bloom-testing-inputs

Test input collections and books for Bloom, plus their reference baselines (e.g. visual-regression screenshots).

This repo holds the inputs for all Bloom testing — both the automated suites (nightly edge-to-edge and visual-regression tests in BloomDesktop) and tests that are still executed manually. Test code does not live here; it lives in BloomDesktop, versioned with the product code it tests. This repo is data.

How BloomDesktop consumes this repo

BloomDesktop pins an exact commit of this repo in build/testing-inputs.pin and fetches it with build/get-testing-inputs.mjs into its gitignored output/testing-inputs/ folder. That means:

• Changing this repo breaks nothing until a BloomDesktop branch advances its pin.
• Each BloomDesktop branch (master, Version6.5, …) pins the commit that matches it.
• To ship a data change: merge it here, then open a BloomDesktop PR that updates the pin (and any test expectations) together.

Layout

collections/
  <collection-name>/
    <collection-name>.bloomCollection
    <Book Title>/
      <Book Title>.htm          # the book DOM (validated by CI)
      meta.json, appearance.json, publish-settings.json, images…
      screenshots/*-reference.png   # committed baselines, when the owning test uses them
manifest.json                   # required entry per collection: purpose, usedBy, automation
scripts/                        # validation run by CI and by `pnpm validate`

Authoring rules

• Author books with Bloom itself, then copy the folder in. Hand-editing the .htm invites invalid DOM; CI runs scripts/validateBloomBook.mjs (vendored from BloomDesktop's edit-bloom-book skill) against every book.
• Commit the rendering-relevant CSS Bloom regenerates (branding.css, appearance.css, defaultLangStyles.css, origami.css, langVisibility.css), captured under the Default branding. These snapshots are forensic evidence: when a reference screenshot later breaks, their diff can name the cause. Refresh them whenever you refresh reference screenshots. Run artifacts (thumbnails, coverImage caches, history.db, branding logo copies) stay uncommitted; the .gitignore explains the split.
• Add a manifest.json entry for every new collection: what it is for, which test suite (or manual test) uses it, and "automation": "automated" | "manual".
• Stay inside the size budgets: 10 MB per file, 50 MB per book folder (CI-enforced; a collection can override in its manifest entry, with justification in the PR). This repo is plain git, no LFS; keep media small (shrink images, trim audio/video).
• Keep content redistributable. Everything here is public. Only include images, audio, and text that we have the right to publish (CC-licensed or team-created).

Running validation locally

pnpm install
pnpm validate