What v0.2.0 delivers, annotated

The site stops being a workflow explorer and becomes a study site. A masthead carries the study identity from config, a snapshot timeline, and a provenance chip; a collapsible sidebar organizes the study into domains — Study Overview, RBQM, Safety, Config — each nesting its own reports, charts and metrics from the study's registry.

1 2 3 4 5

RBQM leads with the ranked site risk-score table — the denominator beside every score, contributing flags and weights on hover, a since-last-snapshot change chip, low-precision sites dimmed rather than ranked — and, beside it, a funnel plot of the same sites against their precision, with the study mean and 95% / 99.8% control limits. Ranked lists of institutions are what funnel plots were proposed to replace; showing both is the point.

1 2 3

Before any finding, the Safety overview states what everything else is read against: census and exposure — enrolled, randomised, dosed, person-years, deaths — then data coverage by visit, the figure that decides whether a quiet visit is reassuring or empty. Below it: the participants the safety metrics flagged for review, and the gallery of rendered safety.viz charts, each opening the live renderer.

1 2

Every number on the site comes from a snapshot tree, and the site says so. The provenance chip expands into the snapshot's reproducibility record — input data version, package snapshot, pipeline status, and the pinned version + SHA of every package in the run. Pick a different dot on the timeline and the whole app re-points at that snapshot's tree.

The reproducibility record. Eight packages pinned with SHAs — gsm.core through open.gismo itself — plus the input data version and pipeline status. "Every number on this site comes from this snapshot tree" is a claim the chip lets you check.

Time travel. One click on ps-001 and the app is the earlier cut: 760 enrolled, 11 red sites, its own provenance. The snapshot date itself now derives from the data cut, not the clock — two cuts run on the same day stay distinguishable.

DEMO-301 is not a mock: both snapshots were produced by this release's pipeline, from CSVs in a plain project folder, by one recorded script. The engine is four verbs, and the site payload they write is exactly what the study site reads.

# A self-contained study in a folder — no GitHub repo, Actions, or Pages required
og_init("demo-301")                      # scaffold: config, workflow YAMLs, example data
og_validate("demo-301")                  # per-domain validation of the input CSVs, in sentences
og_run("demo-301")                       # mappings → metrics → reporting → report modules, via {workr}
og_view("demo-301")                      # serve the study site you just saw

# Snapshot date comes from the data cut (override: og_run(snapshot_date = ...));
# history accumulates under history/, so change columns appear from snapshot two.

A gsm.qtl report, inside the study site. The reports phase collects module HTML wherever the module wrote it and hands it the mapped domains and longitudinal results — so quality tolerance limits get a real report module, not a hard-coded slot.