well-architected

name	well-architected
description	Six-pillar well-architected framework for evaluating architecture quality — operational excellence, security, reliability, performance, cost optimization, and developer experience

Well-Architected Framework

A structured framework for evaluating architecture quality across six pillars. Inspired by the AWS Well-Architected Framework, adapted for startups and modern product development. Technology-agnostic — applies to any stack.

Use this skill when generating the Well-Architected Review deliverable in /architect:blueprint or when running /architect:well-architected as a standalone evaluation.

---

The Six Pillars

1. Operational Excellence

Question: Can you deploy, monitor, and improve the system with confidence?

Criteria	What Good Looks Like	Score Guide
CI/CD pipeline	Automated lint → test → build → deploy on every push	1 = manual deploys, 5 = full CI/CD with rollback
Infrastructure as Code	Deployment config in repo (Vercel config, Dockerfile, Terraform)	1 = manual setup, 5 = fully reproducible
Observability	Structured logs + error tracking + health checks	1 = console.log only, 5 = full observability stack
Incident response	Alerts → runbook → mitigation → post-mortem process	1 = no plan, 5 = documented runbooks and on-call
Change management	Feature flags, staged rollouts, database migrations versioned	1 = YOLO deploys, 5 = staged rollouts with flags

Key questions to ask:

• How do you deploy a change today? How long does it take?
• When something breaks at 2am, how do you find out? How do you fix it?
• Can a new developer deploy to staging on their first day?

---

2. Security

Question: Is user data protected, and are attack surfaces minimized?

Criteria	What Good Looks Like	Score Guide
Authentication	Managed auth provider, MFA for admin, token rotation	1 = DIY password hashing, 5 = managed auth + MFA
Authorization	Role-based or attribute-based access control on every endpoint	1 = no authorization checks, 5 = RBAC/ABAC middleware
Data protection	Encryption at rest and in transit, PII identified and handled	1 = plaintext everywhere, 5 = encrypted + PII policies
API security	Rate limiting, input validation, CORS, security headers	1 = none, 5 = all OWASP top 10 mitigated
Secrets management	No hardcoded secrets, env vars or secrets manager, rotation policy	1 = secrets in code, 5 = secrets manager + rotation
Dependency security	Automated vulnerability scanning in CI	1 = never audited, 5 = automated audit + auto-fix

Key questions to ask:

• What happens if an API key leaks? How fast can you rotate it?
• Can a regular user access admin endpoints by guessing the URL?
• Are you storing anything you shouldn't be? (passwords, full card numbers, unnecessary PII)

---

3. Reliability

Question: Does the system keep working when things go wrong?

Criteria	What Good Looks Like	Score Guide
Fault isolation	One service failing doesn't cascade to others	1 = monolith with no error boundaries, 5 = circuit breakers + fallbacks
Recovery	Automated restarts, health checks, self-healing	1 = manual restart, 5 = auto-restart + health checks + failover
Data durability	Automated backups, point-in-time recovery, tested restores	1 = no backups, 5 = automated backups + tested restores
Retry & timeout	Retries with exponential backoff, timeouts on all external calls	1 = no retries/timeouts, 5 = retry policies on all external calls
Graceful degradation	System works (reduced functionality) when a dependency is down	1 = hard crash, 5 = graceful fallback for each dependency
Scaling	Handles 10x current load without architecture changes	1 = breaks at 2x, 5 = auto-scaling with no code changes

Key questions to ask:

• What happens when the database goes down for 5 minutes?
• What happens when a third-party API (Stripe, SendGrid) is slow or down?
• Can the system handle a sudden traffic spike (e.g., HackerNews front page)?

---

4. Performance Efficiency

Question: Are resources used efficiently, and is the user experience fast?

Criteria	What Good Looks Like	Score Guide
Response time	API p95 < 500ms, page load < 2s, AI response < 5s	1 = > 3s average, 5 = p95 < 200ms
Right-sizing	Resources match actual load (not over/under-provisioned)	1 = fixed large instances, 5 = auto-scaled to demand
Caching	Appropriate caching at each layer (CDN, API, database)	1 = no caching, 5 = multi-layer caching strategy
Async processing	Heavy work offloaded to background jobs / queues	1 = everything synchronous, 5 = async where appropriate
Database efficiency	Indexed queries, connection pooling, no N+1 problems	1 = unoptimized queries, 5 = optimized + monitored
Frontend performance	Code splitting, lazy loading, optimized images, CDN	1 = single bundle, no CDN, 5 = optimized + CDN + edge

Key questions to ask:

• What's the slowest user-facing operation? Can it be made async?
• Are you paying for compute that's idle 90% of the time?
• Is there a caching layer, or does every request hit the database?

---

5. Cost Optimization

Question: Are you spending the minimum needed for the current scale?

Criteria	What Good Looks Like	Score Guide
Free tier usage	Maximizing free tiers before paying	1 = paying for everything, 5 = optimized free tier usage
Right-sizing	Resources match actual usage, not theoretical maximum	1 = over-provisioned, 5 = auto-scaled or right-sized
Cost awareness	Team knows monthly cost breakdown, alerts on overspend	1 = no idea of costs, 5 = cost dashboards + budget alerts
Service selection	Chosen services fit the scale (not enterprise tools for MVP)	1 = enterprise tools for 10 users, 5 = appropriate for scale
LLM cost control	Token optimization, caching, model selection, rate limits	1 = GPT-4 for everything, 5 = tiered models + prompt caching
Scaling economics	Costs scale sub-linearly with users	1 = linear cost scaling, 5 = strong economies of scale

Key questions to ask:

• What's your monthly cloud bill? Do you know what each line item is?
• Are you using the cheapest option that meets your requirements?
• When you 10x users, does cost 10x too, or less?

---

6. Developer Experience

Question: Can developers build, test, and ship features quickly and confidently?

Criteria	What Good Looks Like	Score Guide
Onboarding	New dev productive in < 1 day (README, setup scripts, seed data)	1 = tribal knowledge, 5 = automated setup + docs
Local development	docker compose up or npm run dev and everything works	1 = complex manual setup, 5 = one-command startup
Type safety	TypeScript/Python type hints throughout, shared types across services	1 = untyped, 5 = strict types + shared type packages
Testing	Fast unit tests, meaningful integration tests, CI runs all tests	1 = no tests, 5 = comprehensive test suite < 5 min
Code organization	Clear folder structure, consistent patterns, separation of concerns	1 = spaghetti code, 5 = clear architecture + conventions
Documentation	API docs (OpenAPI), architecture docs, runbooks	1 = no docs, 5 = auto-generated API docs + architecture docs

Key questions to ask:

• How long does it take a new developer to submit their first PR?
• Can you run the entire system locally without cloud credentials?
• Is there a style guide or do code reviews catch inconsistencies?

---

Scoring Methodology

Per-Pillar Scoring

Each pillar is scored 1-5 based on the average of its criteria:

Score	Label	Meaning
1	Critical	Fundamental gaps. Address immediately before building further.
2	Needs Work	Significant gaps. Plan to address in next sprint/milestone.
3	Adequate	Meets minimum bar. Acceptable for MVP, improve for production.
4	Good	Solid implementation. Minor improvements possible.
5	Excellent	Best practices followed. Ready for production scale.

Overall Architecture Rating

Average of all 6 pillar scores:

Overall	Rating	Interpretation
1.0 - 2.0	Fragile	Architecture has critical gaps. Not production-ready.
2.1 - 3.0	Developing	Functional but risky. Acceptable for early MVP with a plan to improve.
3.1 - 3.5	Solid	Good foundation. Typical for well-planned MVP or early startup.
3.6 - 4.0	Strong	Production-quality. Ready for real users and growth.
4.1 - 5.0	Exemplary	Mature architecture. Enterprise-ready. Rare for early-stage products.

Stage-Appropriate Expectations

Not every project needs a 5/5 on every pillar. Set expectations by stage:

Stage	Target Score	Acceptable Lows
Proof of concept	2.0 overall	Security 2, Reliability 1, DevEx 2
MVP	2.5 - 3.0	Reliability 2, Performance 2
Early product (paying users)	3.0 - 3.5	Performance 3, Cost 2
Growth stage	3.5 - 4.0	None below 3
Production / enterprise	4.0+	None below 4

---

Output Format

IMPORTANT: Be comprehensive and thorough in all sections.

1. Visual Score Summary (REQUIRED)

Visualize pillar scores as a horizontal bar table:

Operational Excellence  ████░  4/5 — Good
Security               ███░░  3/5 — Adequate
Reliability            ██░░░  2/5 — Needs Work
Performance Efficiency ████░  4/5 — Good
Cost Optimization      █████  5/5 — Excellent
Developer Experience   ███░░  3/5 — Adequate

Overall: 3.5/5 — Solid

Score table (also provide as markdown table):

Pillar	Score	Label	Key Strength	Critical Gap
Operational Excellence	X/5	[Label]	[1 sentence]	[1 sentence or "None"]
Security	X/5	[Label]	[1 sentence]	[1 sentence or "None"]
Reliability	X/5	[Label]	[1 sentence]	[1 sentence or "None"]
Performance Efficiency	X/5	[Label]	[1 sentence]	[1 sentence or "None"]
Cost Optimization	X/5	[Label]	[1 sentence]	[1 sentence or "None"]
Developer Experience	X/5	[Label]	[1 sentence]	[1 sentence or "None"]
Overall	X.X/5	[Rating]

2. Per-Pillar Detail (MUST be thorough for each pillar)

For each of the 6 pillars, provide comprehensive analysis:

Format:

## [Pillar Name]: X/5 — [Label]

### Score Breakdown

Evaluated against [number] criteria:

| Criteria | Score | Notes |
|----------|:-----:|-------|
| [Criterion 1] | X/5 | [1 sentence why this score] |
| [Criterion 2] | X/5 | [1 sentence why this score] |
| [Criterion 3] | X/5 | [1 sentence why this score] |
| ... | ... | ... |

**Average: X.X/5 → X/5 ([Label])**

### Strengths (minimum 2-4 bullets)

✅ **[Specific strength]**
   - What: [Describe what's implemented]
   - Why it matters: [Business/technical impact]
   - Reference: [Cite specific architecture decision from manifest]

✅ **[Specific strength]**
   - What: [Describe what's implemented]
   - Why it matters: [Business/technical impact]
   - Reference: [Cite specific architecture decision from manifest]

[Continue for all major strengths...]

### Gaps (minimum 2-4 bullets if score < 5)

⚠️ **[Specific gap]**
   - What's missing: [Describe the missing capability]
   - Risk if not addressed: [Specific risk with impact estimation]
   - Severity: [Critical/High/Medium/Low]

⚠️ **[Specific gap]**
   - What's missing: [Describe the missing capability]
   - Risk if not addressed: [Specific risk with impact estimation]
   - Severity: [Critical/High/Medium/Low]

[Continue for all significant gaps...]

### Recommendations (minimum 3-5 specific actions)

Each recommendation must be:
- **Actionable**: Specific enough to implement immediately
- **Measured**: Include effort estimate and impact level
- **Prioritized**: Show order of implementation

**Format for each recommendation:**

**Recommendation #X: [Short action-oriented title]** — [Impact: High/Medium/Low] — [Effort: X hours/days/weeks]

What to do:
[2-3 sentences describing specific implementation steps]

Why it matters:
[1-2 sentences on business/technical impact]

Implementation notes:
 Tool/service to use: [Specific recommendation]
 Code location: [Where to implement, if applicable]
 Dependencies: [What must be done first]
 Definition of done: [How to verify it's complete]

Cost impact: $[amount]/month or [one-time cost]
Timeline: [When to implement — Pre-launch / Month 1 / Quarter 1 / Future]



 — Impact: High — Effort: 4-6 hours

What to do:
Add rate limiting middleware using Vercel's @upstash/ratelimit package. Set default limit of 100 requests per minute per IP, with stricter limits (10 req/min) on authentication endpoints and looser limits (1000 req/min) for authenticated users. Return 429 status with Retry-After header when limit exceeded.

Why it matters:
Without rate limiting, a single malicious user or misconfigured client can overwhelm the API, causing downtime for all users. This is especially critical for authentication endpoints which are common DDoS targets. Rate limiting is a pre-launch requirement for production deployment.

Implementation notes:
 Tool/service to use: @upstash/ratelimit with Vercel KV (free tier: 10K requests/day)
 Code location: src/middleware/rateLimit.ts, apply in src/app/api/

3. Improvement Roadmap (minimum 8-12 items)

Provide comprehensive prioritized roadmap with P0-P3 priority levels:

Priority	Pillar	Action	Effort	Impact	Stage	Cost
P0	Security	Add rate limiting to all API endpoints	4-6 hours	High	Pre-launch	$0
P0	Reliability	Implement health checks with dependency verification	3 hours	High	Pre-launch	$0
P0	Security	Set up automated dependency scanning in CI	2 hours	High	Pre-launch	$0
P1	Operational Excellence	Add structured logging with correlation IDs	1 day	Medium	Month 1	$26/mo (Sentry)
P1	Reliability	Set up database backups with tested restore process	4 hours	High	Month 1	$0 (included)
P1	Security	Implement secrets rotation for API keys	2 days	Medium	Month 1	$0
P2	Performance	Add Redis caching layer for hot queries	2-3 days	Medium	Quarter 1	$7/mo
P2	Operational Excellence	Implement feature flags for gradual rollouts	1 day	Low	Quarter 1	$0 (self-hosted)
P2	Developer Experience	Add OpenAPI schema generation for API docs	1 day	Low	Quarter 1	$0
P3	Performance	Implement database query optimization and indexing review	3-5 days	Medium	Future	$0
P3	Reliability	Add chaos engineering / failure injection testing	1 week	Low	Future	$0
P3	Developer Experience	Set up comprehensive E2E test suite	1-2 weeks	Medium	Future	$0

Priority Definitions:

• P0 (Must-have before launch): Blocks production deployment. Critical security, reliability, or operational gaps. Complete in current sprint.
• P1 (Should-have in first month): Important for stability and user trust. Complete within 30 days of launch.
• P2 (Nice-to-have in first quarter): Improves experience and reduces operational burden. Complete within 90 days.
• P3 (Future enhancement): Optimization or nice-to-have. Evaluate after product-market fit.

Cost Summary:

• Pre-launch (P0): $X/month + Y hours labor
• Month 1 (P0+P1): $X/month + Y hours labor
• Quarter 1 (P0+P1+P2): $X/month + Y hours labor
• Full roadmap (All): $X/month + Y hours labor

Timeline Visualization:

Pre-Launch (Week 0):
├─ P0 items (total: X hours)
└─ Must complete before production deployment

Month 1 (Weeks 1-4):
├─ P1 items (total: X hours)
└─ Critical for stability

Quarter 1 (Weeks 5-12):
├─ P2 items (total: X hours)
└─ Improves operational efficiency

Future (Month 4+):
├─ P3 items (total: X hours)
└─ Evaluate based on growth

4. Quick Wins (REQUIRED if any exist)

Identify 3-5 high-impact, low-effort improvements that can be done in <1 day each:

Format:

Quick Win #X: [Short title] — [Pillar Name]

What: [1-2 sentences on what to implement]
Effort: [X hours]
Impact: [High/Medium impact on pillar score]
How: [3-5 step implementation checklist]
Cost: $[amount or $0]

ROI: [Pillar score improvement: X/5 → Y/5, or specific metric improvement]

Example:

Quick Win #1: Add security headers to API responses — Security

What: Configure Next.js security headers (CSP, X-Frame-Options, HSTS, etc.) in next.config.js to protect against common web vulnerabilities.
Effort: 30 minutes
Impact: High (improves Security pillar from 3/5 to 3.5/5)
How:
1. Add headers configuration to next.config.js
2. Test with securityheaders.com
3. Verify CSP doesn't break any functionality
4. Deploy to staging and production

Cost: $0

ROI: Security pillar: 3/5 → 3.5/5, protects against XSS and clickjacking with minimal effort

5. Critical Blockers (REQUIRED if any exist)

If any pillar scores 1/5 or has critical gaps, call them out explicitly:

🚨 CRITICAL BLOCKER: [Issue name]

Pillar: [Name]
Current state: [What's broken or missing]
Risk: [What bad thing will happen]
Probability: [High/Medium/Low chance of occurrence]
Impact: [Severity if it occurs — data loss, security breach, downtime, etc.]

This blocks: [Production launch / Scaling / User trust / Compliance]

Required action: [Specific fix needed]
Effort: [Realistic time estimate]
Must complete by: [Deadline or stage gate]
Owner: [Who should do this — role/skill level]

Example:

🚨 CRITICAL BLOCKER: No database backups configured

Pillar: Reliability
Current state: Supabase project has default backups (daily, 7-day retention) but no tested restore process. No way to recover from accidental data deletion or corruption.
Risk: Single developer mistake (DROP TABLE, bad migration) or Supabase issue could cause permanent data loss for all users.
Probability: Medium (10-15% chance in first year based on industry data)
Impact: Catastrophic — lose all user data, company trust, potential legal liability

This blocks: Production launch with real users

Required action:
1. Enable Supabase point-in-time recovery (PITR) — provides 7-day recovery window
2. Set up daily automated backup export to S3 (in addition to Supabase backups)
3. Document and TEST restore procedure (actually restore a backup to verify it works)
4. Add backup monitoring (alert if backup fails)

Effort: 4-6 hours (2 hours setup + 2 hours testing + 1 hour documentation)
Must complete by: Before production launch (P0)
Owner: Backend developer or DevOps engineer

6. Scoring Transparency (ALWAYS show)

Show how the overall score was calculated:

Well-Architected Score Calculation:

Operational Excellence: X/5
Security: X/5
Reliability: X/5
Performance Efficiency: X/5
Cost Optimization: X/5
Developer Experience: X/5
                      ─────
Total: XX/30
Average: XX/30 ÷ 6 = X.XX → X.X/5

Overall Rating: X.X/5 — [Rating Label]

Interpretation based on stage:

Stage: [MVP / Early Product / Growth / Enterprise]
Expected score range: [X.X - Y.Y]
Actual score: [Z.Z]

Assessment: [Above/At/Below] expectations

[If below]: Priority focus areas: [list pillars scoring below stage expectations]
[If at/above]: Continue improving [lowest scoring pillars] while scaling

---

When Evaluating Existing Architectures

When used with /architect:well-architected on an existing architecture (not a new blueprint):

1. Ask about current state — Don't assume. Ask what's already in place for each pillar.
2. Be specific — Reference actual services, endpoints, and tools from the architecture.
3. Score honestly — A 3 is not a failure. Most startups are in the 2.5-3.5 range.
4. Prioritize ruthlessly — Don't recommend 20 improvements. Pick the top 5 that give the most value.
5. Match to stage — An MVP doesn't need the same score as an enterprise product.