Skip to main content

Project Statistics — SecurityV0

Snapshot date: February 28, 2026 Project span: 35 days (Jan 24 – Feb 28, 2026)


Timeline

  • Earliest project work: January 9, 2026 (demo-scanner POC for GitHub token/ownership analysis)
  • First Claude Code conversation: January 24, 2026 (graph-mongo MVP)
  • Platform era begins: February 3 (documentation), February 4 (connectors), February 8 (platform)
  • Snapshot date: February 28, 2026

Commits

Active repositories

RepoCommitsFirst CommitLatest Commit
sv0-platform113Feb 8Feb 27
sv0-connectors82Feb 4Feb 28
sv0-documentation112Feb 3Feb 28
Active total307

Archived / predecessor projects

ProjectCommitsPeriodRole
demo-scanner8Jan 9–22First POC — GitHub token ownership scanner, authority reports
sv0_architecture1Jan 23Architecture and design documentation (migrated to sv0-documentation)
sv0-docsv11Feb 2First docs repo attempt (migrated to sv0-documentation)
graph-mongo0 (no git)Jan 24 – Feb 12First platform MVP — MongoDB graph model, 4.5K LOC TypeScript
archive/connectors (correlator)0 (no git)Jan 30 – Feb 1Early Azure-ServiceNow correlator POC, ~6K LOC Python
Archived total10

Grand total: 317 commits across all projects


Code Size

AreaFilesLinesLanguage
Platform backend (src/)9712,257TypeScript
Platform tests (test/)7117,364TypeScript
Platform UI (ui/src/)10211,434TypeScript, React, CSS
Platform scripts75,440TypeScript
Connectors6520,681Python
Active code subtotal34267,176
Documentation19879,874 (~505K words)Markdown
Active grand total540+147,050

Archived / predecessor code

ProjectFilesLinesLanguage
graph-mongo (first MVP)304,541TypeScript
demo-scanner (GitHub POC)93,045Python
archive/connectors (correlator)85,991Python
sv0_architecture (docs v0)186,685Markdown
sv0-docsv1 (docs v1)238,328Markdown
Archived total8828,590

Combined total: 630+ files, 175,640 lines across active + archived projects

Test Suite

  • 496 unit tests (52 test files)
  • 84 integration tests
  • 580 tests total, all passing

Claude Code Usage (All sv0 Projects Combined)

MetricCount
Conversations / sessions99
Human-typed messages2,150
Tool approvals clicked24,085
Claude assistant turns41,536
Claude tool invocations24,109
Conversation data on disk~1.1 GB

Per-session averages

  • ~22 human messages per session
  • ~244 Claude tool calls per session
  • ~11 automated actions per human prompt

Breakdown by project context

Project contextSessionsHuman msgsTool approvalsClaude tool uses
sv0-platform631,44615,71815,734
graph-mongo (first MVP, Jan 29 – Feb 15)253656,3046,305
sv0-workspace (root)7931,7261,730
sv0-documentation338242244
sv0-connectors1269596

The graph-mongo sessions represent the first platform MVP (Jan 29 – Feb 15), before the codebase was restructured into the current three-repo architecture. The workspace root sessions (Feb 10–12) were cross-repo coordination sessions.


Estimated Manual Developer Comparison

What would building the same codebase take a senior full-stack developer working solo, without AI assistance?

ComponentEstimated time
Architecture & design (data model, 9 MongoDB collections, entity types, evidence packs, temporal tracking)2–3 weeks
Backend (12K LOC: ingestion pipeline, evaluator, storage adapter, workers, API, evidence packs with SHA-256 integrity)4–6 weeks
Test suite (17K LOC: 580 tests including MongoDB integration tests)2–3 weeks
UI (11K LOC: React, TanStack Query, ReactFlow graph visualization, ELK layout, 21 pages, 41 components)3–4 weeks
Python connectors (21K LOC: Azure/Entra ID discovery, ServiceNow integration, Graph API, ARM API, transformer pipeline)3–4 weeks
Documentation (80K lines / 505K words: architecture specs, ADRs, product docs, research)2–3 weeks
Scripts & infrastructure (seeding, Docker, deployment, CI)1 week
DevOps (production deployment, MongoDB backups)1 week
Total18–25 weeks (4.5–6 months)

Summary

  • Actual build time: 35 days
  • Estimated manual equivalent: ~22 weeks (5.5 months)
  • Speedup factor: ~4–5x
  • Ratio: Claude performed ~11 automated actions for every human prompt

Token Usage & Cost Estimate

What would this project have cost on the Claude API instead of the Max subscription?

Raw Token Counts

CategoryTokens
Base input tokens1,623,633
Prompt cache write tokens184,935,098
Prompt cache read tokens3,900,715,014
Total input tokens4,087,273,745 (~4.1 billion)
Output tokens637,422
Combined~4.1 billion

Model Distribution

ModelTurns% of usage
Claude Opus 4.633,16079.7%
Claude Opus 4.55,38412.9%
Claude Sonnet 4.62,3085.6%
Claude Sonnet 4.56951.7%

Estimated API Cost

Pricing based on published Anthropic API rates (per 1M tokens):

  • Opus: $15 input / $75 output / $18.75 cache write / $1.50 cache read
  • Sonnet: $3 input / $15 output / $3.75 cache write / $0.30 cache read
ModelCost
Claude Opus 4.6$7,695
Claude Opus 4.5$199
Claude Sonnet 4.6$99
Claude Sonnet 4.5$40
Total (with prompt caching)$8,034
Total (without caching, worst case)$52,827

Actual Subscription Cost

The entire project was built on consumer AI subscriptions:

ServicePlanMonthly cost
Claude Max (5x)Primary development tool (Claude Code)$100/mo
ChatGPT Pro + CodexCode review sessions, second-opinion analysis$20/mo
Total$120/mo

Note: Claude Max 5x usage limits were nearly reached by end of project — likely upgrading to Max 20x ($200/mo) soon. Codex (OpenAI) was used for independent code review sessions but did not hit plan limits.

vs. Claude API (Pay-Per-Token)

ApproachCostMultiplier vs subscription
Actual subscription cost$120/mo1x
Claude API (with prompt caching)~$8,03467x
Claude API (no caching)~$52,827440x

vs. Hiring a Developer (5.5 months equivalent work)

Developer tierAnnual comp (fully loaded)5.5-month costMultiplier vs $120/mo
Senior engineer (remote/mid-market)$180–240K$82,500–110,000688–917x
Senior engineer (NYC/SF tech)$300–400K$137,500–183,0001,146–1,525x
Staff/Principal (NYC fund-level)$400–600K$183,000–275,0001,525–2,292x
Quant/infra eng (top-tier hedge fund)$600K–1M+$275,000–458,000+2,292–3,817x

Fully loaded = base salary + equity/bonus + benefits + payroll taxes + office/equipment (typically 1.3–1.5x base).

Summary

  • $120/mo in AI subscriptions ($100 Claude Max + $20 ChatGPT Pro) delivered ~$8,000 worth of API compute and ~5.5 months of senior developer output
  • Prompt caching alone saved ~$45,000 compared to uncached API calls
  • At the extreme end: a top-tier NYC fund engineer costing $275K+ for 5.5 months of equivalent output vs. $120/mo in subscriptions — a ~2,300x cost difference for comparable deliverables (full-stack platform, 67K LOC, 580 tests, 505K words of documentation, production deployment)

Scaling to a Startup Team: Subscription Options

This project was built on individual consumer subscriptions. What happens when you scale to a small startup (< 10 people)?

Available plans (as of Feb 2026)

The "5x" and "20x" labels refer to multiples of the Pro plan ($20/mo) usage limits — not multiples of each other.

PlanPer seat/moUsage levelClaude Code?Min seatsTermsAdmin/SSO
Pro$201x (baseline)Yes (limited)1ConsumerNo
Individual Max 5x$1005x ProYes1ConsumerNo
Individual Max 20x$20020x ProYes1ConsumerNo
Team Standard$25–30~2x ProNo5CommercialYes
Team Premium$150~6x ProYes5CommercialYes
API (pay-per-token)usage-basedunlimitedYes (BYOK)CommercialN/A

The key pricing tension: Individual Max 5x ($100) is cheaper per seat than Team Premium ($150) and gives comparable Claude Code access, but runs on consumer terms without team management. A solo founder gets more raw usage for less money on Max, but a team loses central billing, SSO, and commercial IP protections.

Cost comparison for a 5-person dev team

ApproachMonthly costClaude Code?Notes
5x Individual Max 5x$500YesConsumer terms, no central billing
5x Individual Max 20x$1,000YesConsumer terms, no central billing
Team Standard (5 seats)$125–150NoCommercial terms, but no Claude Code
Team Premium (5 seats)$750YesCommercial terms + Claude Code + SSO
API (equivalent workload)~$8,000+YesBased on actual token usage from this project

Key distinctions

Consumer terms (Max plan):

  • Individual subscriptions, no team features
  • Anthropic's consumer ToS applies — code inputs may be used for training in general, though Claude Code documentation states it does not train on user code
  • No SSO, no admin controls, no centralized billing
  • Nothing prevents each team member from having their own Max account

Commercial terms (Team/Enterprise):

  • Explicit prohibition on using data for training — no exceptions
  • Stronger IP and confidentiality protections
  • SSO, domain capture, admin controls, centralized billing
  • Required for compliance-sensitive industries

Practical guidance

  • Pre-revenue startup: Individual Max accounts ($100–200/person/mo) are the most cost-effective path. Many early-stage startups do this.
  • Post-revenue / with clients: Team Premium ($150/seat/mo) is the responsible move — commercial terms protect both your IP and your clients' data.
  • API path: Only makes sense if you're building products on top of Claude. For developer productivity (Claude Code), subscriptions are 40–400x cheaper than API pricing based on our measured usage.

Sources: Claude Pricing, Claude Max Plan, Claude Code for Business, Consumer Terms Updates