Generated from smart-base/skills/content/authoring-who-smart-guidelines/grade.md — do not edit here.
✎ Edit this page’s source

grade

Skill id: grade · Package: authoring-who-smart-guidelines · The grading SYSTEM that evidence-appraisal applies when a folio declares GRADE. Origin: the GRADE Working Group (gradeworkinggroup.org); Evidence-to-Decision frameworks per Alonso-Coello et al., BMJ 2016; WHO applies it through the WHO Handbook for Guideline Development.

evidence-appraisal is the procedure and deliberately names no system — which one applies is the folio’s commitment. This is one system a WHO folio can declare. It applies to certainty of evidence for a health recommendation, over a body of evidence answering one PICO question — not to platform or architecture decisions, which are kepner-tregoe, recorded per madr.

Every value below is a code, and the codes are nodes. Each vocabulary is a folio-code-list/v1 in this instance’s code-lists/, with a definition and a source per code, published as SKOS beside the glossary. Record a code, never a paraphrase of one:

list	codes
`grade-certainty`	`high` · `moderate` · `low` · `very-low`
`grade-rating-down`	`risk-of-bias` · `inconsistency` · `indirectness` · `imprecision` · `publication-bias`
`grade-rating-up`	`large-effect` · `dose-response` · `plausible-confounding`
`grade-etd-criterion`	the twelve Evidence-to-Decision criteria
`grade-recommendation-direction`	`for` · `against`
`grade-recommendation-strength`	`strong` · `conditional`

Moved 2026-09-24 (bean wg7r) from smart-kg/methodologies/grade.md. The owner: “needs to be part of skill/SKOS etc. … dont want smart-kg here yet, that is its own repo already”. The method is unchanged; its enumerations became code lists, so a recorded grade is a code with a definition rather than a string.

It has a slot waiting

processes/evidence-retrieval.bpmn frames the question as PICO, retrieves candidates across three classes, and runs Task_AppraiseGrade — “Certainty of evidence per the grading system the folio declares. The grade attaches to the BODY of evidence for one PICO question, not to an individual citation.” The slot stays pluggable, and that is the point: GRADE fills it when declared, and nothing in the diagram names it.

Certainty, in four levels

High · Moderate · Low · Very low — the certainty that the true effect lies close to the estimate. Randomised trials start High and observational evidence starts Low; then:

rated down for risk of bias, inconsistency, indirectness, imprecision, publication bias;
rated up for a large effect, a dose-response gradient, or where plausible confounding would have reduced an observed effect.

The grade attaches to the body of evidence per outcome, never to a citation. A single paper does not have a GRADE. The BPMN task already encodes this by sitting after the join rather than inside the fan-out, and it is the distinction most often got wrong. Record every rating-down and rating-up with the code that moved it — a certainty with no reasons cannot be re-examined.

Evidence-to-Decision: certainty is one criterion of several

A recommendation does not follow from certainty alone. The EtD framework for clinical and public-health recommendations judges each criterion explicitly: problem priority · desirable effects · undesirable effects · certainty of evidence · values · balance of effects · resources required · certainty of the evidence of required resources · cost-effectiveness · equity · acceptability · feasibility. (The file this replaced listed ten, omitting the two resource criteria; the code list follows Alonso-Coello et al. 2016.)

Then a recommendation carries a direction (for / against) and a strength (strong / conditional). Strength is not certainty. A strong recommendation can rest on low-certainty evidence when the balance of effects is decisive, and a conditional one can rest on high certainty when values differ across populations. Conflating the two is the most common misreading of GRADE, and the reason they are two code lists rather than one score.

What this platform must not do with it

No numeric aggregation of EtD criteria. The framework judges each criterion and argues the recommendation; it does not sum them. A weighted total would be the MCDA failure this platform refuses elsewhere — a judgement laundered into a measurement.
could not determine is not very-low. Absent evidence and low-certainty evidence are different findings with different actions. evidence-retrieval models this with EndEvent_GapRecorded — “Gap recorded, no recommendation” — a distinct terminal from an evidence-backed one, and evidence-appraisal names it a third state.
No empirical claim enters a formal proof. A graded body of evidence supports a recommendation, never a theorem. In a paper folio the two live in different block kinds and must not be joined.

Refusals

Never grade a single citation. The body, per outcome.
Never report a recommendation without its direction and strength. One without the other is not a GRADE recommendation.
Never table it in DMN. See dmn — tabling asserts a repeatability the method denies.