Cognitive Sectors

The 5 types of memory and when to use each

Cognitive Sectors

Different information needs different lifespans. The Memory Module organizes memories into 5 cognitive sectors, each with appropriate decay rates.

Your brain doesn’t treat “yesterday’s lunch” the same as “your wedding day.” Neither does the Memory Module.

---

Quick Reference

Sector	Half-Life	Decay	Best For
Episodic	46 days	Fast	Time-bound events (meetings, bugs, deployments)
Semantic	139 days	Slow	Facts & concepts (API docs, definitions)
Procedural	87 days	Medium	How-to knowledge (processes, tutorials)
Emotional	35 days	Very Fast	Sentiment & feedback (morale, satisfaction)
Reflective	693 days	Very Slow	Strategic insights (architecture decisions, “why we did this”)

---

1. Episodic

What

Time-bound events & experiences

Has specific “when” and “where”

Examples: Meeting notes, bug investigations, deployments

Decay

Half-life: 46 days (medium-fast)

Week 1: Fresh, highly relevant Week 4: Still useful Week 12: Fading Week 24: Mostly forgotten

Use for:

• Meeting notes and decisions
• Bug investigation logs
• Deployment records
• Customer interaction logs
• Project milestone events
• Incident reports

Examples:

Standup

"Daily standup - Jan 25, 2025
- Marcus: Finished auth refactor PR #234
- Sarah: Blocked on DB migration
- Aisha: Starting memory docs
Next standup: Jan 26"

Sector: episodic

Why: Specific date/time, useful short-term, less relevant after sprint

Bug Investigation

"Payment bug investigation - Jan 20
- Issue: Stripe webhook timing out
- Cause: 30s timeout, webhook takes 45s
- Fix: Moved to background job
- Deployed: v2.3.1"

Sector: episodic

Why: Event-based, valuable while debugging, fades after fix deployed

Support Call

"Support call - Acme Corp - Jan 20, 2pm
- Issue: API rate limiting (429 errors)
- Resolution: Upgraded to Pro plan
- Follow-up: Jan 27"

Sector: episodic

Why: Specific interaction, useful for follow-up, fades after resolved

---

2. Semantic

What

Facts, concepts & general knowledge

Timeless information

Examples: API docs, definitions, configurations

Decay

Half-life: 139 days (slow)

Month 1: Fresh Month 3: Still strong (0.85) Month 6: Moderate (0.65) Year 1: Fading (0.40)

Use for:

• API documentation and references
• Technical definitions
• Company policies
• Product feature descriptions
• Configuration settings
• Integration docs

Examples:

API Docs

"Our REST API uses JWT bearer tokens.
Tokens expire after 30 days inactivity.
Header: Authorization: Bearer {token}

Generate: POST /api/auth/token
Refresh: POST /api/auth/refresh"

Sector: semantic

Why: Timeless factual information, referenced frequently

Product Knowledge

"ULPI subscription plans:
- Starter: $49/mo, 1K memories
- Pro: $199/mo, 10K memories
- Enterprise: Custom pricing

All plans: MCP integration, API access"

Sector: semantic

Why: Facts about product, useful long-term, rarely changes

Technical Concepts

"Semantic waypoints:
- Graph edges connecting similar memories
- Created when similarity ≥0.75
- BFS search up to 3 hops
- 20% salience decay per hop

Purpose: Automatic context expansion"

Sector: semantic

Why: Conceptual knowledge, relevant as long as feature exists

---

3. Procedural

What

How-to knowledge & processes

Step-by-step instructions

Examples: Deployment processes, coding patterns, tutorials

Decay

Half-life: 87 days (medium)

Week 2: Fresh Month 2: Still reliable (0.80) Month 4: Need refresh (0.60) Month 6: Update needed (0.45)

Use for:

• Deployment procedures
• Setup guides
• Troubleshooting steps
• Coding patterns
• Configuration workflows
• Testing procedures

Examples:

Deployment

"Production deployment process:
1. Run tests: npm run test
2. Build: npm run build
3. Tag: git tag v{version}
4. Push: git push origin v{version}
5. Deploy: vercel --prod
6. Verify: Check /health endpoint
7. Monitor: Watch logs for 10min"

Sector: procedural

Why: Step-by-step process, needs occasional updates

Setup Guide

"Setting up local dev environment:
1. Install Node 20: nvm install 20
2. Clone repo: git clone {repo}
3. Install deps: npm install
4. Copy env: cp .env.example .env.local
5. Start: npm run dev
6. Verify: localhost:3000"

Sector: procedural

Why: Workflow that may change with tooling updates

Troubleshooting

"Fixing 'Module not found' errors:
1. Clear cache: rm -rf .next
2. Delete modules: rm -rf node_modules
3. Reinstall: npm install
4. Restart: npm run dev
5. Still broken? Check imports for typos"

Sector: procedural

Why: Process that may need updates as tools evolve

---

4. Emotional

What

Sentiment & feedback

Feelings, morale, reactions

Examples: Customer frustration, team morale, satisfaction

Decay

Half-life: 35 days (very fast)

Day 1: Fresh Week 2: Fading (0.70) Week 4: Mostly gone (0.50) Week 8: Very faint (0.25)

Use for:

• Customer sentiment
• Team morale tracking
• User satisfaction
• Frustration points
• Positive feedback
• Emotional context

Why fast decay: Sentiment changes quickly, old emotions less relevant

Examples:

Customer Feedback

"Acme Corp frustrated with API rate limits.
Expressed urgency - blocking their launch.
Tone: Frustrated but professional.
Resolution: Upgraded plan, now satisfied.

Sentiment: Frustrated → Satisfied"

Sector: emotional

Why: Sentiment snapshot, less relevant after resolution

Team Morale

"Team morale high after successful launch.
Everyone excited about positive customer feedback.
Celebrating at team dinner tonight.

Context: v2.0 launch week"

Sector: emotional

Why: Emotional state, changes as project progresses

User Reaction

"Users love the new search feature!
Twitter feedback overwhelmingly positive.
Multiple users saying 'game changer.'

Feature: Semantic waypoints"

Sector: emotional

Why: Reaction to feature, fades as feature becomes expected

---

5. Reflective

What

Strategic insights & decisions

The “why” behind choices

Examples: Architecture decisions, lessons learned, retrospectives

Decay

Half-life: 693 days (very slow)

Month 6: Still 0.95 Year 1: Still 0.85 Year 2: Still 0.70 Year 3: Still 0.60

Use for:

• Architecture decisions
• Technology choices
• Lessons learned
• Strategic pivots
• Post-mortems
• “Why we did this”

Why slow decay: Strategic insights remain valuable for years

Examples:

Architecture

"Why we chose microservices over monolith:

Context: Scaling to 100+ developers across teams
Decision: Microservices architecture
Reasoning:
- Team autonomy (deploy independently)
- Technology flexibility (right tool per service)
- Fault isolation (one service down ≠ all down)

Trade-offs accepted:
- Increased operational complexity
- Distributed system challenges
- Network latency between services

Date: Jan 2024"

Sector: reflective

Why: Strategic decision that informs future architecture choices

Lesson Learned

"Production outage retrospective - Dec 2024

What happened: Database migration ran during peak traffic
Impact: 45min downtime, 500 affected users
Root cause: No maintenance window policy

Lessons:
- Always schedule maintenance windows
- Run migrations in off-peak hours
- Test migrations on staging first
- Have rollback plan ready

Actions: Created maintenance window policy"

Sector: reflective

Why: Hard-won wisdom that prevents future incidents

Tech Choice

"Why we chose React over Vue for frontend:

Context: Rebuilding admin dashboard
Decision: React + Next.js
Reasoning:
- Larger talent pool (hiring)
- Better TypeScript support
- Rich ecosystem (libraries)
- Server components (performance)

Alternatives considered:
- Vue 3: Smaller ecosystem
- Svelte: Less mature

Date: Feb 2024"

Sector: reflective

Why: Technology rationale that guides future tech decisions

---

Choosing the Right Sector

Decision tree:

Is it about feelings/sentiment?
  → Emotional (fast decay)

Is it a time-bound event?
  → Episodic (medium-fast decay)

Is it a step-by-step process?
  → Procedural (medium decay)

Is it a fact or definition?
  → Semantic (slow decay)

Is it strategic insight or "why"?
  → Reflective (very slow decay)

When in doubt:

• Events with dates → Episodic
• Timeless facts → Semantic
• How-to guides → Procedural
• Strategic reasons → Reflective

---

Sector Comparison

Aspect	Episodic	Semantic	Procedural	Emotional	Reflective
Question	When/Where?	What is it?	How to do it?	How did it feel?	Why did we do it?
Timeframe	Specific date	Timeless	Periodic	Short-lived	Long-lasting
Example	Bug fix Jan 15	API uses JWT	Deploy: 7 steps	Customer happy	Why microservices
Decay	46 days	139 days	87 days	35 days	693 days
Update	Rarely	When facts change	When process changes	Often	Rarely

---

Auto-Classification

The system can auto-classify based on content:

store_memory({
  content: "Fixed payment bug on Jan 20...",
  sector: "auto"  // System detects: episodic (has date, event)
})

store_memory({
  content: "API authentication uses JWT tokens...",
  sector: "auto"  // System detects: semantic (factual, timeless)
})

store_memory({
  content: "Deploy process: 1. Build 2. Test 3. Push...",
  sector: "auto"  // System detects: procedural (step-by-step)
})

Override when needed:

store_memory({
  content: "Why we chose React...",
  sector: "reflective"  // Manual: strategic decision
})

---

Best Practices

Be Explicit

Specify sector when storing

✅ Good:

store_memory({
  content: "...",
  sector: "reflective"
})

❌ Risky:

store_memory({
  content: "..."  // Auto-classify may guess wrong
})

Match Content to Sector

Write for the sector

Reflective: Include “why” and context

"Why we chose X:
Context: ...
Reasoning: ...
Trade-offs: ..."

Episodic: Include date/time

"Meeting notes - Jan 25, 2025
Attendees: ...
Decisions: ..."

Use Reflective for Decisions

Strategic choices = Reflective

Not episodic (will fade too fast) Not semantic (not just facts)

✅ “Why we chose microservices…” → Reflective ❌ “We use microservices” → Semantic

Emotional for Sentiment Only

Pure sentiment = Emotional

Not for facts about sentiment

✅ “Customer frustrated about X” → Emotional ❌ “Customer satisfaction score: 8.5” → Semantic

---

Next Steps

Decay & Reinforcement

How salience changes over time

Best Practices

Optimize memory usage

Workflows

Put sectors to work

---

Different memories need different lifespans. Choose the right sector, let the system handle the rest.