Public planning explanation: first copy revision
Date: 2026-09-27
Scope
Rewrite the English homepage passages represented in the supplied screenshots and add a paired German/English introduction. Explain the appointment problem before introducing CPIR, candidates, comparison fields, half-open intervals and whole-second measurements. Keep the proposal/execution distinction and the limits of ProvenOptimal explicit.
The initial revision introduced the learning-project context and linked the 1986 Rumelhart, Hinton and Williams paper. The maintainer then supplied the missing personal account. The follow-up below incorporates it without inventing biographical details.
The English homepage links to the German introduction. Both introductions link to each other and are listed in the existing documentation indexes. This is bilingual explanatory content, not a site-wide localization or a German interactive homepage.
Workspace and boundaries
The canonical workspace and active codex/slice-2-evaluation-replay branch were retained
at aa86b82. Fetched origin/main was af62b82; its later visual continuity changes were
not merged into this unrelated evaluation checkpoint. Existing .gitignore changes and
untracked root test results were preserved.
Only public prose, labels, links, corresponding existing test assertions and documentation were changed. CSS, animation, domain calculations, generated runtime inputs, Rust source, versions and release state were not changed. The generated site still uses Rust results. This record supplies implementation evidence, not independent Architecture, Math or Security qualification.
Verification
npm run check: passed, including all 22 version-authority tests, Markdown links, DE/EN counterparts and repository boundaries.npm --prefix apps/cerebri-site run check: passed with zero errors and warnings.npm --prefix apps/cerebri-site run build: passed; Rust-generated examples were rebuilt.- All 30 existing non-visual Playwright tests passed after updating changed text assertions. Existing checks for candidate counts, ordering, lifecycle states, keyboard controls, reduced motion, accessibility, enlarged text and responsive widths remain in place.
- Both new introduction routes were also checked in Chromium at 1440 and 390 pixels: correct article language, reciprocal language links, expected concepts and no horizontal document overflow. Desktop and mobile screenshots were inspected.
- All 10 visual tests ran and reported differences against the prior Windows hashes. The rendered reference captures were inspected, but no new baseline hashes were accepted for this provisional text revision. Linux references were not regenerated or changed.
git diff --check: passed.
The initial browser command encountered an occupied port 4173. The successful browser run imported the same Playwright configuration through a temporary ignored wrapper and used port 4175 with a separate output directory. No assertions or thresholds were disabled. A separate local preview runs on port 4177.
No commit, push, pull request, deployment or remote CI run was performed for this revision. Visual baseline acceptance remains open before publication.
Founder-story follow-up
The maintainer supplied a fuller account of the project's origin on the same day. The paired introductions and the existing homepage finale now connect the neural-network mathematics that sparked curiosity, understanding equations one operation at a time, learning through implementation and tests, and planning as a concrete problem to study. The opening also identifies Cerebri as a learning project rather than reducing its purpose to appointment scheduling.
The story uses the maintainer's first-person perspective. The 1986 paper is credited to David E. Rumelhart, Geoffrey E. Hinton and Ronald J. Williams and linked to the publisher. The broader aim of planning across applications is explicitly future-facing, consistent with Master Specification 0.4, section 1. Current single-target deterministic planning stays distinct from future learned assistance. No unsupported personal anecdotes or current neural-runtime capabilities were added.
The supplied handoff files remain private source material outside the repository. Their broader design, implementation and publication instructions do not expand this text-only task.
After the follow-up, repository checks, the Svelte check and the production build passed again. All 30 existing non-visual Playwright tests passed. Chromium checks confirmed both language links navigate to their counterpart, the article language is correct, and neither introduction nor the homepage overflows horizontally at 1440 or 390 pixels. The updated story passages were captured and inspected at both widths. Visual baseline hashes were not rerun or updated for this follow-up; their earlier mismatch remains unresolved. The changes remain local and unpublished.