# Training Protocols

Use these protocols to make review active instead of passive.

## Recall Scoring

Score after answering from memory and comparing to the source.

| Score | Meaning | Next Due |
| --- | --- | --- |
| 0 | No recall or wrong answer | tomorrow |
| 1 | Fragment only; major gap | tomorrow |
| 2 | Partly right; missing key condition | +2 days |
| 3 | Mostly right; weak transfer | +4 days |
| 4 | Correct and usable | +7 days |
| 5 | Correct, nuanced, and transferable | +14 days |

If the card scores 4 or 5 on three separate days, mark `Stability: stable`. If it is broadly useful, promote it to `principles.md`.

## Daily Consolidation Protocol

1. Read today's inbox and due cards.
2. Group raw events by concept, procedure, mistake, preference, or strategy.
3. Merge duplicates. Do not create multiple cards for the same lesson.
4. Convert each valuable group into a card.
5. Run recall for due cards before reading answers.
6. Update due dates, graph links, mistake prevention rules, and tomorrow's strategy.

## Interleaving Protocol

Mix review topics so the agent learns selection, not just repetition:

1. Pick one concept card.
2. Pick one procedure card.
3. Pick one mistake card.
4. Ask: "Which card matters for the current user task and why?"
5. Apply the chosen card to a concrete next action.

## Transfer Protocol

For each high-value lesson, generate:

- one same-domain application
- one different-domain analogy
- one counterexample where the lesson would mislead
- one experiment that would produce new evidence

Promote only lessons that survive counterexamples.

## Strategic Improvement Protocol

Use this at the end of each day:

1. Name the single bottleneck that limited performance today.
2. Identify whether it was knowledge, process, tool fluency, communication, or judgment.
3. Choose one practice task for tomorrow.
4. Choose one prevention rule.
5. Choose one evidence source to inspect next.

Avoid vague goals such as "be better". Use actions that can be checked tomorrow.