Cost & Velocity Observations

Operational metrics from 8 milestones of AI-assisted solo development on VelaPay.

Per-Milestone Velocity

Milestone	Phases	Plans	Calendar Days	Plans/Day	Focus Area
v1.0	5	20	—	—	Base protocol + checkout
v1.1	8	26	2	13.0	Privacy layer, keeper automation
v1.2	10	36	4	9.0	Admin dashboard, validation, gap closure
v1.3	3	13	2	6.5	Auth migration, Better Auth
v1.4	2	13	1	13.0	Agent mandate protocol
v1.5	7	37	2	18.5	Public billing surfaces, checkout sessions
v1.6	4	16	2	8.0	Docs-only milestone
v1.7	5	29	3	9.7	Protocol/SDK refactor, versioned accounts
v1.8	3	47	Active	Active	Streaming, dashboard, webhooks

Velocity Analysis

Highest velocity milestones:

v1.5 (18.5 plans/day) — Public billing surfaces had clear patterns from prior work. Shared checkout-session abstraction reduced per-plan complexity. High reuse of established patterns.
v1.1, v1.4 (13.0 plans/day) — Focused scope with minimal integration complexity. v1.4's agent mandates were a clean extension of the existing mandate system.

Lowest velocity milestones:

v1.3 (6.5 plans/day) — Auth migration is inherently slow: every plan requires testing against multiple auth states (logged in, logged out, session expired, wallet-only, email-only). Playwright E2E setup consumed significant time.
v1.6 (8.0 plans/day) — Docs milestones have different characteristics. Content writing is faster than code, but review and cross-referencing takes longer. Build-time validation added overhead.

Median velocity: ~9-10 plans/day for feature work. This is the reliable planning number for future milestone estimation.

Phase-Level Breakdown

Plans per Phase by Type

Phase Type	Avg Plans/Phase	Range	Notes
Feature phases	5-6	3-9	Higher coupling, more plans needed
Gap-closure phases	1-3	1-5	Fewer plans but higher debugging ratio
Docs-only phases	3-4	2-6	Lower coupling, content-driven size
Protocol/SDK phases	6-8	4-12	Highest coupling, careful sequencing
Infrastructure phases	4-5	2-7	CI/CD, deployment, configuration
Auth phases	4-6	3-8	Every plan touches sensitive paths

Plan Duration Distribution

Duration	Plan Type	Frequency
5-10 min	Simple implementations (config, types, utilities)	~30%
10-20 min	Standard features (CRUD, UI components, API routes)	~50%
20-30 min	Complex features (protocol instructions, auth flows, integrations)	~15%
30+ min	Debugging, refactoring, gap-closure	~5%

Average plan execution: ~15 minutes across all types.

Gap-Closure Overhead Analysis

v1.2 Gap-Closure (Phases 20-23)

Metric	Value
Gap-closure phases	4
Gap-closure plans	~8
Feature phases	6
Feature plans	~28
Overhead ratio	~22%
Root cause	Deferred wave-0 test stubs

v1.7 Gap-Closure (Phases 43-44)

Metric	Value
Gap-closure phases	2
Gap-closure plans	~6
Feature phases	3
Feature plans	~23
Overhead ratio	~30%
Root cause	Skipping verifier on Phase 42
Specific gaps	Integration drift + traceability drift

Gap-Closure Cost Model

Factor	Impact
Each skipped verifier	+1-2 gap-closure phases
Each gap-closure phase	+1-3 plans
Deferred validation debt	Compounds ~10% per milestone if not addressed
Dedicated closure vs. in-band	Dedicated is 2x more effective (catches more issues)

Recommendation: Budget 1-2 gap-closure phases per milestone (~3-6 plans, ~45-90 min). This is cheaper than the compound interest of deferred validation.

Model Usage & Cost

Model Configuration

Role	Model	Profile	Purpose
Primary (orchestrator)	Opus 4.6	quality	Planning, architecture decisions, complex debugging
Executor subagent	Sonnet	balanced	Plan execution, code generation, file creation
Planner subagent	Sonnet	balanced	Phase planning, task breakdown, dependency analysis
Researcher subagent	Sonnet	balanced	Context gathering, documentation search, web lookup
Reviewer subagent	Sonnet	balanced	Code review, security review, quality checks

Execution Time Breakdown

Component	Time	Percentage
Plan execution (subagents)	~25 hours	~69%
Planning + orchestration	~6 hours	~17%
Verification + debugging	~3 hours	~8%
Milestone audits	~2 hours	~6%
Total	~36 hours	100%

Context Window Usage

Milestone	Avg Context at Plan Start	Peak Context	Notes
v1.0-v1.1	~20K tokens	~60K tokens	Small codebase, low context needs
v1.2	~40K tokens	~100K tokens	Growing codebase, more cross-referencing
v1.3	~35K tokens	~80K tokens	Focused scope, moderate context
v1.5	~50K tokens	~120K tokens	Multi-repo work, high context needs
v1.7	~60K tokens	~150K tokens	Protocol refactor, maximum cross-referencing
v1.8	~55K tokens	~130K tokens	Streaming + multi-repo, high context

Commit Activity

Per-Milestone Commits

Milestone	Approximate Commits	Key Pattern
v1.0	~50	Initial scaffolding + protocol
v1.1	~60	Privacy + keeper additions
v1.2	~90	Admin dashboard + gap closure
v1.3	~35	Auth migration
v1.4	~30	Agent mandates
v1.5	~85	Public surfaces + checkout
v1.6	~40	Documentation
v1.7	~70	Protocol refactor
v1.8	~70+	Streaming + webhooks (active)
Total	529+	—

Commit Message Patterns

The project follows conventional commits:

feat(phase-N): description — feature implementations
fix(phase-N): description — bug fixes
docs(phase-N): description — documentation
ci(phase-N): description — CI/CD changes
chore(phase-N): description — maintenance

Infrastructure Costs

Cloudflare (Primary Platform)

Service	Usage	Monthly Cost
Workers	5 deployments	Free tier
D1	2 databases (dev + staging)	Free tier
KV	Session storage	Free tier
R2	Static assets	Free tier
Queues	Billing events	Free tier (10k ops/day)

Total Cloudflare cost: $0/month during development.

Railway (Dashboard SSR)

Service	Usage	Monthly Cost
Railway compute	Dashboard SSR	~$5/month (developer plan)

Helius (RPC + Webhooks)

Service	Usage	Monthly Cost
RPC provider	Devnet + staging	Free tier
Webhooks	Event indexing	Free tier
DAS API	Asset queries	Free tier

Development Tooling

Service	Usage	Monthly Cost
AI assistance (Opus 4.6 + Sonnet)	~36 hours execution	Variable
GitHub (private repos)	5 repos	Free tier
Domain (velapay.com)	Cloudflare DNS	~$10/year

Total infra cost during development: ~$5/month + AI costs.

Estimation Guide for Future Milestones

How to Estimate a Milestone

Count the number of feature phases (from roadmap)
Add 1-2 gap-closure phases (standard allocation)
Estimate plans per phase: 5-6 for features, 3-4 for docs
Calculate total plans: phases × plans/phase
Estimate duration: total plans ÷ 9 plans/day
Add 20% buffer for gap-closure and debugging

Example Estimation

Milestone v2.0: Multi-token billing
- Feature phases: 4
- Gap-closure phases: 1
- Plans per feature phase: 6
- Plans per gap-closure phase: 3
- Total plans: (4 × 6) + (1 × 3) = 27 plans
- Estimated duration: 27 ÷ 9 = 3 days
- With buffer: 3.6 days → ~4 days

Velocity Adjustments

Factor	Adjustment	Reason
Protocol changes	-20% plans/day	Higher coupling, more testing
New auth patterns	-30% plans/day	Every plan touches sensitive paths
Docs-only	-10% plans/day	Review takes longer than writing
Reusing established patterns	+30% plans/day	Less design, more execution
Multi-repo coordination	-15% plans/day	Context switching overhead

Cost & Velocity Observations ​

Per-Milestone Velocity ​

Velocity Analysis ​

Phase-Level Breakdown ​

Plans per Phase by Type ​

Plan Duration Distribution ​

Gap-Closure Overhead Analysis ​

v1.2 Gap-Closure (Phases 20-23) ​

v1.7 Gap-Closure (Phases 43-44) ​

Gap-Closure Cost Model ​

Model Usage & Cost ​

Model Configuration ​

Execution Time Breakdown ​

Context Window Usage ​

Commit Activity ​

Per-Milestone Commits ​

Commit Message Patterns ​

Infrastructure Costs ​

Cloudflare (Primary Platform) ​

Railway (Dashboard SSR) ​

Helius (RPC + Webhooks) ​

Development Tooling ​

Estimation Guide for Future Milestones ​

How to Estimate a Milestone ​

Example Estimation ​

Velocity Adjustments ​