The Evidence on Teaching

The ledger

The archive’s open books, at both grains: every decision a school builder faces — the 0 we haven’t answered left visibly empty — and every paper read, including the 129 read and not used. Most evidence reviews show you only what survived, which makes a thorough search indistinguishable from a lucky one.

54/54
decisions answered
100% of the declared tree
1171
papers read
1042
counted as evidence
cited by a verdict
103
read, set aside
with a stated reason
26
could not obtain
recorded, not dropped

The decisions · 54 of 54 answered

An empty cell means not yet read — not “no evidence exists,” and not “nothing works.” We would rather show you the hole than quietly crop it out of frame. The plan lives in TAXONOMY.md; the build fails if this page and the database disagree.

Subjects — how to teach each

12/12

Methods — how humans learn (apply across every subject)

9/9

Structure — how to organize the school

11/11

Early childhood — before and around school entry

5/5

Don't-bother — debunked / fad methods

4/4

Inputs — non-academic factors that move outcomes

6/6

Assessment & intake — reading what a test actually says

3/3

Homeschooling — the delivery context

3/3

Subject topics still to build

1/1

The papers · all 1171

How closely these were read

363 of 1171 were graded from the full paper. An effect size taken from an abstract carries no attrition, no moderator table and no limitations section behind it — so abstract-only is debt, not a neutral state, and it is shown rather than smoothed over. 308 sources standing behind a live verdict were graded without the full paper.

Checking status

271 of 1171 have had their extracted numbers re-checked against the paper by a second, adversarial pass. The rest are marked unchecked below — they are not wrong, they are unaudited, and the difference matters enough to show.

grade A: 56grade B: 371grade C: 482grade D: 262