AP Agency · Project Evidence Report

SwiftBrand

AI-powered business name generator and full brand operating system — name generation with live domain checks and confidence scoring, tiered deep-dive brand intelligence reports, brand kit (AI logo + palette + fonts + voice), brand-anything templates, social kit, pitch deck, merch mockups, programmatic SEO engine covering 180+ pages across 30 niches, first-party analytics, and a multi-tier SaaS billing layer.

$515K – $1.13M traditional build cost
6.1 person-years of engineering
Built in ~3.5 weeks (Jun 15 – Jul 9, 2026)
The Defensible Quartet
$515K
– $1.13M
Traditional Build Cost
What a normal agency shop would charge to design, architect, and ship this from scratch.
6.1 yrs
Engineering Effort
Mid estimate — two independent models: productivity ratio (LOC/dev/day) and COCOMO II post-architecture.
6 ppl
× 12 mo
Traditional Team × Schedule
The team this replaces — derived from COCOMO II schedule equation applied to 24,858 LOC.
~15.1×
Build Speed
Shipped in ~3.5 calendar weeks versus the 12.2-month normal schedule. One engineer, zero meetings.

The Real Number

6.1 engineering-years delivered under 1 operator — the sustained ~6-person team you'd otherwise staff, run by one person supervising Steve. ~15× faster than the conventional 12.2-month schedule.

0 incidents — every action backed up, owner-scoped, approval-gated

Traditional cost avoided: $515K – $1.13M (supporting figure — the headline is the leverage, not the dollar).

The Hard Evidence

No estimates — counts straight from git and the live codebase.

Signal Value
Code LOC 24,858
JavaScript 24,628
Python (tooling) 230
Total files 609
Git commits 122
Distinct edit days 16
Lines inserted 144,542
Lines deleted 21,770
Calendar span 2026-06-15 → 2026-07-09
Calendar weeks 3.5
Contributors 2
Defects tracked & resolved 8
Regression tests 36
Test runs logged 641
What's Actually In It — Feature Enumeration

Verified by direct inspection of route handlers, module headers, and public/ directory (on-disk name: CompanyNameFinderACN). Started as an AI company-naming tool; grown via 40+ recent commits into a full multi-tenant Business Operating System. 107 distinct Express routes · 206 pre-generated static SEO pages · 9 table groups + job-store/analytics/taste-engine DBs.

Naming Engine (V1 + V2)
  • V1 coordinator — orchestrates generation → domain-check → trademark → market analysis → report → email, pluggable across Claude/OpenAI/DeepSeek
  • V2 staged pipeline — Stage A territories, Stage B persona fan-out, Stage C1 deterministic zero-AI quality gates (word-list based), Stage C2 anchored LLM judge, Stage D domain check, Stage E curated trial-10
  • Self-tuning ranking — scores every generated name by template + thumbs-up/down + domain-availability feedback, biasing future generation; Taste Engine records name events/impressions for personalization
  • Per-report AI call-volume breakers — free 250 calls/600K tokens, paid 400/1.2M — with a "zero-name honest failure" mode (no fabricated results on total failure)
Domain & Trademark Research
  • DNS + WHOIS availability across TLDs; social-handle checks routed through an IPRoyal residential rotating proxy
  • USPTO trademark search, common-law (Google-scraped) unregistered-mark detection via Puppeteer
  • International scrapers — CIPO / EUIPO / UK IPO (no public API exists for these)
  • US state-registry checker — top-10 states, free tier
  • Domain-history (WHOIS history), 0–100% composite confidence score across all checks
  • Smart TLD selection — .com + 4 industry-relevant TLDs
Brand OS — Pillar 1: Site/Doc Generation
  • Brand kit — logo, palette, font, voice, generated from the naming report
  • Pure-templating branded business card / letterhead / invoice — no AI cost at render
  • Website-gen — real deployable one-page branded site, downloadable standalone HTML
  • "Talk to your website / business card" — plain-English instruction → DeepSeek proposes an allowlisted set of field edits (never raw HTML), validated against a strict schema before applying — a prompt-injection fence reused across every AI-edit surface; version history + rollback
  • Merch mockups (t-shirt/mug/tote/cap/sticker/box, CSS/SVG-drawn, PNG export); social kit (profile/banner/post/story); investor pitch deck (keyboard-driven, print-to-PDF); business-plan generator v2 (Chart.js, Document/Slides toggle)
Business-OS Pillar 3 — Company Memory & Skills
  • Persistent structured facts store (key/value/source/timestamp) + a learned-skills store
  • AI extraction agent + transparency panel — feeds downstream document generation
Business-OS Pillar 4 — The Boardroom
  • One-click P&L → branded shareholder-grade deck
  • Architecturally enforced separation — the compute module is PURE (zero AI/network — numbers can never be AI-generated/edited by construction); the AI module is narrative-only and cannot touch numbers
  • Charts are inline SVG only — CSP blocks external chart libs
Document Center
  • DB-driven template registry — "adding template #41 is a row, not a release"
  • AI drafting with an allowlisted field/clause fence
  • Explain / rollback / edit history per document; financial category added recently
Agency / Multi-Tenant White-Label (Phase 4)
  • Tenants table — slug provisioning with reserved-slug guard; seats (owner/admin/member, seat-limit enforced)
  • Host-resolution middleware — <slug>.swiftbrand.studio → req.tenant; vanity subdomain paints agency name/logo/colors, "Powered by SwiftBrand" footer, signup hidden for team members
  • Agency API — provision, seats, branding upload; seat-invite emails with claim tokens
  • E2E + Playwright-vision verified per commit — seat 200 / non-seat 403 / root 200
Billing
  • Stripe AND LemonSqueezy code both present; LemonSqueezy active — webhook fails CLOSED on missing secret/signature
  • Proration via PATCH-not-new-checkout, feature-flagged off by default
  • Billing portal + cancel in Settings (staging)
Auth
  • JWT auth — email/password + verification
  • OAuth — Google / GitHub / Facebook via Passport
  • API-key issuance for programmatic/business tier — bulk research, generate/revoke keys
Programmatic SEO
  • Registry-driven niche pages — 206 static HTML live, each with a real domain-checked name grid, FAQPage + BreadcrumbList + WebApplication JSON-LD, hub-and-spoke linking, sitemap
  • --modifiers mode generates niche × modifier pages with genuinely distinct AI name sets per tone
Analytics
  • First-party, privacy-safe — daily-salted visitor hashing (no raw IP), UTM/source classification, bot flagging, zero external dependency
Admin Console
  • Dashboard, economics, traffic views
  • Jobs — list/cancel/restart
  • AI-provider live toggle; name-template CRUD + toggle; SEO analyze; UAT run
  • User Journeys panel — auto-captured narrator+critic screenshot gallery per flow
Infrastructure / Scale
  • Postgres-backed durable job queue (SQLite fallback), pluggable admission backend — in-memory default, swaps to Redis/BullMQ for true horizontal multi-box scaling with pub/sub progress, zero caller changes
  • Report-scheduler admission control — paid lane drains before free lane, per-user concurrency cap
Recent Hardening (chronological proof of active engineering)
  • Payment webhook hardened fail-closed
  • Site/document AI agents hard-pinned to DeepSeek for security
  • EU cookie-consent banner (CF-IPCountry); Privacy Policy page (LS MoR framing)
  • Domain-checker false-positive fix
Planned / Not-Yet-Live
  • PLANNED — full 50-state registration (top-10 live now; OpenCorporates path noted)
  • PLANNED — international trademark APIs (scraping now — genuine constraint, not a shortcut)
  • PLANNED — LS proration (present in code, flagged off)
  • PLANNED — billing portal/cancel + version rollback marked "staging only" (not confirmed promoted)
How We Costed It — Methodology

Model. The same model with the same constants is applied to every project in this portfolio — there are no per-project complexity multipliers and no schedule compression applied. The generator is console/backend/tools/evidence_merge.py.

Low bound. Productivity model — 24,858 lines of production code ÷ 25 finished LOC per developer-day ÷ 21.7 work-days per month = 45.8 person-months (3.8 person-years). The 25 LOC/day figure is the conservative industry baseline for finished, tested, reviewed code (McConnell, Code Complete); it excludes generated scaffolding and build artifacts.

High bound. COCOMO II post-architecture — PM = 2.94 × (24.86 KLOC)1.0997 = 100.7 person-months (8.4 person-years). Applied nominally, with no effort-adjustment multipliers, so the ceiling is not inflated.

Mid (reported). The arithmetic mean of the two models: 73.2 person-months = 6.1 person-years. This is the figure quoted throughout this report.

Rate assumption. $135,000/year fully-loaded blended developer (salary, benefits, overhead) = $11,250 per person-month. Applied uniformly to all three effort bounds: low $515K, mid $824K, high $1.13M.

Calendar. Git-verified: 2026-06-15 → 2026-07-09 = 0.81 months of real calendar. COCOMO's schedule equation gives the calendar a conventional team would need: TDEV = 3.67 × 73.20.28 = 12.2 months at a sustainable 6.0 engineers. Speed advantage = 12.2 ÷ 0.81 = 15.1×.

Honesty & Caveats

LOC includes generated SEO pages. The 609-file count and some of the 24,858 LOC include the 180+ programmatically generated HTML SEO pages under public/. The core server code (routes, AI agents, report generator, brand-kit, templates, analytics) accounts for the large majority of the JavaScript LOC; the SEO output is templated but is still shipped, maintained, and served as live, indexed pages.

Backup files are counted in the file total. The working directory contains *.bak_* timestamped backup files (STEVE-1 safety protocol). These are excluded from LOC counts but inflate the file total modestly; the active source file count is lower.

Single contributor. All 122 commits and 144,542 inserted lines are from one engineer (STEVE-1 / AI-assisted development). Traditional productivity models assume human teams; the AI advantage is the source of the ~15× speedup, which is the point of this report.

Numbers are verbatim from the evidence payload. No figures were recomputed or adjusted for this report. If a number appears inconsistent with another, it reflects the raw git/DB artifact as measured.