The School Builder's Playbook

How to read this

Every claim below traces to a verified topic page (linked) with graded sources behind it. The grading constitution (METHODOLOGY) privileges RCTs, lotteries, and natural experiments; flags genetic confounding in observational work; discounts researcher-designed outcome measures ~2×; and tracks fadeout separately from end-of-treatment effects. Three database-wide regularities frame everything:

1. Every famous number deflates. Hattie's 0.79 feedback, Black & Wiliam's 0.7 formative assessment, Bloom's 2-sigma tutoring, mastery learning's 1.0 — all shrink 3-20× on independent measures at realistic scale. The honest big effects in education are ~0.2-0.4 SD.
2. Nothing moves g. Effective instruction moves domain skills, attainment, and life outcomes — sometimes dramatically — while general ability stays put (with one small Swedish exception noted below). Good instruction raises the heritability of skills by removing environmental bottlenecks: it is floor-raising, not gap-erasure.
3. Test-score fadeout is the default; attainment persistence is the prize. Early gains revert toward genetically-anchored trajectories — but class size, tutoring, and double-dose algebra all show credential and life-outcome effects that outlive their score effects. Build for the long game and judge levers on the right ledger.

---

The ten commitments

1. Teach everything explicitly, and never use minimal guidance. The single most consistent finding across domains: unassisted discovery reliably loses (d≈−0.38); the constructivist curriculum lost the only multi-curriculum RCT by 0.22 SD; explicit instruction for strugglers is the most-replicated finding in intervention research. Guided inquiry and structured problem-first work are fine — minimal guidance is the poison. → explicit vs inquiry · explicit vs reform math · worked examples

2. Build retrieval and spacing into the curriculum's bones. The two strong-support methods that apply to every subject: low-stakes quizzing with feedback (durable-retention effects across 222 classroom studies) and distributed practice (the most robust finding in learning science). Students won't self-administer either — schedule them: cumulative reviews, spiral sequencing, cumulative exams, spaced homework. → retrieval practice · spaced practice · interleaving

3. Teach reading as code-then-knowledge. Explicit systematic phonics K-1 (the settled part), fluency through volume, and then the under-rated lever: a knowledge-rich cumulative curriculum as the actual comprehension program — not comprehension-strategy units (a one-time trick, ~null at scale). The Core Knowledge lottery (ITT 0.24 on state reading) is the best curriculum bet in the database. → teaching children to read · background knowledge

4. Teach math as explicit modular strands. Fluency (timed, low-stakes — the anti-timed-test movement has no causal evidence), word-problem schemas, and concepts↔procedures in alternation. Math sub-skills don't create each other; teach each on purpose. Place students in algebra by measured readiness; accelerate the ready; double-dose the rest (+3.3pp BA attainment, twelve years out). → teaching children mathematics · algebra timing

5. Hire for demonstrated performance; fire the credentials budget. Teacher quality is the largest within-school lever (1 SD ≈ 0.10-0.15 SD/yr — more than a 10-student class-size cut) and invisible on paper: certification null, master's null. Open the funnel, judge on 2-3 years of measured performance, gate retention on it. Replace workshop PD (~$18k/teacher/yr, causally dead) with content-specific coaching and evaluation-with-feedback — the two development levers that survive. → teacher quality

6. Tutor the students who fall behind — the expensive boring way. Education's most reliable intervention at its honest size (~0.2-0.3 SD; never Bloom's 2.0): groups of 2-4, paid consistent tutors, inside the school day, 3-5×/week, 30+ delivered hours, $1,400-4,800/student. Never opt-in, after-school, once-weekly, or rotating tutors. Assume decay; re-dose across years. → tutoring · Reading Recovery cautionary tale

7. Group by measured readiness, flexibly — and accelerate. Within-class and cross-grade grouping with genuinely differentiated instruction are ~free wins; the one tracking RCT raised the whole distribution via teach-to-level. Keep placements permeable and re-measured (locking 10-year-olds into tracks is where causal harm lives). Screen everyone objectively (universal screening tripled disadvantaged gifted placement); accelerate on mastery criteria (no harm + a saved year); run achievement-selected advanced classrooms (+0.5 SD for high-achieving minority students). Skip the gifted-label apparatus. → grouping & tracking · acceleration & gifted

8. Give feedback as task information; never as person evaluation. Feedback is the only method with a ~1/3 backfire rate. What works: specific task information, comments without grades on formative work (the grade cancels the comment), and slashed marking volume (no causal evidence for the hours). What backfires: praise, ego comparisons, discouragement. Treat "formative assessment" as a cheap ~0.1 bet, not a 0.7 transformation. → feedback & formative assessment

9. Choose curricula once, by exclusion list — then buy implementation. The whole cost is attention (~$36/student): exclude minimal-guidance programs (worth ~0.1-0.2 SD/yr of avoided harm), pick structured knowledge-rich materials, and spend on training and actual use — the field's default (3 days per career) guarantees the null. Don't chase rankings among mainstream texts (~0-0.05 SD at stake). → curriculum choice · math curricula

10. Buy the bundle, not the inputs. Five practices explain ~45% of school effectiveness — high-dosage tutoring, extended time, data-driven instruction, frequent teacher feedback, high expectations — while class size, spending level, and credentials explain ~zero. The bundle transplants into ordinary schools at ordinary budgets ($355-1,837/student, IRR 13-29%). Keep classes ~22-25 with excellent teachers; consider small classes only in K-1 and never by lowering the hiring bar. Match intensity to intake: the No-Excuses evidence is grade-A for disadvantaged populations and absent-to-negative for advantaged ones. → spending & school effects · class size

---

The refuse list

Money and hours the evidence says to withhold: whole-language/balanced-literacy and three-cueing · minimal-guidance discovery curricula in any subject · comprehension-strategy programs as the comprehension plan · standalone speech-only phonemic-awareness curricula · sustained-silent-reading as an intervention · dropping timed practice on anxiety grounds · concepts-first sequencing dogma · rich toy-like manipulatives and unguided hands-on time · algebra-timing mandates in either direction · strict mastery gates that tax fast learners (keep the diagnostic core, drop the gate) · master's-degree pay bumps · workshop PD · written-marking volume and grades on formative work · person praise · gifted-label gatekeeping by subjective referral · detracking on equity grounds · opt-in/after-school/once-weekly tutoring · commercial interim-assessment data platforms · Reading Recovery at its price · chasing textbook rankings and WWC curriculum labels · class-size reduction that dilutes teacher quality — and Bloom's 2-sigma, Hattie's 0.79, and Black & Wiliam's 0.7 as planning numbers.

The budget hierarchy

Per unit of learning, spend in roughly this order: (1) curriculum exclusion list + knowledge- rich adoption (attention-priced) · (2) retrieval/spacing architecture (free, scheduling) · (3) grouping-by-readiness + acceleration + universal screening (free, saves money net) · (4) teacher selection & retention machinery (reallocates existing payroll) · (5) coaching + evaluation-feedback ($3-7k/teacher, replaces dead PD spend) · (6) high-dosage tutoring for whoever is behind ($1.4-4.8k/student served) · (7) extended time · (8) small K-1 classes for disadvantaged intakes — the only defensible class-size buy · (9) unallocated spending increases (the worst buy: ~$12.7k per 0.1 SD).

The honest limits

• Fadeout is undefeated. No intervention reliably prevents early score gains from reverting; the defensible responses are cumulative curriculum, maintained practice, re-dosed tutoring, and judging levers on attainment ledgers where they exist (STAR, double-dose, finance reforms).
• Reading comprehension is the hardest thing to move (math moves 2-5× more in every structural study) — which is exactly why the knowledge-curriculum bet and the early-code investments matter.
• The right tail is under-evidenced. Most causal evidence concerns struggling and average students; for profoundly gifted students the honest offer is acceleration (no harm + saved years) and advanced content, on thinner data.
• Implementation is the graveyard. The 2004 double-dose cohort, the Nashville tutoring program, the national tutoring schemes, the comparison-methods scale-up — every one died on delivered-dosage and fidelity, not on theory. Whatever you adopt, measure what was actually delivered.
• One Swedish exception to "nothing moves g": a class-size RD moved IQ-type measures durably (+0.02 SD/pupil, for low-income sons). Small, unreplicated, honestly recorded.