SYSTEM1 / TEACH BY DOING

Your choices become
a small, useful skill.

File a document where it belongs. Build a candidate from your choices, compare it with your approved skill, then choose whether to adopt it. No prompts to engineer. No teaching JSON to write.

Local · No LLMFictional sample workspaceYour real files stay untouched
One skill, taught through an ordinary task.

Drag the document to a folder, or click the folder. A choice is an example of what you want.
1Show a few choices2Review a candidate3Adopt a passing skill

Documents

Pick a card, then choose its folder.

DEMONSTRATION

Loading sample documents…

Your folder choice becomes one teaching example.

The sample filing policy

“Calibrate” supplies separate answers for uncertainty. The 30 sample filing documents below are recurring development checks for candidate comparison. The six unusual requests are preview-only challenges. Neither group can become a lesson or calibration example; repeated use is not a fresh test.

COMPARE BEFORE ADOPTING

Candidate review

Your approved skill stays available while you edit lessons or review a candidate. Passing these authored checks is a minimum gate, not proof of real-world readiness.

Record lessons, then review a candidate.
TRY IT ON DIFFERENT DOCUMENTS

The sample inbox. Your approved skill.

System1 places accepted suggestions into sample folders. Uncertain documents wait for you. Predictions never teach themselves.

Adopt a passing candidate to preview 30 recurring filing checks and 6 unusual requests.

Try your own text

A suggestion does not add a lesson. To teach from this text, choose the correct folder yourself and save it explicitly. Saved lessons remain in your local session.

What gets learned?

The visible title and text become the observation. The folder you choose becomes the answer. That is a labelled example—created by doing the task.

Change an earlier folder choice to correct a mistake, then review and adopt a new candidate. Repeated choices update the same example; they do not inflate the count.

The approved skill contains the learned decision and its review calibration. It can be loaded without this page.

This first experiment handles text previews. It does not read PDFs, watch your desktop, or move actual files.