Case study · educational validation pending
This page keeps William’s recorded decisions, AI-assisted implementation, simulation output, automated checks, and the learner study that has not been run. It does not report learning gains.
The code-native overview is the repository fallback for the editable product map. It keeps the four-Lab sequence and the study gate visible even when Figma access is unavailable.
A beginner can meet praxeology as a list of conclusions. The intended experience is plainer: one ordinary situation, a response the learner controls, a visible consequence, and a private note of what changed. The path starts at the first lesson, then the Choice Machine, and keeps the market lab easy to find.
The repository records that William conceived and directed Praxeos. He chose the progression from individual choice to exchange, entrepreneurial discovery, and money across time; made Market Without a Manager the featured demonstration; kept the warm editorial identity; and separated checklist checks from optional guidance. A private motive story is not in the project record, so this page does not invent one.
The earlier product scattered attention across Monetary Garden, Signal Orchard, Calculation Labyrinth, and Coordination Engine. It also carried static motion fallbacks, answer-shaped analyzer feedback, and unreachable preview code. Those approaches were retired after their useful sources and design lessons were preserved.
Codex assisted with implementation, fixtures, documentation, and verification under William’s product direction. Claude is an optional in-product guide: it can ask about reasoning using a fixed list of cited sources, and it cannot score a conclusion.
Next.js renders the learning system. Four pure TypeScript engines create replayable state from a seed, assumptions, and action log. One versioned local browser store holds lessons and Lab records. Only the bounded Guide endpoint is dynamic; no account or transcript database is required.
Every Lab uses semantic HTML and SVG, large controls, visible focus, structured state, live change summaries, and the same complete interaction under reduced motion. Mobile focuses the current action and turns secondary evidence into a readable sequence.
Breadth originally arrived before a clear first action. Visual fallback was mistaken for educational equivalence, and keyword-shaped feedback risked pretending to understand prose. Local-only records improve privacy and simplicity but do not sync across devices.
Automated results and remaining blockers are recorded in the build report. Technical checks can verify deterministic engines, privacy boundaries, keyboard flows, rendering, and performance. They cannot verify semantic understanding or educational impact.
90-second demonstration
Open Market Without a Manager. Meet five participants, inspect their ranked priorities, attempt the first barter offer, introduce money, and compare the completed-trade price with the still-unmet wants. Then show the structured market record and the assumption labels. The complete 8–10 minute journey continues through incomplete information, shortage, a price ceiling, interpretation, revision, Notebook, and Explore mode.
The lab engines replay the same state from the same steps. Market offers stay guided so one demonstration can be compared with another.
Writing stays in the browser. It is not synced, and it is not placed in a share link.
If the guide is missing, offline, or limited, the self-review still works. The guide does not grade the learner.
Five-person study · Pending
Five real beginners still have to complete and evaluate every lab without William explaining the interface. That study has not been run. Hesitation, misclicks, terminology, and unclear cause and effect would be recorded anonymously. Automated checks can show that the models replay, that old records still load, and that share links omit writing. They cannot show that a beginner understood praxeology.