/brain:review

Run a spaced repetition review session using the SM-2 algorithm. Surfaces memories that are due for review at optimal intervals for long-term retention.

Usage

/brain:review [scope]

Arguments:

• [scope] — (Optional) Limit the review to a specific category or subtree (e.g., professional/skills). If omitted, reviews all due memories.

What It Does

/brain:review implements the SM-2 (SuperMemo 2) algorithm to schedule memory reviews at increasing intervals. The review queue is generated during /brain:sleep and stored in ~/.brain/review-queue.json.

The Review Process

1. Load due memories — Reads the review queue and filters for memories whose next review date has passed
2. Present each memory — Shows the memory content one at a time
3. Collect quality grade — You rate your recall quality from 1 to 5
4. Update scheduling — Adjusts the review interval and ease factor based on your grade
5. Apply reinforcement — Each reviewed memory receives a strength boost via spaced reinforcement

Quality Grades

Interval Calculation

After a successful review (grade >= 3), the next interval is calculated:

next_interval = current_interval * ease_factor

The ease factor starts at 2.5 and adjusts based on recall quality:

new_ease = ease + (0.1 - (5 - grade) * (0.08 + (5 - grade) * 0.02))

The ease factor has a minimum of 1.3 to prevent intervals from shrinking too aggressively.

Interval Progression

With consistent grade-4 reviews (ease factor 2.5):

Failed recalls (grade 1 or 2) reset the interval to 1 day, so you review the memory again the next session.

Example Interaction

User: /brain:review

Agent: 📋 3 memories due for review:

Review 1/3:
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
📝 Chose Event-Driven Architecture for Project Alpha

We decided to use event-driven architecture with Kafka instead of
synchronous REST calls. Kafka was chosen over RabbitMQ for replay
capability. 3-week implementation estimate.

Last reviewed: 12 days ago | Strength: 0.78
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
How well did you recall this? (1-5): 4

✅ Next review in 30 days. Strength boosted to 0.82.

Review 2/3:
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
💡 TypeScript satisfies operator preserves narrowed types

The `satisfies` operator checks assignability while preserving
the narrowed type, unlike `as` which widens to the target type.

Last reviewed: 8 days ago | Strength: 0.65
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
How well did you recall this? (1-5): 2

⚠️ Interval reset to 1 day. Review again next session.

Review 3/3:
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
🔧 Git rebase workflow

Fetch upstream, rebase onto main, resolve conflicts file-by-file,
force push to branch.

Last reviewed: 20 days ago | Strength: 0.72
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
How well did you recall this? (1-5): 5

✅ Next review in 50 days. Strength boosted to 0.78.

Review complete! 2 of 3 memories reinforced successfully.
1 memory needs re-review tomorrow.

Queue Generation

The review queue is populated during the Expertise Detection phase of /brain:sleep. Memories are selected based on:

• Decay trajectory — memories at risk of falling below thresholds
• Expertise gaps — knowledge areas where retention could be improved
• Recall history — memories that have not been reviewed in a long time