Break the Wall v0.2.0-alpha

What's Changed

• fix: readme file change by @lucariggi92 in #25
• Record the witnessed install and complete the macOS app plan by @simoneriggi92 in #26
• A Windows alpha: CI, platform-aware wording, x64 installer and docs by @simoneriggi92 in #27

New Contributors

• @lucariggi92 made their first contribution in #25

Full Changelog: v0.1.1-alpha...v0.2.0-alpha

Break the Wall v0.1.1-alpha

What's Changed

• Show the whole loop in the README demo by @simoneriggi92 in #22
• Record the v0.1.0-alpha pre-release in the plan by @simoneriggi92 in #21
• Package the app as an ad-hoc signed DMG for Apple Silicon by @simoneriggi92 in #23
• Build the DMG in CI and attach it to the tag's release by @simoneriggi92 in #24

Full Changelog: v0.1.0-alpha...v0.1.1-alpha

Break the Wall 0.1.0-alpha

Take the text apart. AI gives you a wall of text. Break the Wall turns it into source-linked parts you can explore, test and put back together.

This is a pre-release that runs from source on macOS. There is no installer or binary; the prepared examples work without an account.

Try it

git clone https://github.com/simoneriggi92/BreakTheWall.git
cd BreakTheWall
git checkout v0.1.0-alpha
npm ci
npm run build
npm start

Requirements: macOS, Node.js 22.23.2 and npm 10.9.8. The first start downloads Electron's binary.

What's in it

Identity

• A new identity: the name stays Break the Wall, with the tagline "Take the text apart." / "Smonta il testo." and a mark of four lines of text, one piece pulled out in terracotta and still linked to its place.
• The mark and tagline in the header of both themes, the home's new opening, and the app icon in the Dock. Paper and blueprint keep their colours; terracotta appears only in the logo.
• Brand assets in docs/images/brand/; npm run brand renders the app icon and the social card from them.

Disassemblies

• A disassembly format with typed joints for three kinds of material (mechanism, argument, story), sub-assemblies as deep as the source allows, and a bill of materials that accounts for every line of the source. The host checks every file before showing it.
• An automatic layout: mechanisms and stories flow left to right; arguments stack the claim above its grounds, with attacks and limits beside it. Part numbers follow the reading order.
• The player: exploded view with a part card (a short lead, then marked points), runs (scenarios with a prediction before each step, load tests that pull out one brick at a time, new cases), reassembly on a blank drawing, and the bill of materials.
• Resizable drawing, card and source panes; a way back to the level above with ↑, Escape or Alt+↑.
• Sources read as numbered lines or as a rendered document with tables and Mermaid diagrams, where citations highlight their blocks.
• Paper and blueprint themes, and an English and Italian interface. Progress, runs and notes resume after a restart.
• Two prepared examples, each in English and Italian: an original release runbook (mechanism) and Mill's harm principle (argument, public domain).

Generation

• Studio takes apart pasted text or .txt / .md files around the learner's question, in three phases: structure with the bill of materials, one request per part with only its lines, then runs and reassembly hints. Each phase is checked and repaired at most once; a repair that leaves only over-long texts gets one small shorten request.
• Suggested break-downs on request: 3–5 questions with the kind and parts they would give.
• The top level is drawn while it is written, with each part's status.
• A failed or cancelled generation keeps what was accepted and can be resumed.
• Any part with enough lines can be taken apart further on demand, as a new revision of the disassembly.
• Two routes: Sign in with ChatGPT (experimental, uses the learner's plan), or a local model server on this Mac (llama.cpp, LM Studio, Ollama or an OpenAI-compatible server).
• npm run eval measures generation on fixed public cases.

Earlier lessons

• Lessons generated before disassemblies stay readable, with their progress, under "Earlier paths".

Known limits

• Explanations are generated and can be wrong: the app checks citations and source coverage, not truth.
• Sign in with ChatGPT is experimental; local models are measured on lessons, not yet on disassemblies.
• PDF documents and repositories as sources are planned, not available.

Feedback is welcome in Discussions and issues. Full history: CHANGELOG.md.