85-years-of-learning figure as if it were a property of the intervention rather than a property of a specific delivery channel: there is no standardized, validated instrument that predicts, before a government scales a pilot nationally, how much of the effect size will survive the handoff from NGO-run implementation to routine bureaucratic delivery.

's Kenya teacher-contract experiment, the Action Schools! BC cluster-RCT in Canada, decades of "voltage drop" anecdotes across global health and education — that all document the same pattern after the fact.

What we do not have is a leading indicator that a finance ministry or a donor can check before signing a scale-up budget: something equivalent to a stress test that says "this program's effect depends on implementer-level monitoring intensity X, your civil service currently delivers at intensity Y, and Y is below the threshold where the effect held in trials."

Instead the field relies on qualitative "readiness assessments" that are rarely pre-registered, rarely validated against actual scale-up outcomes, and almost never publicly scored the way effect sizes themselves are.

Implementation science has produced frameworks — RE-AIM, the Consolidated Framework for Implementation Research — that are good at describing after the fact why something degraded, and much weaker at forecasting the magnitude of degradation before money moves.

J-PAL and its partners have been unusually honest that TaRL's own scale-up required years of iterative adaptation with Pratham still embedded in delivery; that adaptation process is exactly the tacit knowledge that a standardized instrument would need to capture and that currently lives in program officers' heads rather than in a replicable protocol.

The cost of not solving this is not abstract.

Every government that scales an education, health, or cash-transfer pilot on the strength of an RCT effect size and then gets the Kenya result — a measured near-zero — has burned a budget cycle, a political mandate for "evidence-based policy," and, in the education case, a cohort of children who did not get the learning gains the citation promised them.

Given that roughly 70% of 10-year-olds in low- and middle-income countries were estimated to be below basic reading proficiency post-pandemic per the World Bank's learning poverty metric, the stakes of getting scale-up prediction wrong are not a rounding error, they are the difference between hitting or missing foundational learning targets for hundreds of millions of children.