safety.viz release review 2026-08-14

What v1.6.0 adds, annotated

The gallery crosses into nephrotoxicity, and the Hepatic Safety Explorer gets back the feature the original renderer was best known for. Each section below is a line on what changed, why it matters when you are reading a study, a capture of it running, and a way into the live demo to check it yourself. Every capture was taken on 14 August from the release candidate — the build this cut promotes, the same bytes the dev site deploys.

5 PRs, all merged to dev 1 new renderer — 12 in the gallery 1 178 unit + 236 browser tests promotes dev → main via sv#124

sv#121 · hub#35

Nephrotoxicity Explorer

Experimental

The twelfth renderer, ported from SafetyGraphics/nepExplorer: KDIGO acute-kidney-injury screening as one point per participant, at their maximum post-baseline fold change in serum creatinine against their maximum absolute change, over the KDIGO stage zones. Kidney sibling of the eDISH scatter, on the same shell, the same lifecycle, the same evidence discipline. Design decisions D1–D9 are signed in design #35; Phase 1 implements them.

The zones are an L, not nested boxes

Stage 1 is the 1.5–2× fold band at any absolute change, plus everything at or above 0.3 mg/dL below 1.5× fold. The unpainted box in the lower left is the one region where both KDIGO criteria are clear — and participants whose creatinine only fell keep their place on the chart below zero rather than being dropped.

Nephrotoxicity Explorer: creatinine scatter over L-shaped KDIGO stage zones with two triangular absolute-rule marks, data-handling notes above the chart, and a stage summary table below. 1 2 3 4
  1. 1 Nothing vanishes silently: plotted count, baseline-fallback count, and a CSV download beside the 24 dropped participants
  2. 2 The L: the ≥ 0.3 mg/dL arm reaches left of the 1.5× fold line; the clear box below it stays unpainted
  3. 3 Triangles are the ≥ 4.0 mg/dL rule — two sit inside the Stage-1 band and are Stage 3 anyway
  4. 4 The summary counts three ways: fold, absolute, and the combined KDIGO stage the zones show

The mark that overrides the plane

KDIGO's third Stage-3 criterion is a value, not a change: creatinine reaching 4.0 mg/dL. That cannot be a region of a change-vs-change plane, so it is a property of the mark instead (design D5) — larger, triangular, with its own tooltip line.

Tooltip on a triangular mark inside the Stage 1 zone: participant AKI-9027, KDIGO stage Stage 3, fold change 1.7× (Stage 1), absolute change +1.89 mg/dL (Stage 1), baseline 2.71, maximum 4.6 — Stage 3 by the absolute-value rule.

The case the mark exists for. AKI-9027 starts at 2.71 mg/dL — a high chronic-kidney-disease baseline — and rises 1.7×: Stage 1 on either change axis. But the maximum reached 4.6 mg/dL, so the participant is Stage 3 by the absolute-value rule, and the tooltip says which rule staged them. In the original renderer this participant would read as mild.

Dashes where KDIGO defines no stage

KDIGO stage summary table: fold change, absolute change and combined KDIGO stage columns; the absolute-change column shows dashes rather than zeroes for Stage 2 and Stage 3.

The first two column pairs are separate marginal distributions — the shape the R app's table has; the third is the combined stage the zones show. The absolute-change column's Stage 2 and 3 cells are dashes, not zeroes: KDIGO defines no such stages on absolute change, and a zero would read as "nobody qualified".

Why it matters

This is the portfolio's first renal display, and it ships with the data honesty the KDIGO criteria demand: per-record mg/dL ↔ µmol/L unit resolution (one unresolvable record suppresses absolute-change staging chart-wide rather than guessing), a baseline fallback that is counted and disclosed, and every dropped record downloadable with its reason.

It ships Experimental on purpose: the staging ladder is defensible from the KDIGO criteria and from the source's own chart geometry, but it should be confirmed by someone who owns the clinical content before the evidence page claims KDIGO conformance — that call is flagged in sv#121. Phase 2 (profile drill-down, CKD-EPI eGFR) is scoped in the design and unfiled.

Try it

Open the Nephrotoxicity Explorer
  1. Hover the two triangles inside the Stage-1 band and read which rule staged them.
  2. Filter Treatment Group to the AKI: arms — the injected cohort — and watch the counts re-derive.
  3. Click Download participants next to the dropped count and open the CSV: every row names its reason.
  4. Then read the evidence page — every NEP-* row traces to a named test.
sv#118 · hub#88

Study-day playback, with motion trails

The animation the original hep-explorer was best known for, restored to the eDISH scatter. Press play and every participant walks their own lab trajectory against the Hy's-Law quadrants, trails accumulating behind the moving points; the population's story stops being a single peak-values snapshot.

7 s clip · press play

Four drawing rules are ported verbatim from the original: a point sits on its most recent result at or before the shown day, holds at its first result before it is measured at all, shrinks outside its own measured span, and is not drawn before its participant's first record. At the end of the clip the slider is scrubbed — the playback yields to the reader rather than fighting for the day. That behaviour is HEP-ANIM-008, and it has its own story: it shipped implemented but unevidenced, and sv#119 closed that gap with a named browser test — under the done-gate, a requirement row is only as good as the evidence it points at.

Why it matters

A peak-values scatter cannot distinguish a participant who drifted into the Hy's-Law quadrant over twelve weeks from one who arrived in three days. Watching the walk — and the trail it leaves — is the fastest read of temporal pattern the display offers, and it was the loudest gap left from the v1.3 port.

On the public demo the day axis runs over the visit sequence, because the vendored dataset carries no study-day column — the label says so. With real ADY data the same control plays calendar study days.

Try it

Open hep-explorer, switch View to the eDISH scatter
  1. Press ▶ under the chart and watch the trails accumulate.
  2. While it plays, drag the day slider — playback stops and lands on your day.
  3. Change any control mid-play: the play-through stops rather than writing into a redrawn chart.
  4. Reset returns to peak values (all days).
sv#118 · hub#88

The P_ALT estimate, opt-in

Opt-in

With calculate_palt: true, the participant profile header carries an estimate of the fraction of hepatocytes lost — the quantitative companion to the eDISH read — with the arithmetic behind it shown, not just the figure.

Participant profile header showing the P_ALT hepatocyte-loss estimate alongside R Ratio, with the calculation's inputs displayed.

The canonical evidence capture (HEP-PALT-001, Linux CI environment). This one is not on the public demo, deliberately: the estimate integrates ALT over study day × 24 hours, and the demo dataset's visit sequence cannot stand in for that clock — so the demo leaves it off rather than showing a number built on the wrong axis. The evidence page carries the assertions (HEP-PALT-001…003), including that it stays off unless the caller opts in.

Why it matters

It is the first quantitative injury estimate in the portfolio, and its gating is the point: unit and sampling assumptions belong to the data owner, so the module refuses to guess them by default. The EX exposure half of sv#49 stays open for a dataset that can feed it.

Try it

Read the HEP-PALT evidence rows
  1. Find HEP-PALT-001–003 and open the capture.
  2. In your own call, pass calculate_palt: true with real study-day data.
sv#122 · hub#88

The eDISH axes finish their polish list

Four items migrated from the upstream backlog close out the v1.2 UI-polish list (sv#54): a log-base choice, manual axis limits, full measure names in the drill-down, and the R/nR primary sources in the Clinical guide.

Doublings, if that is how you read fold changes

The eDISH scatter on log axes with gridlines at 0.125, 0.25, 0.5, 1, 2, 4 and 8 — every doubling — with marginal box plots and rugs.

Axis Type log now offers a Log Base picker: log10 (decades) or log2 (doublings). Here the gridlines sit on every doubling — 0.25 / 0.5 / 1 / 2 / 4 / 8 ×ULN. The base is a tick generator, not a transform: log positions are base-independent, so choosing doublings moves the gridlines and never the cloud. A base whose powers cannot span the domain is not offered.

Limits you can see, names you can read

X-axis Limits and Y-axis Limits sections: Lower and Upper boxes pre-filled with 0/5.72 and 0/6.24, each with its own Reset Limits button.

Manual limits on both eDISH axes, and the boxes arrive filled with the limit actually in force — not blank. Type an upper bound and that axis follows while the other keeps re-deriving; Reset Limits hands one axis back to auto; changing the measure or Display Type returns both, since a limit typed for ALT ×ULN says nothing about ALP.

The participant rail for CLD-9045: R Ratio 3.08, a labs-over-time chart whose legend names each measure in full — Alanine Aminotransferase, Aspartate Aminotransferase, Bilirubin, Alkaline Phosphatase — with each line's short key written at its last point, and the per-measure summary table below.

The drill-down names its measures in full. Click a point, and the rail's labs chart legend reads Alanine Aminotransferase, not ALT — while each line still carries its short key at its own last point, in the line's colour, pushed apart when two lines end at the same height. The Clinical guide now links the R / nR primary sources and states the nR formula rather than only naming it.

Why it matters

Each item is small; together they close the list, and sv#54 — open since the v1.2 port — closes with them. The upstream asks these migrate from (#112, #238, #290, #335) date back years in the original repo.

Try it

Open hep-explorer on the eDISH view
  1. Set Axis Type to log; a Log Base picker appears. Choose doublings and watch the gridlines, not the cloud.
  2. Read the X/Y-axis Limits boxes before touching anything, then nudge one.
  3. Click a point and read the rail's legend and line-end keys.
notes

Reading this page

How the captures were made

  • Playwright drove Chromium against the release-candidate build — the site as built from the commit sv#124 promotes, byte-identical to the dev deploy — on 2026-08-14. Nothing was staged or mocked. The one exception is the P_ALT still, which is the canonical Linux evidence capture (HEP-PALT-001), because the public demo deliberately leaves that estimate off.
  • The clip is silent WebM, ~7 s, recorded during the real interaction. Stills are 2× JPEG.
  • The demo data is the shared adbds extract, now carrying two injected cohorts: 24 crafted CLD-* Hy's-Law cases for the hepatic views and 46 synthetic AKI-* participants for the KDIGO zones (design #35 D8). The nep demo plots 276 participants and reports 24 it could not plot, with reasons.

Also in the release, not shown here

  • HEP-ANIM-008 evidence (sv#119) — the scrub-stops-playback rule gains its named test; the clip above shows the behaviour itself.
  • Release prep (sv#123) — dist/safety.viz-1.5.0/ frozen back to the tag's bytes, dist/safety.viz-1.6.0/ vendored, fixtures repointed.
  • Every module's evidence.json was refreshed on the canonical Linux environment on 14 August as part of the nep-explorer landing.

Where this fits

This page is the review surface for RC sv#124 (dev → main). The user-facing release notes live in that PR's body and publish on the v1.6.0 tag when it merges — on your approval, via the attested lane, and not before.

The rest of the candidate

  1. The dev gallery — all twelve renderers
  2. nep-explorer evidence — requirement-keyed test records
  3. nep-explorer clinical guide — the KDIGO reading, stated