safety.viz release review 2026-10-03

What v1.9.0 adds, annotated

safety.viz opens its shared parts to a second chart library, and the demo app now carries that library’s four biomarker charts in a tab of their own, with R’s tests one press away. No existing chart, setting or API changes: what is new is the Biomarkers tab, R started on request, the chart list’s format version 2, and the kit those charts are built on.

17 of 17 charts ready on the pilot study R on request: 13.26 MB, once the kit: 36 parts, public from v1.9.0 single file: 1,165,774 bytes, no requests safety.viz dev at 096cc26

What to look at

  1. Open the Biomarkers tab: four charts, each ready, and the app counts 17 of 17 charts on the pilot study.
  2. Choose Alanine Aminotransferase in the group comparison before pressing anything: every visit’s line says statistics need R and what starting it costs, and nothing has been fetched.
  3. Press Start R. It reads Starting R…, then R started, and each visit prints R’s one-way ANOVA across the three arms: at Week 4, p = 0.005 for 75, 67 and 65 participants. Desktop R gives the same.
  4. Open the single-file build with the network off: the same tab draws, there is no Start R, and each line says why statistics are unavailable there.

Patch release v1.9.2

A patch on v1.9.1, and this note is its whole demo. Two things are new. The demo app can be installed and run on your own machine with one script. And the app’s biomarker charts are rebuilt on bio.viz v0.3.0 and gsm.bio v0.3.0, with the cross-tabulation as a fifth. No safety chart changes, and nothing in the library’s API changes.

In this note

Where it stands

  • It is released, on 2026-10-07: safety.viz v1.9.2. The owner approved the release candidate before it was merged to the main branch and tagged (safety.viz#218).
  • Where this note sends you to the demo app, it is to the released demo app. Its footer reads “safety.viz 1.9.2”.
  • The pictures here were taken from the dev site at dfbb4f4, one step before release preparation set the version; nothing a picture shows has changed since.
  • Captured on 2026-10-07 from the dev site, with real R running in the browser. A script beside this page holds 30 things this note says to the page that shows them, and all 30 held. Desktop R gives the same p-values and counts from the study’s own files.

1. Run the demo app on your own machine

One file, scripts/install-demo.mjs, puts the demo app on your own computer. It downloads safety.viz, builds the demo app and opens it in your browser with the pilot study loaded, served to that computer and to nothing else. In a clone you already have, npm run demo does the same. This is the task for a script that installs and runs the demo app (safety.viz#214), merged by its pull request (safety.viz#215).

The demo app served from this machine by the installer, on its Data view: the Pilot study chosen, four files loaded, the step Open a chart reading 18 of 18 charts ready, and the tabs reading Labs and vitals 9 of 9, ECG 1 of 1, Adverse events 3 of 3 and Biomarkers 5 of 5.

The demo app as the installer served it on a Mac for this page, at http://127.0.0.1:8711/, on the Data view: the pilot study’s four files loaded, 18 of 18 charts ready, Biomarkers 5 of 5. To load and to open the Biomarkers tab, the page asked that address for everything and no other host for anything.

$ node install-demo.mjs --ref dev --no-open --port 8711
Downloading safety.viz (dev) into /…/safety.viz…
Cloning into '/…/safety.viz'...
Installing what it needs to build (this can take a minute)…

added 129 packages in 1s
Starting the demo…
Building the demo app…

The safety.viz demo app is running at http://127.0.0.1:8711/
It answers this computer only. Files you load are read in your browser.
Press Ctrl+C to stop it.

What the installer printed on that run, made before the release and so with --ref dev, with the folder’s path shortened. The browser was left closed and the port named so that a script could open the page and take the picture above.

Try it

curl -fsSLO https://raw.githubusercontent.com/jwildfire/safety.viz/main/scripts/install-demo.mjs
node install-demo.mjs
  1. Check that you have Node.js 22 or later and git: node --version and git --version.
  2. In a folder of your choice, run the two lines above. In Windows PowerShell write curl.exe in place of curl.
  3. Wait for the line “The safety.viz demo app is running at http://127.0.0.1:8642/”. Your browser opens at that address. Here it took under a minute.
  4. Read the tabs: Biomarkers 5 of 5. Click Data: “18 of 18 charts ready”, on the Pilot study.
  5. Press Ctrl+C in the terminal. The demo stops, and the address no longer answers.
  6. Run node install-demo.mjs again, or cd safety.viz and npm run demo. It starts in a few seconds, with nothing downloaded or installed a second time.

What it does, in order

  1. It checks before it changes anything. Node.js 22 or later, git and npm must be there, and the folder it installs into must be absent, empty or already a safety.viz clone. If not, it says what it found and where to get what is missing, and stops with “Nothing was changed.”
  2. It downloads one release with git, without its history, into ./safety.viz. With no --ref that is the main branch, which holds the latest release.
  3. It installs the tools the release’s lock file names, with npm ci, inside that folder.
  4. It builds the demo app into build/demo inside that folder: the page, the app, the single-file build, every demo study, the typefaces, bio.viz’s bundle and gsm.bio’s statistics file.
  5. It serves that folder at http://127.0.0.1:8642/, or at the next free port if that one is busy, prints the address and opens it in your browser.

What it needs

  • Node.js 22 or later, which brings npm, and git.
  • A network connection, for the download from GitHub and the tools from npm.
  • About 285 MB of disk. Measured on the run above, one step before release preparation: 275 MB, of which 127 MB is build tools and 14 MB the built demo. Release preparation added the 1.9.2 bundle, about 10 MB.
  • That is the cost of building from source. A demo already built and attached to the release would need no build tools and no git, and is not in this release.

What it touches on your machine

  • One folder: safety.viz in the folder you ran it from, or the one you name with --dir. The download, the tools and the built demo are all inside it.
  • npm’s own cache of downloads in your home folder, as any npm install does.
  • One port on 127.0.0.1 while it runs.
  • Your browser, which it opens at the address unless you pass --no-open.
  • Nothing else. It installs nothing outside that folder and changes no setting, so the folder is the whole install.

Who can reach the page

  • The server listens on 127.0.0.1 only, so no other computer can open the page.
  • It refuses a request that names any host but 127.0.0.1 or localhost, so a page on another site cannot read the demo through your browser.
  • It only reads. It gives no file from outside the built demo’s folder, and it accepts nothing sent to it.
  • Files you load are read in your browser. The one thing the page asks the network for is R, from webr.r-wasm.org, and only when you press Start R.

Stopping, starting again, and the options

  • Ctrl+C stops it. Stopping the installer stops the demo with it and frees the port.
  • Run it again to start it again. A clone already in the folder is used as it is. Run again with the same --ref it was installed with, the installer uses the install and starts the demo, a fault the release review found and its task fixed before the release (safety.viz#219).
  • An install is never moved to another release. Give a later release a folder of its own: --ref v1.9.3 --dir safety.viz-1.9.3.
  • --port names the port, --no-open leaves the browser alone, and --help lists every option.
  • If the download is cut off, the installer stops with git’s own message and leaves nothing behind. Run it again. That happened once while this page was made.

What was tried, and what was not

  • macOS with Node 24: the installer was run end to end from an empty folder, for its pull request and again for this page on 2026-10-07. Stopped with an interrupt, it freed the port. Run again, it started without downloading or installing anything.
  • Linux with Node 22: the test runner. The installer’s tests run the file itself against small made-up releases, with nothing installed from the network. A browser test starts the demo command, reads the address it prints, and holds the page to the pilot study with every chart supported, R starting on request and a clean stop.
  • Windows: written for, and not yet tried. There the installer starts npm through a shell, and the demo opens the browser with cmd /c start.
  • A Node older than 22: the file is written so that an old Node can read it and be told it is too old. It has not been run on one.
  • Ten behaviours were broken on purpose, and each failed its test. Among them: the server answering on every network interface, a path climbing out of the folder, and a release named over an existing install.
  • An independent reviewer read the first version and found three faults, each fixed with a test: a named release was ignored once a clone existed, every run installed again, and stopping the installer alone left the demo running.

2. The biomarker charts, rebuilt on bio.viz v0.3.0 and gsm.bio v0.3.0

In the demo app’s Biomarkers tab, the group comparison opens on a trend tile for every biomarker. A tile opens that biomarker across the visits, with R’s test under each visit once R is started. A visit opens alone, and a trail leads back. The cross-tabulation joins as a fifth biomarker chart. This is the requirement that the group comparison opens on trends over time and drills down to one visit (obot.roadmap#367), brought to the app by the task to rebuild it on bio.viz v0.3.0 (safety.viz#212) and its pull request (safety.viz#213).

The demo app’s Biomarkers tab, reading 5 of 5, with five charts beneath the tabs: Group comparison, Association scatter, Correlation matrix, Biomarker screen and Cross-tabulation. The group comparison is open on a grid of small tiles, one per biomarker, each with a blue, a green and an orange line from Baseline to Week 26 and its own value range printed beneath. Above the tiles: 9 unscheduled visits not drawn.

The Biomarkers tab on the pilot study: 5 of 5, and the group comparison open on 28 trend tiles, a line for each arm through its median at every scheduled visit. No statistic is printed, and R has been asked for nothing. The first line says 9 unscheduled visits are not drawn.

Alanine Aminotransferase over time after R was started: box plots for the three arms at ten visits from Baseline to Week 26, and beneath them a table with the number in each arm at each visit and a row headed One-way ANOVA reading p = 0.515, 0.127, 0.005, 0.037, 0.013, 0.380, 0.170, 0.050, 0.517 and 0.252. Above the chart a trail reads All biomarkers, then Alanine Aminotransferase over time.

The Alanine Aminotransferase tile, opened, after Start R: ten visits, the number in each arm, and R’s one-way ANOVA under each visit. The trail above the chart leads back to the tiles.

Week 4 opened alone: the trail reads All biomarkers, Alanine Aminotransferase over time, Week 4; the Visit control reads 1 of 10; three boxes for Placebo n = 75, Xanomeline High Dose n = 67 and Xanomeline Low Dose n = 65; and under the chart, One-way analysis of variance: p = 0.005.

Week 4, clicked under the picture, opens alone. The Visit control reads “1 of 10”, and the line under the chart is R’s test of that one visit.

The cross-tabulation in the app: a table of ARM by EOSSTT with counts 58 and 28 for Placebo, 27 and 45 for Xanomeline High Dose, 25 and 71 for Xanomeline Low Dose, total 254, each with its percentage of the row; the same table as stacked bars; and beneath, Pearson's Chi-squared test: p < 0.001 (n = 254).

The cross-tabulation, the fifth biomarker chart: treatment arm by end-of-study status, the same table as bars, and R’s chi-square test beneath.

The trend tiles at 390 pixels wide: two tiles to a row, beneath the note about unscheduled visits and the key to the three arms.

At 390 px: the tiles, two to a row.

Alanine Aminotransferase over time at 390 pixels wide after R started: the box plots and the table beneath them, showing Baseline to Week 8 with the later visits off to the right, and the row of p-values starting 0.515, 0.127, 0.005.

At 390 px: one biomarker over time. The picture and its table scroll sideways inside their own box, and the page does not.

Try it

Open the released demo app
  1. Click the Biomarkers tab. It reads 5 of 5: Group comparison, Association scatter, Correlation matrix, Biomarker screen, Cross-tabulation. On the Data view the app counts 18 of 18 charts.
  2. The group comparison opens on 28 trend tiles, a line for each arm and no statistic. The first line says 9 unscheduled visits are not drawn.
  3. Click the Alanine Aminotransferase tile. Ten visits are drawn, Baseline to Week 26, and the One-way ANOVA row reads “Statistics unavailable”.
  4. Press Start R, at the left beneath the tabs. It downloads about 13 MB, once, from webr.r-wasm.org, and the study’s data stays in the browser. The row then reads p = 0.515, 0.127, 0.005, 0.037, 0.013, 0.380, 0.170, 0.050, 0.517, 0.252.
  5. Click Week 4 under the picture. That visit opens alone: the Visit control reads “1 of 10”, and the line reads “One-way analysis of variance: p = 0.005 (Placebo n = 75, Xanomeline High Dose n = 67, Xanomeline Low Dose n = 65)”.
  6. Read the trail above the chart: All biomarkers › Alanine Aminotransferase over time › Week 4. Click a step to go back to it.
  7. Open Cross-tabulation. ARM by EOSSTT counts 58 and 28, 27 and 45, 25 and 71, total 254, and the line reads “Pearson's Chi-squared test: p < 0.001 (n = 254)”.
  8. Under Test, choose Fisher’s exact test. The line reads “Fisher's Exact Test for Count Data: p < 0.001 (n = 254)”.

The detail the release notes leave out

  • The copies: bio.viz’s bundle and gsm.bio’s statistics file are copied from their release tags, bio.viz v0.3.0 and gsm.bio v0.3.0. The test runner holds each copy to its tag, byte for byte.
  • How the three levels work, their settings and what R is asked at each are on the demo page of those two releases: the tiles, one biomarker over time and one visit. This note shows them only as the demo app draws them on the pilot study.
  • Before R: the tiles ask R for nothing. Over time, the test row reads “Statistics unavailable” and the line under the chart says what starting R costs. No request had gone to R’s host before the press.
  • Starting R, measured here: 13.25 MB in 7 requests, all to webr.r-wasm.org, once. Every biomarker chart then uses that one R. Opening the visit, the cross-tabulation and the biomarker screen fetched nothing more from it.
  • The numbers: desktop R, with base R alone on the study’s own files, gives the same ten p-values with the same number in each arm, and the same table and both tests for the cross-tabulation. The script and its answers are beside this page.
  • The ten p-values are also the ones in the table in section 02 below. The test and the data have not changed, only where the chart prints them.
  • Who is counted: the subject-level file has 254 participants. The labs file also carries 110 synthetic participants who are in no other file, and the biomarker charts leave them out and say so.
  • The single-file build, opened with the network off, draws every chart and each level of the group comparison and sends no request. A browser test holds it to that.
  • What did not move: no safety chart’s source, bundle, demo page or data is in the change. The app’s count goes from 17 charts to 18, and from four biomarker charts to five.
  • Tests: 2,242 unit tests and 385 browser tests pass. One browser test walks this tab with real R started in the browser and holds each of R’s answers to desktop R’s on the same rows.

Not in the release

  • The group comparison’s difference grid, a second opening view, is paused and has a requirement of its own (obot.roadmap#371). bio.viz v0.3.0 does not ship it, so the app does not draw it and asks R for nothing for it.
  • bio.viz’s sixth chart, the stratified survival chart, is in the copied bundle and is not listed in the app. It reads an outcomes table, which the app’s chart list cannot name yet (bio.viz#63).

Known, and not fixed

The Biomarker screen’s Compare control in the app, set to Difference between two groups, with a sentence beneath it: A hazard ratio needs an outcomes table: give outcomes, one row per participant and endpoint, with a time and a flag, as init({ results, participants, outcomes }).
  • On the Biomarker screen, under Compare, the chart prints a sentence telling the reader to give an outcomes table by calling init(). A reader of the app cannot do that. It has been bio.viz’s wording since v0.2.0, and rewording it is in the backlog (bio.viz#119).
  • On a phone, the row labels of the table over time break inside a word, as the second phone picture above shows. It was seen while this page was captured, and it is in bio.viz’s backlog (bio.viz#120).
  • The installer has not been tried on Windows, as said above.

Where to look further

Patch release v1.9.1

A patch on everything below. Nothing is removed, and no chart’s drawing changes except the histogram’s, which now draws its two deprecation notes.

  • The demo app’s footer says what happens to the data you load: files are read in the browser and never uploaded, and starting R downloads R from webr.r-wasm.org while your data stays in the browser. The single file says it loads nothing (safety.viz#196).
  • The histogram’s two JavaScript p-values, the normality screen and the group comparison, are deprecated. They still work, and the chart says so. v1.10.0 removes them, so remove test_normality and compare_distributions from your settings; nothing in safety.viz replaces them (safety.viz#188).
  • A library chart named data is refused, because the app keeps that name for its data view (safety.viz#197).
  • Each requirement ID now names one requirement, and every evidence page shows its own requirement text (safety.viz#195).
  • Evidence screenshots can be refreshed by name, and the obot app can start that refresh itself (safety.viz#204, safety.viz#207).
01 · the Biomarkers tab

The Biomarkers tab

bio.viz’s four charts sit beside Labs and vitals, ECG and Adverse events: the group comparison, the association scatter, the correlation matrix and the biomarker screen. They read the labs and vitals file as their results and the subject-level file as their participants, through the mapping already made, so a study mapped for the safety charts is mapped for these.

The demo app on the pilot study with the Biomarkers tab open, reading 4 of 4. Beneath the tabs, a Start R control with the hint About 13 MB, once, and the four biomarker charts; the group comparison overview shows Alanine Aminotransferase across the visits, by arm.

The Biomarkers tab on the pilot study. The tabs read 9 of 9, 1 of 1, 3 of 3 and 4 of 4, and the Data view says 17 of 17 charts ready. The group comparison opens on an overview of every measure at every visit, which prints no test: “12 of 28 biomarkers shown”.

What you should see

  • Biomarkers 4 of 4, each chart marked ready.
  • On the Data view: 17 of 17 charts ready.
  • The group comparison overview: 12 of 28 biomarkers on page 1 of 3.

Try it

Open the demo app
  1. Click the Biomarkers tab.
  2. Open each of the four charts from the row beneath the tabs.
  3. On the Data view, choose the renamed-column study: the one mapping readies both libraries’ charts.
02 · R on request

R on request

The biomarker charts print statistics only from R. Until you ask for R, every biomarker chart draws and its statistics line says what starting R costs, and the page fetches nothing. One press starts R in your browser, once, from webR’s public host with gsm.bio’s statistics functions, and every biomarker chart uses that one R from then on.

The group comparison with Alanine Aminotransferase chosen, before R is started: panels for Baseline, Unscheduled 1.1 and 1.2 with box plots by arm, and under each the line Statistics need R. Start R to compute them: it downloads about 13 MB, once, from webr.r-wasm.org, and the study’s data stays in this browser.

Alanine Aminotransferase chosen before the press. Under each visit: “Statistics need R. Start R to compute them: it downloads about 13 MB, once, from webr.r-wasm.org, and the study’s data stays in this browser.” No request had gone to R’s host at this point.

The app header just after the press: the control reads Starting R…

Just after the press: Starting R…

The app header once R has answered: the control reads R started.

1.5 seconds later, here: R started.

The same chart after R started: the Week 4 panel prints One-way analysis of variance, p = 0.005, Placebo n = 75, Xanomeline High Dose n = 67, Xanomeline Low Dose n = 65; Week 6 prints p = 0.037; Unscheduled 5.1 says there is no test because only one arm has values; Unscheduled 7.1 prints R’s reason for not computing, a minimum group size of 5.

After the press, the chart keeps what you chose and prints R’s test under each visit. Where a visit has too few participants, it prints R’s reason, not a number.

VisitOne-way ANOVA, across the three armsn: Placebo, High Dose, Low Dose
Baselinep = 0.51576, 66, 88
Week 2p = 0.12782, 70, 73
Week 4p = 0.00575, 67, 65
Week 6p = 0.03771, 55, 58
Week 8p = 0.01370, 52, 54
Week 12p = 0.38066, 42, 40
Week 16p = 0.17067, 33, 32
Week 20p = 0.05061, 29, 26
Week 24p = 0.51757, 28, 24
Week 26p = 0.25250, 24, 22

Alanine Aminotransferase by arm on the pilot study, as the app printed them, each labelled exploratory and unadjusted. Desktop R, running the same gsm.bio statistics file on the app’s own data files, gives the same p-value and counts at every visit. Unscheduled visits with fewer than five in a group print R’s reason instead.

The biomarker screen in the app after R started: Result at Baseline, Placebo against Xanomeline High Dose, twenty rows sorted by estimate, Platelet first at 0.3404 with p = 0.042 unadjusted and 0.295 after Benjamini-Hochberg.

Another biomarker chart, the screen, on the same R with no second download: Placebo against Xanomeline High Dose at Baseline across 28 biomarkers. Platelet is first at 0.3404 (0.01 to 0.6696), p = 0.042 before adjustment and 0.295 after. No row survives the adjustment.

The app at 390 pixels wide with the Biomarkers tab open: the Start R control with About 13 MB, once, beside the Group comparison chip, and the chart’s notes beneath.

At 390 px, before the press.

At 390 pixels wide after R started: the Baseline panel with its three boxes and R’s one-way ANOVA, p = 0.515.

At 390 px, after it: R’s test at Baseline. Neither page scrolls sideways.

What you should see

  • Before the press: no request to webr.r-wasm.org.
  • The press: 13.26 MB in 7 requests, all to webr.r-wasm.org, here in 1.5 seconds to R started. Opening the screen afterwards fetched nothing more.
  • Week 4: p = 0.005 (75, 67, 65).

Try it

Open the group comparison in the app
  1. Set Biomarker to Alanine Aminotransferase and read a visit’s line.
  2. Press Start R, at the left beneath the tabs.
  3. Switch Test to Kruskal-Wallis: each line waits for R and prints the new answer.
  4. Open the biomarker screen: it uses the R that is already running.

The single file, offline

The single-file build carries bio.viz’s charts inline and still loads nothing, so it cannot start R. It has no Start R control, and its biomarker charts say why statistics are unavailable.

The single-file build opened from disk with the pilot study’s four files dropped on it: the Biomarkers tab reads 4 of 4, and beneath the tabs the four biomarker charts with no Start R control.

The single file, opened from disk with the network off and the pilot study’s four files dropped on it: the Biomarkers tab, and no Start R.

The single file’s group comparison with Alanine Aminotransferase chosen: the same box plots, and under each visit the line Statistics are unavailable in this file: it loads nothing, so it cannot start R. The hosted demo app can start R in your browser.

“Statistics are unavailable in this file: it loads nothing, so it cannot start R. The hosted demo app can start R in your browser.” The file is 1,165,774 bytes and made no request.

03 · a second library joins

A second library joins the app

A page that mounts the demo app can hand it another chart library: its chart factories and its own chart list. The app lists those charts in a tab of their own, gives each a status, counts them, builds their settings and data from the loaded files and the mapping, and draws one at a time, as it does its own. To describe charts like these, the chart list’s format goes to version 2: an entry can name its library, the group it is listed under, and the named tables its chart takes. The thirteen safety charts’ entries mean what they meant.

The Domains page: the standard domain set of four tables, with a summary strip reading 17 charts read the standard set and the manifest described in format version 2, then the subject-level domain, whose list of charts that add it when supplied names the four biomarker charts.

The Domains page: 17 charts read the standard set, from a manifest in format version 2. The subject-level domain now lists the four biomarker charts among those that use it when it is supplied.

The Domains page’s Biomarkers, from bio.viz section: four charts from a second library, each with links to its live demo and API reference, the tables it takes and a table of its column settings with their defaults.

A section of its own, “Biomarkers, from bio.viz”: each chart with the tables it takes, its column settings and their defaults, and links to its demo and reference on bio.viz’s site.

Why it matters

Before this, the app knew only its own charts. Now any library that publishes a chart list in this format can join it, on the same files and mapping, without the app being rebuilt around it.

Try it

Open the Domains page
  1. Find “Biomarkers, from bio.viz” near the end.
  2. Follow a chart’s API reference link to bio.viz’s site.
04 · the kit

The kit

SafetyViz.kit hands out the parts every safety.viz chart is built from, so a library loaded beside it builds charts that look and behave the same, without a copy of that code drifting away: the control sidebar and its builders, the filter contract, the axis-limit helpers, the record listing, the participant rail, the box drawing, the measure list, the Kaplan–Meier estimator and the Chart.js the bundle already contains. Each member is the very function the charts call. bio.viz’s four charts are built on it.

The kit API reference on the dev site: 36 members from 8 shared modules and Chart.js, reached as SafetyViz.kit, drawing with Chart.js 4.5.1, public surface from v1.9.0, with a contents list of the member groups and a snippet loading safety.viz before another library.

The kit’s reference: 36 members from 8 shared modules and Chart.js 4.5.1, reached as SafetyViz.kit.

The one commitment in this release

From v1.9.0 the kit is public surface: a change to any member’s name, signature, return value, or the elements and class names it produces is a breaking change. Adding a member is not. The kit, explained in one page, and its API reference.

before you approve

What is not in this release

  • R in the single file: it loads nothing, so its biomarker charts say statistics are unavailable.
  • A library other than bio.viz in the app: the seam is general, and bio.viz is the only one using it.

Where to look further