Working proof · built for this proposal · Astro 5 + TypeScript
A two package Astro monorepo built to demonstrate the exact discipline this project demands: a donor component library powering a second brokerage site with its own brand, metadata, analytics and CMS models, and an automated gate proving that zero donor branding or configuration reaches the built output. Every result below is from real runs, reproducible from the repository.
Prepared by Yinka Aderibigbe.
astro-reuse-proof/
├── packages/donor-ui the "existing production site" component family
│ ├── Hero.astro all content by props, all color through CSS custom properties
│ ├── ReviewList.astro static first, accessible, no DOM cloned slides
│ ├── PropertyCard.astro
│ ├── tokens.css DONOR brand tokens ← never imported by the new site
│ └── donor-site-config.ts donor GA4 id, Turnstile key, site name ← never imported
└── apps/harborline the new site (fictional brand, created for this demo)
├── src/styles/tokens.css the ONLY brand tokens in the build
├── src/layouts/Layout own metadata, canonical, GA4 id, favicon
├── src/content + data cities collection, professionals, reviews, featured
├── public/admin/config.yml Decap models mirroring the brief's profile fields
├── src/server/leadValidation.ts honeypot, min submit timing, dedupe, formats
└── tests/reuse.test.mjs the automated gates below
Donor components carry zero embedded brand: content arrives by props, color by custom properties. The donor package still ships its own tokens and config, exactly like a real donor repository would, and the consumer simply never imports them. The gate then proves the negative.
$ node tests/reuse.test.mjs PASS donor leakage gate: 0 occurrences of 5 donor markers across 11 built files PASS JSON-LD valid on index.html (@type RealEstateAgent) PASS JSON-LD valid on professionals/amara-okafor/index.html (@type Person) PASS robots.txt and llms.txt shipped PASS review content appears exactly once (no DOM cloning) PASS lead validation: honeypot, min submit timing, format, duplicate replay all enforced ALL GATES PASS
The leakage gate scans every built HTML, CSS, JS and text file for the donor brand name, donor GA4 id, donor Turnstile key, donor brand hex and donor font. It fails the build on a single occurrence. The same pattern extends to any marker list a real donor repository needs.
| Gate | Target | Result | Verdict |
|---|---|---|---|
| Desktop Performance | ≥ 90 | 100 | PASS |
| Desktop Accessibility | ≥ 95 | 100 | PASS |
| Desktop Best Practices | ≥ 95 | 100 | PASS |
| Desktop SEO | 100 | 100 | PASS |
| Mobile Performance | ≥ 85 | 100 | PASS |
| Mobile Accessibility | ≥ 95 | 95 | PASS |
| Mobile Best Practices | ≥ 95 | 100 | PASS |
| Mobile SEO | 100 | 100 | PASS |
| LCP | < 2.5 s | 0.2 s desktop / 0.8 s mobile | PASS |
| CLS | < 0.1 | 0 | PASS |
The QA loop, demonstrated on itself. The first audit run scored accessibility 95 desktop and 89 mobile and flagged two defects: insufficient contrast on the accent color and a console error from a missing favicon. Both were fixed in the consumer layer only, by darkening the Harborline accent token and adding the site favicon; the donor package was not touched. The re run produced the table above. That catch, fix in the right layer, re verify cycle is the working method, not a story about it.
Open the live demo site to browse it directly; screenshots below for the record.
Decap CMS models mirroring the brief's professional profile fields (portrait, contact, biography with word guidance, specialties, areas, languages, credentials, individually toggled social links) plus cities, source verified reviews and featured properties with active flags and display order. A reusable city page template driven by a content collection. A typed server side lead validation pipeline covering honeypot, minimum submit timing, format checks and duplicate replay, unit tested, with Turnstile server verification and the owner webhook documented at the endpoint boundary. robots.txt and llms.txt shipped in the build.
Built August 31, 2026. Every gate result and Lighthouse figure above comes from real, repeatable runs against the built output, and the first audit run's findings and their fixes are recorded exactly as they happened. Harborline Realty is a fictional brand created for this demonstration. Repository available for the audit stage.