Catalogue
| ID | Title | Summary | Practice | Status |
|---|---|---|---|---|
| ND-001 | Null Operator The studio's own operating layer: a lead agent decomposes work into narrow, test-specified jobs, dispatches them to interchangeable worker models (OpenCode/GLM, Codex, a self-hosted agent), gates every result behind tests and a human decision, and records provenance for each run. | |||
| ND-002 | Qlass Self-hostable, open-source learning management infrastructure (Next.js, PostgreSQL, Prisma) with a teacher-gated agentic unit builder: plan → author → review, every run logged to an audit table, nothing published without a human. A candidate learning environment for Agentic Education. | |||
| ND-003 | Atlas Curriculum Audit An agent skill that reconciles a school's gradebook records against its curriculum-mapping system and a published course guide: build the assignment inventory, classify evidence, correct misplacements, verify every unit — with a human confirming every destructive step. Practical agentic orchestration across educational systems of record. | |||
| ND-004 | arXiv Weekly Autonomous weekly research infrastructure: search 21 arXiv categories, enrich with citation data, filter 100 candidates to 5–10, synthesise a briefing into a knowledge vault, write a script, produce audio, publish an RSS podcast feed. Running unattended every Monday since June 2026. | |||
| ND-005 | MDSolids Web An open-source, browser-based, clean-room reimplementation of the classic mechanics-of-materials teaching suite: section properties, axial, torsion, beams, trusses, columns, Mohr's circle, combined loading, pressure vessels. Built by a human directing a fleet of agents against a shared module contract. | |||
| ND-006 | Stock Forecast in the Browser A price forecaster that runs entirely in the browser: a two-layer LSTM, geometric Brownian motion with Monte Carlo confidence bands, a decision-tree regressor over engineered indicators, and a disagreement-widened ensemble. Built as a research exemplar for a secondary-school research course. | |||
| ND-007 | Political Polarity An NLP pipeline for measuring political lean in text corpora: a TF-IDF + logistic-regression baseline, a fine-tuned DistilBERT path, synthetic-data generation, and a model card stating out-of-scope uses. A study in what a single researcher can now run that once required a team. | |||
| ND-008 | Six Models, One Lesson Six model routes given the identical lesson-design brief (an AI-literacy lesson for a statistics course), scored on a fixed rubric, with the route of invocation recorded and a disclosed conflict: the reviewer graded its own submission. The first Agentic Education experiment (AE-001). |
| ID | Title | Summary | Practice | Status |
|---|---|---|---|---|
| ND-009 | Course Operations Compiler Treat syllabi, block plans, warm-ups, exit tickets and slide decks as compiled artifacts: single-source calendar and course specs, deterministic generators, a tested date library, an agent skill for deck authoring, and programmatic QA for layout overflow. Architecture described publicly; repositories stay private. | |||
| ND-010 | Approval Gate for Consequential Agent Actions A small service and two automation flows that let an agent propose actions near money without being trusted: position caps, a falsifiable thesis required for every proposal, a push-notification approval loop, and a 24-hour auto-reject. The pattern is public; the data is not. | |||
| ND-011 | Knowledge Base as Agent Memory A git-synced Markdown vault embedded into pgvector on a schedule and exposed to agents as a single read-only search tool. Long-term memory that humans can read, edit and version — and agents can only query. |