FLOOD RECORDING STUDIOS — MIRROR — Master Breakdown (Living Doc)

The old WordPress site (floodrecordingstudios.com on WP Engine) snapshotted page-for-page and served from our own Cloudflare Worker flood-recording-studios. This is M1 of the takeover — Mirror → DNS → Stage on the Website Builder → Flip — so the studio never goes dark while the domain moves and the new site is built elsewhere.

Preview: https://flood-recording-studios.appolis.workers.dev · the real domain attaches in M2 · the flip (M4) re-points the domain at the Website Builder and keeps this Worker as rollback.

Single source of truth for this app. Updated every time it changes. Board: Kosmos project flood-recording-studios — read 🧠 STATE (note_2696) first; the takeover brief and decisions are in 🗺 ROADMAP (note_2694); the measured as-is site is 📋 INVENTORY (note_2695); traps in note_2760; the lane split with the Website Builder in 🔗 LOCKSTEP (note_3084).

• Deploy: cd "F:\Claude Code\flood-recording-studios" && npx wrangler deploy
• Version: GET /__version → {"app":"flood-recording-studios","role":"mirror","version":"0.1.0","date":"2026-09-09","pages":4,"indexable":false}
• Build: v0.1.0 (2026-09-09) — 🎙 THE MIRROR EXISTS. scripts/harvest.js snapshots the 4 real pages (home, /booking/, /contact/, /availability-policy/) and everything they reference — 259 assets, 15,787,504 bytes, including the whole 76-file WordPress media library — into site/, rewrites same-origin URLs to root-relative, removes the WordPress discovery and endpoint tags that would point at doors this Worker does not have (api.w.org, EditURI, shortlink, alternate/oEmbed/RSS, generator), the unused Contact Form 7 script/style, WP Engine's per-response Cloudflare bot-detection beacon, and WordPress's own robots meta (the visibility decision moves into lib/routes.js — see §2), keeps the Cal.com booking embed, the GoHighLevel chat bubble, the Google Maps iframe and the Google Fonts loads exactly as they were, and adds the one content addition Tyler asked for (todo_2698 #8): a LocalBusiness JSON-LD block and an address / hours / socials strip above the copyright line. The Worker serves site/ through the static-assets binding, answers /__version, sends the 7 leftover theme-demo pages and every WordPress endpoint to 410 Gone, redirects www → apex, and — while SEARCH_VISIBLE is false — marks every host noindex, so nothing here can be indexed before Tyler says go-live. Tests: npm test → 20 pass, 0 fail (routes 8 · site 12). Not a registered Appolis app (see §7). Blueprint: note_3090 on the Flood board.

1. WHAT IT IS

A static snapshot, served from our own Worker. Nothing here talks to WordPress at run time: the pages are HTML files, the assets are the theme's own CSS/JS/images copied to the same paths, and the Worker's only logic is the routing table in lib/routes.js. That is the whole design — the mirror exists so that the DNS move (M2) has a target that cannot break, and so that the Website Builder (the sister lane) has a byte-exact reference of what the old site looked like.

What it deliberately is not: a redesign, a CMS, a booking system or a store. Booking and chat stay on Cal.com and GoHighLevel until the flip; afterwards Hermes owns them (Hermes board todo_3078). The new site is built on the Website Builder (project website-builder), never here.

2. THE HARVEST — scripts/harvest.js

node scripts/harvest.js (or npm run harvest) rebuilds site/ and harvest/ from the live WordPress site; --offline redoes only the rewrite from the raw pages already on disk:

1. Fetches the 4 real pages with a Chrome user-agent. WP Engine's bot challenge blocks curl but not node's fetch with a browser UA (measured 2026-09-08) — the raw HTML is kept under harvest/raw/ for provenance.
2. Collects every same-origin reference — absolute, protocol-relative (//floodrecordingstudios.com/…, which is how the theme loads Slider Revolution's CSS), JSON-escaped (https:\/\/…) and root-relative /wp-content|/wp-includes — from the pages, then from every stylesheet (url(), @import) and script it downloads, then the whole media library via /wp-json/wp/v2/media. Everything is saved under site/ at the same path. WP Engine rate-limits bursts (HTTP 429 after ~35 requests, measured), so requests are spaced 250 ms apart and a 429/5xx is retried with exponential backoff; a genuine 404 is recorded in harvest/manifest.json → missing[] and tolerated by the tests (the live theme has five: three owl.video.play.png paths, a LESS placeholder @{img-path}, a misspelled Barlow-SemiBoold.ttf), a 429 left behind fails them. Known quirk carried over as-is: the theme's placeholder.png files are JPEG data under a .png name; browsers sniff the bytes and render them, so they are served unchanged.
3. Rewrites each page: same-origin URLs → root-relative (the canonical link keeps the absolute real domain, on purpose); strips the tags listed in the build line above; injects the JSON-LD before </head> and the address strip immediately before the © Copyright paragraph.
4. Writes sitemap.xml (the 4 real URLs on the real domain), robots.txt, 404.html, and harvest/manifest.json (pages, every asset with its byte size, the media-library count, what was stripped, what was missing). Prints one line: HARVEST pages=4 assets=N missing=M media=K bytes=B stripped=S — at v0.1.0: HARVEST pages=4 assets=259 missing=5 media=76 bytes=15787504 stripped=68.

Re-run it whenever the WordPress site is edited before the flip (the routine session's page-count check tells you when). The tests then prove the new snapshot is complete and closed.

The two deliberate changes. Everything else on the pages is byte-for-byte what WordPress served.

1. The address, hours and socials (Tyler, 2026-09-08, todo_2698 #8). The old site never states the address; listings do. Every page now carries <script type="application/ld+json" id="frs-localbusiness"> (LocalBusiness · 6003 28th St E Unit B, Bradenton, FL 34203 · +1 941-705-9356 · floodrecordingstudios@gmail.com · Instagram/Facebook · the map pin's coordinates) and <div id="frs-address"> with the same facts plus "Sessions by appointment — booking open 24/7", above the copyright line. The JSON-LD carries no opening hours on purpose: nobody has measured the studio's hours, and "open 24 hours" would have been invented.
2. Search visibility becomes one switch, and it is OFF. Measured 2026-09-08: all four live pages carry <meta name='robots' content='noindex, nofollow' /> — WordPress's "Discourage search engines" switch is ON in the WP Engine install. Tyler's call, 2026-09-08: "Just keep it off because we're going to be remaking everything anyways. When we're done with the staging site and ready to push it live on our platform, then we can just not make anything be hidden." So: WordPress is not touched, the harvest strips the baked-in tag so the snapshot's HTML is clean, and the decision lives in lib/routes.js as SEARCH_VISIBLE = false — while it is false the Worker sends x-robots-tag: noindex, nofollow on every host, the apex included, and /__version reports "indexable":false. At go-live one constant flips; the tests and this section are coupled to it on purpose, and a preview host stays noindex either way.

3. ROUTES — lib/routes.js (pure, tested)

wrangler.jsonc: assets.directory = ./site, binding = ASSETS, run_worker_first = true (the Worker sees every request first, then falls through to the assets), html_handling = auto-trailing-slash, not_found_handling = 404-page, workers_dev = true.

4. FILES

worker.js            the Worker — turns lib/routes.js decisions into responses
lib/routes.js        VERSION · SEARCH_VISIBLE · ORPHANS · DEAD · decide(url, opts)
scripts/harvest.js   the snapshot script (§2)
site/                the snapshot: index.html, booking/, contact/, availability-policy/, wp-content/, wp-includes/, sitemap.xml, robots.txt, 404.html — 266 files
harvest/raw/         the 4 pages as WordPress served them (provenance)
harvest/manifest.json what was harvested, byte for byte
test/routes.test.mjs the routing table + the search-visibility switch (8 tests)
test/site.test.mjs   the snapshot is complete, closed to WordPress, carries the changes (12 tests)
wrangler.jsonc · package.json · APP_BREAKDOWN.md (this file) · README.md
.gitattributes       `* -text` — byte-exact checkouts (this machine's git has core.autocrlf=true; without it every clone/worktree would rewrite 64 text files and break the manifest test)
.gitignore           /node_modules/ /.wrangler/ *.log /archive/ — ANCHORED: the theme ships isotope.pkgd.min.js under site/…/vendor/node_modules/, and an unanchored ignore dropped it from every clone (found by cloning); the WP Engine backup zip lives in archive/, never committed

5. TESTS

npm test (= node --test test/routes.test.mjs test/site.test.mjs) — no dependencies, node's own runner. 20 tests, 0 failing at v0.1.0.

• routes: the version door · www → apex keeps path + query · all 7 orphans gone with and without the slash · every WordPress door gone, itself and beneath · the site is hidden from search on every host while the switch is off · flipping the switch makes only the apex indexable · no false positives (a path that merely contains a dead word) · VERSION agrees with package.json and this doc's build line.
• site: the 4 pages exist with their WordPress titles · no page points back at the old host except canonical + JSON-LD · canonical stays absolute · discovery/endpoint/CF7 tags, the robots meta and the WP Engine beacon gone · every local reference on every page resolves to a file in site/ · stylesheets pull nothing that is not on disk · Cal.com on /booking/, GoHighLevel + fonts everywhere, the map on /contact/ · the address strip and JSON-LD on every page, above the copyright line · sitemap/robots/404 · no image, font or script on disk is secretly an HTML error page · this doc carries measured numbers · the manifest agrees with the disk and no 429 was left behind.

6. DAILY CHECKS (the routine session, after the first deploy)

1. https://flood-recording-studios.appolis.workers.dev/__version → version equals lib/routes.js VERSION and "indexable":false until Tyler says go-live.
2. The 4 preview pages → 200 with their titles; /sample-page/ → 410; /wp-json/ → 410; every page answers with x-robots-tag: noindex, nofollow.
3. /booking/ in the browser pane shows the Cal.com iframe (app.cal.com/flood-studios-amn0ru/…).
4. The live WordPress page count (/wp-json/wp/v2/pages?per_page=100&_fields=id) is still 11 — if it moved, re-run the harvest and the tests.
5. After M2 add: the real URL 200 · MX still aspmx.l.google.com · a test mail delivers · the apex still answers x-robots-tag: noindex (the studio stays hidden until the new site goes live).

7. COMPLIANCE

The Appolis connector does not open a flood-recording-studios app (measured 2026-09-08: appolis_compliance_list → "this connector does not open flood-recording-studios"), so nothing can be attested under this name. The suite-wide directives open on 2026-09-09 were read against this build: overlay lock (cmp_8c8134e18f) — the mirror opens no overlays of its own (the chat bubble is GoHighLevel's); schema proof (cmp_c7b85307b7) — no database, no migration; token discipline (cmp_18c8b53201) — no AI call anywhere; email identity (cmp_34f01cf47c) — sends no email (there is no form); progress on every wait (cmp_progress0001) — static files, no wait; Hephaestus first (cmp_b80e65baad) — the mirror makes no AI call to route; answers are a form (cmp_0f71c92a20) — this lane asks through Kosmos decision forms, not text lists. If the mirror ever grows a form, a wait or an AI call, it registers as an app first and attests then.

8. FEATURES DECK (plain-English rows)

• The studio's site, on our own servers. The old website was copied page for page — every picture, every style, the booking calendar and the chat bubble — and now runs from our own infrastructure, so moving the domain later cannot take the studio offline.
• Still hidden from Google, on purpose, with one switch to change it. The old site had been telling search engines to ignore it. That stays true while the new site is being built, but it is now one setting in one file instead of a checkbox buried in WordPress — and the site says out loud whether it is hidden.
• Dead pages finally retired. Seven leftover demo pages from the theme (a dance school!) were still public. They now answer "gone", which tells Google to forget them.
• The address, at last. The old site never said where the studio is. Every page now shows the address, phone, email, hours and social links, and tells search engines the same in the format they read.
• Nothing to break. No WordPress, no plugins, no database, no logins — just files and a tiny routing table, with 20 automated checks proving the copy is complete.

9. NEXT

• M2 DNS — enumerate every consumer of the domain (MX, SPF/DKIM/DMARC, Cal.com, GoHighLevel, Google Business, the old Appointy link) → zone in Tyler's Cloudflare with identical records → nameservers → verify site + mail → then apex + www as custom domains on this Worker. The domain is not in Tyler's own GoDaddy account; he reaches it through GoDaddy Pro delegate access (all 18 records are in INVENTORY note_2695; there is no DKIM and no DMARC, and a pay CNAME to GoDaddy pay links is a consumer the move must carry).
• M3 the new site on the Website Builder (that lane) · M4 the flip; this Worker stays as rollback until WP Engine is cancelled. SEARCH_VISIBLE flips at go-live, not before.