Avalon Build Package
Build Roadmap
Six phases from consent foundation to native apps and white-label engine. MVP (Phases 0-2) delivers a fully functional visualization app with 57 features. Post-MVP phases add the stylist marketplace, commerce layer, and sovereign inference migration.
Rev A September 2026 Prepared by Swift Innovation
6
Build phases, Phase 0 through Phase 5
57
MVP features across Phases 0-2
40
Post-MVP features, Phases 3-5
515
MVP all-in agent-directed hours
Tier 1
$8K / mo
MVP in 9 months · Phase 3 by month 13
Tier 2
$15K / mo
MVP in 5 months · Phase 3 by month 7
Tier 3
$22K / mo
MVP in 3 months · Phase 3 by month 4.5
MVP feature (tagged Y in feature matrix)
New from gap analysis (Rev E addition)
Post-MVP feature
MVP Scope
0
Consent & Foundation
MVP
~2-4 wks
Legal infrastructure and prospect access
Before a single face image is processed, the consent architecture must be in place. Illinois BIPA and CCPA compliance is not a post-launch concern - it is the foundation. This phase also delivers the prospect portal so Cliff can track progress in real time.
Consent
CN-01Consent gate UI and immutable consent record CN-02Retention scheduler and hard purge job CN-03Consent versioning and policy text hashing
Platform
P-09Prospect portal for Cliff
1
Core Engine
MVP
3-6 mos
Visualization engine, corpus ingestion, and infrastructure
The engine is the core technical risk. Phase 1 is all about proving it works before building the application around it. Hair segmentation, Face Lock composite, fidelity gating, and the ArcRouter integration are all Phase 1 deliverables. The corpus shoot pipeline and held-out eval set also happen here so quality measurement is built in from day one.
Engine
E-01Hair region segmentation E-02Mask refinement and edge handling E-03Conditioning assembly E-04Generation call via ArcRouter E-05Face Lock composite E-06Fidelity gate E-07Edge blend and color harmonize E-08Variant fan-out and job queue E-09Versioned engine HTTP contract E-14Direction-of-change classifier E-18Photo-based hair type classification
Capture
C-02Client-side quality gate C-03On-device landmark and crop
Consent
CN-04Data subject access / delete / export CN-06Age gate
Catalog
CT-01Style taxonomy and graph schema CT-08Volume and length bands on every style CT-09Seed shoot ingestion pipeline
Platform
P-01Avalon graph instance with backups P-02Object store with TTL and encryption P-11Held-out eval set construction P-12Capture rejection-rate monitoring by skin tone
2
Full Application
MVP
longest phase
Consumer app, feasibility engine, stylist layer, and safety
Phase 2 is the largest build phase. The PWA shell, all capture flows, the match system, catalog admin, the full consumer UX, and the initial stylist directory all land here. By the end of Phase 2 the app is usable end-to-end and ready for a closed beta with stylists and early consumers.
Capture
C-01Guided selfie capture with live coaching C-04Upload from photo library C-05Retake and coaching loop
Engine
E-16Product-effect render mode
Feasibility
F-01Feasibility rule table and verdict engine F-02Feasibility panel in result view F-03Growth estimation with honest ranges F-04Damage and physical-cost copy per style
Catalog
CT-02Style catalog admin CRUD CT-03Reference image ingestion with release record CT-04Style browse / filter / search
App
A-01PWA shell and design system A-02Result view with variant selection A-03Before and after compare A-04Lookbook save and organize A-05Share card with AI disclosure A-07Account and authentication A-08Feedback capture A-09Onboarding and expectation setting A-14Style match quiz and preference capture A-15Side-by-side reference photo comparison A-16Multi-style session flow
Stylist
ST-01Stylist directory stub ST-02Send-to-stylist consultation handoff ST-08Geographic stylist discovery
Platform
P-03ArcTrace integration with face-image exclusion P-04Rate limiting and abuse controls P-05Moderation queue P-07Per-session cost and usage metering P-13Third-party subject attestation and takedown route P-14Input and output safety filtering
Post-MVP
3
Engine Hardening
Fine-tune, stylist supply, and edge cases
After beta data comes in, Phase 3 applies it. Segmentation fine-tunes on the proprietary corpus, stylist portfolio upload is live, and edge cases like baby hair treatment and second-vendor failover are addressed. The 360 preview and cap sizing features are conditional on what beta usage signals.
Engine
E-10Second vendor failover E-11Segmentation fine-tune on corpus E-15Reductive render with hairline ground truth E-17Edge and baby hair treatment E-19360-degree multi-angle style preview
Feasibility
F-06Cap sizing and head shape (conditional)
App
A-17Editorial content layer
Stylist
ST-04Portfolio upload with signed release capture
Platform
P-06Fidelity eval harness
4
Commerce & Growth
Revenue activation, full stylist marketplace, product affiliate
Phase 4 is where the business model fully activates. Subscriptions, product affiliate, the complete stylist marketplace with booking and dashboards, push notifications, and lifecycle features all ship. Commerce on the wig/unit side is conditional on whether that market segment proves out in Phase 3.
Capture
C-06Multi-angle capture for side views
Engine
E-12Hair color-only fast path E-20Lace tinting and density simulation (conditional)
Feasibility
F-05Growth tracking over time
Catalog
CT-06Style suitability against hair profile CT-07Trending and seasonal merchandising
App
A-10Push and lifecycle notifications A-12Between-install lifecycle prompts A-18Hair health timeline and condition tracking
Stylist
ST-03Claimed stylist profile ST-05Booking link and deposit capture ST-06Stylist dashboard and lead view ST-07Stylist verification workflow ST-09Stylist-curated product recommendations
Commerce
CM-01Product attach to style CM-02Affiliate link tracking and attribution CM-03Subscription and paywall CM-06Wig and unit commerce object CM-07Buy button and vendor handoff
Platform
P-15Corpus and weights portability path
5
Scale & Optionality
Native apps, brand data platform, and white-label engine
Phase 5 is the optionality layer. Native iOS and Android apps for distribution at scale. The brand demand dashboard converts aggregate visualization data into a product brands will pay for. White-label packaging turns the engine into a licensable product. Sovereign inference migration cuts external AI provider dependency entirely.
Engine
E-13Live AR filter (color and simple silhouettes)
App
A-11Native iOS and Android apps
Commerce
CM-04Brand placement slots CM-05Aggregate demand dashboard for brands
Platform
P-08Migration to sovereign Grid inference P-10White-label engine packaging