you build learning systems, run a classroom, evaluate skills, fund research, or just have a problem with measurable stakes — get in touch. We want collaborators who care about the evidence and the implementation in equal measure.