Reading the evidence codes¶

The letters are abbreviations:

q — design tier. Where the study sits on the causal-design ladder: 4 a pre-registered trial or strong meta-analysis, 1 a case study, qualitative study or argument.
i — impact. A legacy bin for a reported statistic, in the study's own terms.
n — sample. Sample size, free text — often a description ('large (multiple course sections)') rather than a number, because a meta-analysis and a classroom study do not count the same thing.
kind and r — rigour. How well the study does what its kind can do, judged against that kind's own standard and never another's (criteria in eval/kind-rigor/RUBRIC.md).

A subclaim is prefixed with the first two, bare: q3 i2 means a q3 study with an i2 effect. An evidence entry spells them out in words beside the code and adds n= and the study's kind with its rigour, qualitative · r3.

Kind — what sort of evidence the study is

Kind	What it covers
`causal`	Randomised or quasi-experiment, or single-case design: a manipulated condition and a comparison
`quant-synthesis`	Meta-analysis, or systematic review with quantitative pooling
`review`	Narrative or scoping review, or a book-length synthesis without a systematic search
`associational`	Correlational, survey, longitudinal or observational study, with or without statistical controls
`qualitative`	Interviews, focus groups, ethnography, case study, thematic or discourse analysis
`design`	Design-based research, development or evaluation of an artefact, or a practitioner's report of a design
`theoretical`	An argument, model or framework, with or without cited support

r — rigour, judged within the kind

Code	Meaning
`r3`	Strong for its kind — Meets its kind's criteria: e.g. a randomised, powered, low-bias trial; a qualitative study with clear aims, systematic analysis shown and credibility checks
`r2`	Adequate for its kind — Most of its kind's criteria met, with clear gaps
`r1`	Weak for its kind — Few of its kind's criteria met: tiny or confounded comparison, thin description, bare assertion

r?: r? means the text available (often an abstract) could not show whether the criteria are met; with no text of the study at all, rigour is always r?

q — position on the causal-design ladder

Code	Criteria
`q4`	Pre-registered RCT or well-powered meta-analysis
`q3`	Peer-reviewed experiment (not pre-registered) or systematic review
`q2`	Quasi-experiment, observational with controls, or narrative review
`q1`	Case study, expert opinion, or theoretical argument

i — impact magnitude

Code	Rough effect size
`i3`	Large — d >= 0.8 or equivalent
`i2`	Medium — d 0.4-0.79
`i1`	Small — d 0.2-0.39
`i0`	Negligible: an effect size is reported and it is below d 0.2 or equivalent

These legacy bins describe a reported statistic, not educational importance. Compare effects only with the learner population, assignment and comparator, outcome measure and its alignment, score range, follow-up, and uncertainty in view. Kraft's education benchmarks concern causal K–12 interventions measured on standardized achievement tests; they do not automatically replace these bins for other outcomes.

A Kraft tier requires source-verified learner, design and measurement context for the particular effect, recorded as impact_context on its observations/ record. Where none is recorded, its eligibility and comparability remain unreviewed; see Comparing learning intervention effects.

A ? in place of a digit — q?, i? — written but not yet established. q? is not the same as an absent q: it says someone looked and could not say, which is a finding rather than a gap.

Strength here describes the research. Whether anyone has checked that this page reports it faithfully is a separate axis — see trust in the page metadata below, and the Schema & Guide.