--- title: Mathematical Sight-Reading emoji: 🎼 colorFrom: indigo colorTo: purple sdk: gradio sdk_version: 6.29.0 python_version: "3.11" app_file: app.py pinned: false hf_oauth: true short_description: Drill reading dense maths notation like sheet music --- # Mathematical Sight-Reading Framework (MSRF) Train mathematical literacy by treating notation like sheet music: decode the symbols, explain the structure, then ground it in more than one domain. You get a **naked formula** (no name, no context). You answer in three parts (formal sight-reading, concept and mechanics, application and metaphor). A judge model scores **Precision** (0-10), **Concept** (Superficial / Competent / Exemplary) and **Transfer** (A / B / C), then shows what you missed and an exemplary sight-read. ## How it works - **Formula bank** (`bank.py`): 16 formulas over four tiers, each with a hidden reference sight-read. The judge grades against that reference, not its own memory, which is what keeps scoring stable. Add formulas by appending to the list. - **Adaptive tiers** (`engine.py`): rolling mean of your composite score moves you up or down a tier. Plain code, no LLM involved. - **Spaced repetition**: poorly answered formulas return after a few drills (Leitner boxes). - **Hint ladder**: hint 1 = types, hint 2 = mechanics. Each costs 0.1 of the composite. - **Weak-spot report and export/import** on the Progress tab. - **Judge** (`judge.py`): your own OpenRouter key, any model id. Default set with the `JUDGE_MODEL` variable; dropdown choices with `JUDGE_MODELS` (comma separated). ## Privacy Your key, progress and model choice live in your browser's localStorage. The key passes through the Space server in memory only to make the OpenRouter call; it is not logged or stored. ## Tests ``` pip install gradio pytest playwright itsdangerous authlib python -m pytest tests/test_core.py python tests/test_ui.py # browser test against a fake OpenRouter ```