Nobi Landing – Vercel/Netlify Starter

Local dev (optional)

1. Install Node 18+
2. npm install
3. npm run dev

Deploy to Cloudflare Pages

• Simply push your changes to the main branch and Cloudflare Pages will build and deploy automatically.

Deploy to Vercel (deprecated)

• Push this folder to a GitHub repo.
• Import the repo in Vercel.
• Framework Preset: Vite
• Build Command: npm run build
• Output Directory: dist
• Hit Deploy.

Static assets

Put images/videos under public/media and reference them as /media/filename.ext.

Static page content

npm run build renders the homepage, pricing, and product components into HTML using React's Node renderer. It does not require Chromium. Browser JavaScript still mounts the same components for the bundle slider, navigation, and forms. Blog and glossary pages retain their existing content prerendering. The blog's static Markdown renderer preserves explicitly authored literal HTTPS links with rel="sponsored nofollow", matching their runtime MDX links. It keeps other HTML escaped. A direct affiliate anchor records a network referral; it does not invoke the Shop Nobi product-click handler or prove an order.

The homepage is served through an exact Cloudflare Pages rewrite so the empty index.html shell remains available for other client-side routes. This avoids serving homepage text as the content of unrelated pages.

Run npm test to build the site and check page content, pricing facts, URL forms, asset references, and signup attribution. Register additional pages in both scripts/prerender-pages.js and scripts/render-pages.jsx.

Local pricing page

Run npm run preview:pricing with Node 18+ and open http://127.0.0.1:5174/pricing. The page shows a bundle slider, the full bundle table behind "See all bundles", what counts as a message, and the going-over settings. The bundle list lives in src/constants/pricing.js. The previous /pricing/ideas preview URL redirects to the pricing page on loopback hosts. Rebuild to see edits.

npm test builds the prerendered pages and checks the bundle slider, the bundle table, the pricing facts, and existing page regressions.

Shop Nobi search pages

Published search pages keep the API's catalog refresh timestamp separate from recorded editorial revisions. shop-nobi-editorial-revision-dates.js records substantive source revisions by slug; generic rules retain independent revision dates. Dates are in UTC and are not an inventory check or a claim of public release time. Ecru and the silver FAQ rule use their verified deployment receipts for known release dates. Only changed editorial fragments that survive final publication contribute a date; suppressed or unmatched guidance does not make a page newer. The sitemap uses the later valid catalog or editorial date from the same published data as HTML. Visible labels distinguish “Catalog refreshed” and “Content revised”. Stories, categories and product-offer check dates keep their separate meanings. Never set all dates to the build time or a shared module's mtime. When materially revising an authored entry or rule, update its recorded revision and provenance; a no-op rebuild or date-only edit is not a new content revision. The offline tests cover suppression and stability.

Generic Best ... search-page titles (including the from Top Stores suffix) are replaced at publication with catalog-backed titles and descriptions by scripts/shop-nobi-catalog-copy.js. Related-search labels also remove that generic ranking wording at publication, retaining custom labels, link order, destinations and original query fallbacks. The copy names the displayed listing count and up to two actual retailers, and only invites price comparison when multiple displayed records have prices and currencies. The shared shop-nobi-publication-quality.js rule also removes fixed monetary assertions from source guide sentences and FAQs and supplies a price summary computed from the displayed records. Mixed currencies are not combined into a range. Clear visitor-entered search examples retain their budget amounts. When publication filters remove records, dependent source guide/FAQ copy is withheld because it may describe products no longer shown. Empty pages offer a search path, have matching noindex metadata in HTML and the browser, and are omitted from the search sitemap and llms.txt links. Populated pages return to normal publication automatically on a later build when their listings return. Reviewed editorial guidance remains subject to the same publication checks. The green ink/refill page also withholds three independently verified non-green retailer options at normalization. Each hold requires the exact page, merchant, variant ID and literal URL. Other pages and unresolved color options are retained; this is not a keyword-based color filter or a catalog mutation. The existing filtered-output path rebuilds counts/prices and withholds dependent guide/FAQ claims. Only an actual exact hold carries a membership revision into surviving publication text; the original catalog refresh timestamp stays unchanged. Empty output remains undated. It does not certify every remaining listing's ink color. The generic-copy branch replaces 16 reviewed exact suede-care FAQ pairs and three waterproof-selection FAQ pairs on uncited Best ... from Top Stores source pages. The tailored answers ask shoppers to check exact-model care or waterproof specifications instead of assuming protection from a material or finish alone. Thirty-four reviewed jewelry FAQ pairs containing unsupported skin-suitability assurances also receive exact-item composition guidance. The everyday-wear question retains ring fit, finish and care advice. Each correction requires its original slug, title and complete question/answer pair. Titles without from Top Stores are permitted only for those exact records; this does not widen the suede or waterproof rules. Other custom titles and cited source guidance are preserved. The rule does not match broad care or water-related keywords or change reviewed guides, other FAQs, or product data. Unfamiliar wording remains subject to separate review. Six reviewed lens FAQs and seven related guide paragraphs receive exact-source corrections in shop-nobi-blue-light-guidance.js. The guidance separates tint, prescription design, filtering and anti-reflective coatings without promising eye-strain or sleep benefits. It runs after wearable guidance, preserving that correction on metal vision-care pages. Original slug, title and complete text must match; cited, authored, empty and filtered pages remain protected. No new public citations or product fields are introduced. The generic branch also corrects 37 reviewed wearable guide paragraphs through . Each replacement requires its exact original slug, title and whole paragraph. Tailored copy retains useful fit, component and finish comparisons while removing unsupported skin, safety or performance assurances. All FAQs and other paragraphs remain unchanged. Authored, cited, empty and filtered pages keep their existing protections. The static source revision is recorded only when new text survives final publication; it does not change the catalog refresh timestamp or date unmatched pages. Ten reviewed textile paragraphs and nine related FAQs receive exact-source corrections in . These retain useful fiber, fill, fit and care-label comparisons without inferring skin suitability from an organic, cotton, fleece or down-alternative label. Adult baby-tee styling is distinguished from infant clothing. Existing source protections, product data and prior corrections remain unchanged. Only surviving corrected text contributes the rule’s static editorial revision. Three reviewed cosmetic pages receive five exact paragraph and nine FAQ corrections in . The copy separates product labels from unverified sourcing, skin-suitability or performance claims and directs shoppers to exact-formula ingredients and instructions. Existing buying advice and all product data are preserved. Original source identity and whole-text gates retain authored/cited/empty/filtered protections; only surviving new fragments contribute a static editorial revision. Nine reviewed garment pages receive nine exact FAQ corrections and one blazer paragraph correction in . Care advice uses the selected garment's label, lining and trim instead of granting wash or dry-clean permission from a fiber name. The existing exact-passage helper protects custom, cited, empty and filtered pages; only surviving changed text contributes the rule's revision date. Products and earning links remain intact.

The brown-chukka page has three reviewed paragraphs and three FAQ answers corrected through the same exact-passage helper. It distinguishes named upper finishes and model-specific weather/fit checks without promising rain protection, break-in outcomes or walking comfort. The styling FAQ, computed price summary and all product records remain unchanged; only surviving corrections revise its date.

The single laundry-detergent page has two reviewed paragraphs and two FAQs corrected by shop-nobi-detergent-dose-guidance.js. Formula-specific dosing, washer and fabric instructions replace universal half-dose and compatibility assurances. The detergent and cosmetic rules share shop-nobi-exact-guidance.js, which retains exact source-identity and complete-text gates. Each rule keeps its own revision date, and existing product cards and other advice remain unchanged. This is not a general fact-check of qualitative product advice. The same publication data supplies HTML, metadata, FAQ schema and browser hydration; source guides, product ordering, product facts, affiliate links, Pinterest feeds and Stories are unchanged. This is a site-wide rollout, not a controlled click-through experiment, and it does not guarantee Google's chosen title/snippet or indexing.

The existing laptop-stands page also has a concise manufacturer-specification comparison in its buying guide. scripts/shop-nobi-laptop-comparison.js requires the exact reviewed listing name, brand, merchant, product ID and variant URL for each row. A changed or removed listing loses its row and source; empty or publication-filtered pages show no comparison. Prices and stock remain in the unchanged product cards. The optional semantic table scrolls on narrow screens and follows the guide's existing landing-experiment position. These specifications are attributed references, not hands-on performance rankings.

The womens-khaki-pants page uses the same comparison layout with clothing-specific headers. Its four reviewed models distinguish petite/tall hiking cuts, cotton-twill wide-leg trousers and cotton reverse-sateen double-front pants. The unresolved 25-inch travel color is excluded from the comparison. Exact identity guards and publication-filter suppression apply; no size availability, office suitability or universal best ranking is inferred.

scripts/prerender-shop.js builds one page per Shop Nobi search at /shop/s/<slug>, with the products in the HTML, one page per top category at /shop/c/<slug>, and a list of popular searches on /shop. It also writes /shop/pins.xml, an RSS feed with one item per search page, which Pinterest reads to publish selected product Pins linking to existing search pages. It also writes one feed per top-level category at /shop/pins/<slug>.xml, one for each Pinterest board, and lists them with their item counts in /shop/pins/index.json. Pinterest has its own selection in scripts/shop-nobi-pin-selection.js: explicitly available, priced USD products with safe native photos; a relevant earning product is preferred within the page's existing results. Product/photo identities are deduplicated across board feeds, with alternate eligible products keeping distinct pages represented. The pictured product supplies factual Pin copy, and mixed earning pages start with #ad and the commission disclosure. Canonical GUIDs stay the same; refresh times do not become publication dates. This does not change search ranking, website cards, SEO previews or Web Stories.

scripts/shop-nobi-editorial-copy.js combines reviewed, page-specific buying guidance dictionaries, including the original sterling-silver-rings guide. The page writer applies each selected page's title, description, guide and FAQs together to rendered HTML, metadata, structured data and embedded browser content. Product facts, shopping links and snapshot dates stay intact. The shared source plan, Pinterest feeds and Web Stories retain their existing copy; replacements apply only to the explicitly selected search pages. The same publication path supplies the pencil-skirts buying guide, replacing the fixed price-range and general fabric-care advice with exact-item checks and an invitation to enter preferences in the existing search box. Catalog prices, product links, snapshot dates, feeds and Stories are unchanged. The clothing/accessory copy module also covers fourteen observed-query pages: red-blouses, navy-cardigans, multicolor-sweaters, gray-cargo-pants, yellow-midi-dresses, green-wallets, pink-swim-trunks, blue-navy-blouses, pink-hoodies and charcoal-hoodies, navy-suits, blue-navy-suits, blue-navy-casual-dresses and green-maxi-dresses. Their guidance distinguishes sizing, construction and variant choices, replaces unsupported superlatives and care/performance claims, and suggests visitor-entered searches. These edits do not certify stock, merchant claims or ranking improvements. shop-nobi-additional-buying-copy.js covers black-stud-earrings, cream-rugs, brown-mini-dresses, sundresses, silver-watches and black-maxi-dresses, amethyst-earrings, emerald-jewelry, brown-suede-boots brown-boots, diamond-bangles, aquamarine-rings, mens-signet-rings, amethyst-jewelry, gold-white-bangles and purple-violet-jewelry, gold-bracelets, blossom-jewelry, saw-accessories, table-saw-accessories, miter-saw-accessories, ecru-puzzles, red-electric-guitars, mens-diamond-bracelets and yellow-gold-diamond-earrings through the same publication-only path. It replaces allergy-prevention and blanket cleaning claims with quantity, measurement, component, size-variant and exact-item care checks. Saw guides distinguish accessory systems from add-ons and category placement from verified compatibility; they provide buying comparisons, not cutting or setup instructions. The Ecru clarification uses the and to distinguish the named puzzle line from a color and separate difficulty from piece count. Red electric guitar guidance distinguishes exact finish from a builder name, compares condition and specifications, and avoids unsupported tone or tuning promises. Men's diamond bracelet guidance separates link shape from stone coverage and replaces universal stone, allergy and care claims with exact-item disclosures. Yellow gold diamond earring comparisons clarify selling quantity, component materials, dimensions, piercing requirements and the stated stone-weight basis.

The care, material, apparel, equipment and jewelry guide modules correct specific observed claims against retained catalog records. The fit modules distinguish product names from compatible models, material labels from care instructions, and accessories from complete products. The accessory-identity module separates bag charms from jewelry, selected metal construction from a color or karat label, and exact sizing services from blanket resizing promises. These corrections do not authenticate materials, establish skin suitability, or certify fresh stock. The parameterized guide-review test checks removal of recorded defects and agreement between visible copy, FAQ schema and browser hydration while retaining product records, query and snapshot date.

The equipment-claim module also covers bicycle chains, hubs and shifters, tent footprints, snowboard binding parts, camping lanterns, earplugs, electric massagers and beige/brown electric guitars. Its buying guidance separates complete products from accessories, exact compatibility from broad category names, and finish labels from materials. It does not supply servicing procedures or unsupported safety and performance promises.

scripts/shop-nobi-pin-editorial.js records reviewed page/product/photo pairs, with the reason and observed Pin. A reviewed page is omitted from Pinterest if its chosen product or photo changes or becomes unavailable, until another hero is reviewed. This does not ban its merchant or country of manufacture. URL checks do not certify resolution or composition; new native heroes still need visual review, including recognition at a 236px card width. Source media stays intact. Existing published Pins are not promised to update when RSS changes, and no feed reconnect or artificial GUID is used to force a repost. Non-affiliate pages also receive pictured-product copy, without a commission claim. Availability comes from the saved page snapshot; it is not rechecked when Pinterest imports the item. Mature pages stay out of every feed.

Prebuilt CUCKOO cards use text, price and the existing product link, with the programme's exact Disclosure: Compensated Affiliate. notice. Its current terms require Awin-provided imagery, and no matching blender creative was found. The normalized view omits CUCKOO pictures from HTML, embedded hydration data, structured data and image-dependent previews, Stories and Pins. Stored catalogue images remain intact, and other merchants keep their existing presentation. This is a prebuilt-page rule, not certification of live assistant results, every SKU's commission or other channels. Feed connections and canonical GUIDs are unchanged; existing published Pins are not edited or deleted by this rule.

Prebuilt ESR product records are omitted while permission for the retained product materials and published prices is unresolved. This happens before cards, hydration, product schema, previews, Stories and Pin selection, so those surfaces cannot retain a partial product promotion. Source catalog tuples, inclusion and assigned links are untouched; this is not a store rejection or a live-search filter. Generic guide text is not rewritten. Mixed pages retain the other cards; an ESR-only page retains its editorial content without product records or image-dependent Story/Pin items. No supplied homepage creative is presented as a product link. Feed/board mappings and canonical GUID rules remain unchanged, although eligible hero, frame and item sets can change. Existing Pins are not removed or edited by this build rule.

The outside accounts behind this (Search Console, Bing, IndexNow, Pinterest) are recorded in nobi-marketing's search engines and discovery sites doc. It fetches the page data from https://api.nobi.ai/v1/shop/pages. SHOP_PAGES_API_URL points the build at another API base address. If the fetch fails, the build fails, so Cloudflare keeps the last good deploy live.

The API may initially serve its previous answer while rebuilding the cached listing. The build accepts listings up to 15 minutes old, accommodating the API's ten-minute cache and a rebuild allowance. Older valid listings receive at most four reads of the same route, two seconds apart, under one shared 60-second budget. Persistently old data, invalid generation times, HTTP errors or malformed responses stop the build before any Shop pages or feeds are written. No cache-bypass query or source mutation is used.

Related searches keep their existing prebuilt page links. If a known related search falls outside the build's file limit, its link opens /shop?q=... with the original query instead of becoming plain text or pointing to an unbuilt page. Older embedded data falls back to the related search's title. Root and category navigation use the same exact generic-title topic parser as related search labels: Best Boots from Top Stores and Best Boots From Top Stores become Boots. The initial Best and Top Stores casing remain exact; custom labels, routes and ordering stay unchanged. Still-generic populated metadata is normalized after authored clarification selection, preserving those guides and FAQs. Newly recognized metadata and related-label fragments receive their own recorded editorial revision only when they survive final publication. Existing normalized text retains its earlier revision; navigation does not trigger a blanket date refresh.

In the browser, a search page keeps its prebuilt products above the live search, and a category page keeps its searches under the search box, until the assistant bundle draws live products. A refused or failed live search therefore leaves the prebuilt content in place. On /shop, browse links remain below the idle assistant without hiding its headline or suggestions. Twelve categories are expanded; native disclosures retain the remaining category and search links in the HTML. The mounted browse section uses an H2 instead of repeating the initial page intro. Only a live product draw hides this navigation. A click on a prebuilt product sends the API the same "Click On Search Result" event as a live product, with the same entry point, "AI Search Bar". Prebuilt clicks omit the conversation ID so the API records them without waiting for a live conversation to be saved. This also means a prebuilt card gives no cross-merchant AI conversation credit after a shopper uses the assistant. Live assistant results keep their own conversation attribution. Crawlers, automated browsers and local copies send none. In a smoke test's tab, the click carries the bundle's nobi_smoke_test marker, so the API drops it as it drops the bundle's own smoke-test events.

Each page view of a search page also sends one "Displayed Product Collection" event, the event the bundle sends for the products it shows. Its shownProducts list holds every prebuilt product, in order: productVariantId, isPaidLink, and owningMerchantId when the store is not Shop Nobi, which every event is posted under. A product built before the API sent ids keeps its place with a null productVariantId, and a page with no ids sends no event. The event's shownProductsStartPosition is 1, so the product at list index i (from 0) is at place 1 + i, the positionInGrid its click sends. It carries pageSlug and the collection name Prebuilt search page, so a report can tell it from the live search. The build keeps a product's productVariantId and owningMerchantId only when the API sent both as real ids, and prebuilt clicks then carry both too. The same people-only rules apply.

Each search page with at least two short guide pages and two product pictures, one of them a Shopify picture whose 3:4 cut is at least 640 by 853 pixels (the build downloads the start of each poster to check), also gets a Google Web Story at /shop/stories/<slug>, an AMP story built by scripts/shop-nobi-web-story-plan.js and scripts/shop-nobi-web-story.js. It shows a cover, the guide in pages of at most 180 characters, up to three products with nofollow store links, questions that fit, and a last page that links back to the search page with utm_source=google_web_stories. The search CTA on the sofas story instead links to the /shop search home with utm_campaign=story_to_search and utm_content=sofas, so a visitor must enter their own query. Its product links remain available. The search page links to its story, and sitemap-shop.xml lists the stories. Each store link goes through https://api.nobi.ai/v1/shop/story-go with the plain store link and the story's slug; the API counts the tap and forwards to the store, through the store's affiliate link when one pays us. The story's amp-analytics sends https://api.nobi.ai/v1/shop/story-view the slug and the id of each page in view. A product page's id is product-<place>-<variant id>-<store id> when the API sent both ids, and the API records that product as shown, with its place as the event's shownProductsStartPosition; it ignores every other page and every bot. Stories get only the file room left after the pages. Mature pages get no story. To check one, run npx amphtml-validator dist/shop/stories/<slug>.html.

Store links can earn a commission. For each product the API may send storeLink (the finished link for search pages) and paidLink (true when the link earns a commission). The website never builds or tags a paid link itself, and keeps one only for paid products. A search page keeps the plain store link in its HTML and holds the paid link only in base64, so crawlers never see it. A person's click opens the paid link, with its sub-ID parameter (the one whose value is shop_nobi) set to <slug>.<visit id>, at most 50 characters and stable within the page view (src/utils/shopNobiClickId.js). The click event records that finished link along with incoming traffic tags and referrer. A crawler's click opens the plain link with Shop Nobi's tags. Plain prebuilt links remove any stale nobi-conversation-id. A paid link, on search pages and in stories, gets rel="sponsored" and a small "We may earn a commission" notice under it (CommissionNotice in src/utils/shopNobiPage.js).

Every published Shop Nobi search page includes its eligible displayed product photos in sitemap-shop.xml, deduplicated per page and capped at 1,000 images. Existing merchant restrictions and unsafe-image filtering apply before sitemap generation. Categories, Stories and image-free pages keep their normal URL entries without product-image tags. Image discovery and indexing are not guaranteed; Google documents Search Console verification for externally hosted image domains.

Search entries in sitemap-shop.xml use their resolved catalog/editorial modification date. Other entries retain their own refresh/offer-check timestamp, with the API export time as a fallback.

The last build step, scripts/submit-indexnow.js, tells IndexNow (Bing, Yandex and the other engines that share it) about new or changed Shop Nobi pages from sitemap-shop.xml and NEW canonical authored pages from sitemap-pages.xml, including hubs and blog articles. Shop pages are compared with the live sitemap by address and lastmod. Utilities are selected from the authored sitemap and compared with the live utility-revisions.json, which prerendering writes from each delivered HTML page and its referenced local image bytes. This covers rendered facts, metadata, disclosures, links and shared JS/CSS bundle fingerprints. A shared bundle change can conservatively notify all utilities, but unchanged builds and same-day content changes are distinguished without a build clock or git. Utility sitemap entries omit lastmod rather than invent a modification date. Existing nonutility authored pages are not replayed for date-only changes. Only canonical HTTPS destinations on nobi.ai are accepted. Duplicate addresses are removed before batching up to 10,000 URLs. The key file is public/b20309cb84825b184afe0ffcf36630a7.txt; the key is public by design. Only a Cloudflare Pages build of main from the real page data submits. Local builds, preview builds and builds with SHOP_PAGES_API_URL skip it. An unreadable live Shop sitemap skips only Shop notifications; unknown live authored sitemap metadata skips authored notifications, while unknown utility revisions skip only utilities, preserving new nonutility and Shop notifications. A missing built file, incomplete revision artifact or failed submission remains visible in the build log. A missing referenced local image stops prerendering. The IndexNow step never fails the build and saves dist/indexnow-submission-receipt.json with exact public URLs, batch HTTP acceptance/failure and explicit skip reasons. No keys or response bodies are saved. Google sitemap submission is a separate authorized release operation; IndexNow does not notify Google or prove indexing. It runs before Cloudflare publishes the build, so an engine that fetches a new page in the first minutes can still see the last deploy. Acceptance acknowledges the notification; it does not prove indexing, search traffic or commission. The first revision-manifest deployment cannot compare against a nonexistent live manifest. A separate bounded initial notification is an operator action; the build never treats unknown live metadata as unchanged or force-replays it.

The GitHub workflow nightly-shop-nobi-rebuild.yml rebuilds the site every day at 10:00 UTC, after the API's overnight refresh, by calling the Cloudflare Pages deploy hook in the CLOUDFLARE_DEPLOY_HOOK_URL secret.

To build without the network, serve the test data: node tests/shop-nobi-pages-server.js 4789, then SHOP_PAGES_API_URL=http://127.0.0.1:4789 npm test.

Gift chooser prototype

/tools/gift-chooser uses twelve source-dated product examples for coffee, painting and charging. A visitor explicitly chooses an interest, per-item USD budget and applicable preferences. The local filter returns up to three explained options, including fewer or no options when constraints cannot be met. It does not run a model, backend search or automatic query.

The main shopping link opens the actual existing Nobi collection, without pretending the answers are carried over as filters. Native retailer URLs are secondary source/details links; no affiliate template is constructed from saved objects. An optional editable request can be copied into the supported manual search at https://nobi.ai/shop. These absolute canonical handoffs also keep a local preview from presenting fixture Shop results as real inventory.

src/content/giftChooserProducts.js holds the reviewed records and explanations; src/utils/giftChoice.js handles native form data and selection. Prices and availability are dated snapshots, not live quotes. Tests use actual product records, native FormData and the actual React result component through Vite SSR, plus the normal built page and utility-revision checks. They are not a browser interaction test or verified purchase/commission evidence.

Election Night checklist

/election-night-survival-kit is a static, nonpartisan November 3, 2026 checklist with a disclosure above optional shopping links. It reuses the standalone product-exit handler and an actual offline Java builder record for the exact Baseus CR11 variant in src/content/electionNightLink.js. It does not construct an affiliate template or promise stock, arrival, runtime or a purchase. The Marley link opens the existing cost calculator and carries only incoming UTM tags so its existing offer-exit event retains acquisition context. No calculator or model call was added. Original diagrams, three Pins and the silent 15-second video are under public/media/election-night-*; the video loads only when requested. The page CSS is loaded with the main stylesheet so the static HTML stays styled without JavaScript. After November 4, review the dated seasonal copy before promoting it again.

tests/election-night-page.test.js checks static content, disclosure order, media/metadata/sitemap, the exact product exit and calculator attribution tags.

Cove/Cubby comparison utility

/tools/cove-vs-cubby renders the comparison facts in HTML and runs its floor-space calculator locally in the browser. It keeps the site's existing floating assistant behavior outside /shop; the calculator sends no query. Manufacturer facts, checked date, exact catalog variant IDs and reviewed server-built store links are authored in src/content/coveCubbyComparison.js. Recheck these inputs before changing the dated price claims. Do not construct a second affiliate URL template in the website.

Purchase clicks reuse the standalone visit handler with the actual Shop Nobi Utility surface, page URL, acquisition tags and variant ID. Unfiltered click reporting retains the surface; the current backend report validator does not yet accept it as an exact entry-point filter. Standalone utility clicks have no conversation-derived search text; the current report's /shop/s extractor also leaves their page slug empty. Page URL, details.query, traffic tags and the final affiliate reference remain available for acquisition joins. Do not create a synthetic search or conversation. Run the footprint tests with node --test tests/cat-footprint-fit.test.js; the full offline website suite also checks the built comparison HTML.

Marley Spoon subscription cost utility

/tools/marley-spoon-cost is one authored SUBSCRIPTION/PER_DELIVERY offering, not a physical product catalog entry or generic MCP/service search. It models only the dated two-person/three-meal plan and food-discount schedule in src/content/marleySpoonCost.js. Food rounds once in integer cents before shipping is added. Delivered-box totals never promise future checkout prices.

src/content/marleySpoonLink.js contains actual Java store-link-builder output from the independently verified non-catalog owner. Validation runs on import and fails the build for missing/invalid reviewed link configuration. Do not construct a second affiliate redirect or create a fake product to represent an offer. New prices, codes or ownership changes need source review and fresh builder output before publication.

The merchant exit shares person/visit/smoke/local and beacon/fetch behavior with product exits, but sends a distinct Click On Offer generic event with the offering kind, price basis, authored owner, final link and acquisition context. It includes no product/variant or conversation IDs and sends no calculator selections. Owner metadata is authored, not server-resolved. Current counted physical-click SQL selects only Click On Search Result, so offer adoption needs a separate raw-event-to-affiliate-reference ledger. The current exact entry-point report filter also omits Shop Nobi Utility.

Run node --test tests/delivery-cost.test.js tests/shop-nobi-offer-click.test.js for calculator, configuration and click behavior, then the full offline website suite for static content and existing product regressions.

Baseus independent-screen selector

/tools/baseus-two-screens compares only the US NU1 Air Mac B0114400 and RD1 Pro WIN B0114600. The local pure selector separates documented conditional display fit, known incompatibility and missing evidence, then evaluates laptop charging independently. It does not infer hardware capabilities from a laptop name, run a search, install software or claim physical testing. Manufacturer diagrams and software/manual sources are dated October 17, 2026. The NU1 dual-4K marketing conflict and the mirrored manual's unverified hardware revision remain explicit.

The exact catalog IDs and actual offline Java builder outputs are authored in src/content/baseusScreenLinks.js. Missing product identity or invalid destinations fail the build. Exits use the existing standalone product handler with Shop Nobi Utility, a real variant and the fixed topic; selector answers are not added to click events. Generic event ingestion and reporting-filter registration are separate backend responsibilities. Tests cover decision boundaries, emitted exit attribution and both built HTML URL forms, sources and disclosures. The original diagram is under public/media/baseus-independent-screens.svg; Pin creative specifications are in agents/creative_specs/baseus_screen_pins.md.

MindTheColor digital-license checklist

/tools/mindthecolor-fit checks declared Photoshop, platform, installation, connectivity and single-user license prerequisites. Missing answers stay unknown; known mismatches take priority. It does not inspect hardware or guarantee compatibility. Changing an answer clears a previous result. Answers remain local. The source price is €179 plus VAT, checked October 17, 2026. Adobe Photoshop is separate. The original dependency diagram uses no merchant images or software UI.

This is one authored DIGITAL_SOFTWARE/PER_LICENSE offer, separate from physical search and MCP catalogs. src/content/mindTheColorLink.js contains actual offline Java builder output from the independently verified excluded Gumroad marketplace owner and its parameter configuration. Import validation rejects altered identity, source tags, duplicate/missing publisher parameters or incomplete provenance. Regenerate through the real builder after source/configuration changes; never construct a second affiliate URL in the website.

Exits reuse Click On Offer and preserve Gumroad's publisher parameter. Gumroad's checked link has no verified per-visit sub-ID, so our acquisition event does not establish an exact order-to-visit join. Marketplace access and the generated link do not prove retained commission or payout; account payout onboarding remains unfinished. This page does not change it. Run node --test tests/mindthecolor-fit.test.js tests/mindthecolor-offer-click.test.js, then the full offline suite for the built page, discovery metadata and existing regressions.

Duvet dimension comparison

/tools/duvet-cover-size compares one insert with up to three covers. Entered hundredths of an inch or centimetre normalize to integer micrometres, preserving exact cross-unit equality and nonzero differences. Outputs compare each axis, without a fit tolerance, automatic rotation or product ranking. Inputs remain in component memory. The existing event sender receives only fixed completion labels once per page mount after a valid comparison; normal local/crawler QA is excluded. The internal shopping link retains existing acquisition tags and a fixed tool marker, never dimensions. Event persistence and later report completeness remain separate from the local delivery tests.

USB-C shared power worksheet

/tools/usb-c-power-check compares two entered device targets with documented simultaneous port outputs in integer tenths of a watt. Unknown outputs remain unknown; the stated total never supplies a power split. Source examples apply only to Apple’s named 35W dual-port adapter. The worksheet does not certify compatibility, fast charging or safety. Inputs stay in component memory, with no storage or input telemetry. Static controls remain disabled until JavaScript is ready, avoiding a fallback form submission of the entered values.

Phone case checklist

/tools/phone-case-check records shopper-entered phone and selected case identities, self-reported listing evidence and separate optional charging/accessory questions. Notes remain in component memory and reset to unanswered; the tool does not match model names, certify compatibility, store notes or send a new analytics event. Apple examples illustrate named models only. The /shop landing page links to it.

Reviewed guide and FAQ corrections

The care, material, jewelry, apparel, equipment and safety shop-nobi-*-guide-copy.js modules contain page-specific corrections for care, fiber composition, product format and sizing. The shared editorial selector applies their metadata, visible guidance and FAQs together. Publication filters still take priority: empty or filtered listings cannot resurrect dependent inventory claims. These reviewed pages are a bounded batch, not a claim that every catalog guide has been fact-checked.

The reviewed apparel and household comparison dictionaries add twenty targeted guide/FAQ corrections through shop-nobi-editorial-copy.js. They distinguish product types, exact composition, documented performance and model compatibility. Product records, ranking, earning links and source queries remain unchanged. tests/shop-nobi-guide-review.test.js checks removal of each retained legacy claim and consistency across visible HTML, FAQ schema and browser data.

Landing-page product-click experiment

Populated prebuilt search pages participate in shop-landing-products-v1. A stable slug hash assigns each page to the existing guide-first layout or to products-first with an early same-page search jump. Crawlers and browsers see the same layout. Guide text, metadata, product ordering and earning links stay unchanged. The jump focuses the existing live-search input without submitting any query or navigating away from the original acquisition URL.

The existing product-display and prebuilt-click events carry experimentId, experimentVariant, pageSlug and landingVisitId. The visit ID belongs to the loaded document, does not identify a person, and needs no persistent storage. A nobi-analytics listener remains mounted after live results replace the prebuilt grid. It sends separate Shop Nobi Landing Outcome events for live product-click intentions and submitted messages, never a duplicate legacy click/search. Live click intentions count toward the primary measure only when joined to an accepted canonical click with matching conversation, product and shopper within a bounded time window. Unmatched intentions remain diagnostic. A message can be a question, so message submission is not labeled a completed product search. Shop Nobi Refine Search records successful focus of the search input. Existing local/bot/automation suppression and smoke-test tagging apply to all events.

Read raw event partitions, not the current reporting view (which drops product exposures and referrer fields). Count exposed visits with either kind of product click, divided by exposed visits; retain separate prebuilt and live outcomes. Filter known operator/test visits, retain missing-provenance counts, and never call client-supplied Google referrers verified people. Compare complete periods, page/query mix and cohort exposure; this is a page-level observational split, not visitor randomization or a revenue result. The plan is in agents/feature_plans/shop_landing_product_conversion.md.

Teak set comparison

The existing teak outdoor furniture page compares three reviewed Outer models by package contents, layout and ordering checks. Manufacturer model facts require exact catalog identities; filtered or removed products cannot retain their comparison rows or sources. No selected cushion color, current stock or offer is inferred. The catalog records, search query, earning links and landing experiment remain unchanged.

Cloth napkin comparison

The existing cloth napkin page compares three exact catalog models by format, set count, dimensions and specific care instructions. Dinner and cocktail formats remain distinct; linen composition does not imply common washing or drying instructions. Unknown Matouk shape details are omitted. Exact identity and publication-filter gates control each row and source, preserving product cards and landing allocation.

Men’s silver chain model comparison

shop-nobi-silver-chain-comparison.js distinguishes sterling-silver and steel base metals, plating, chain construction and model length options using three manufacturer sources. Exact catalog identity gates and publication-filter suppression match the existing comparisons. Product cards and landing experiment allocation remain unchanged; the table does not guarantee pendant compatibility, durability or stock.

Brown shirt model comparison

shop-nobi-brown-shirt-comparison.js compares documented fabric composition, cut and model-specific care for three shirts. Exact identity gates and publication-filter suppression preserve the existing comparison contract. All product records, source context and landing experiment allocation remain unchanged; model facts do not promise current stock, comfort or durability.

Brown suede boot comparison

shop-nobi-brown-suede-comparison.js compares the reviewed Bedell, Javier and Remy variants by silhouette, suede construction, entry details, sole and stated heel height. Exact name, brand, merchant, variant and literal URL must match for each row and source. Filtered or empty output suppresses the comparison. Brand size systems remain separate; the table makes no stock, traction or durability promise. All product records, links, ordering, catalog dates and landing allocation remain unchanged. The existing revision mechanism records this substantive copy change.

Hand lotion comparison

The existing hand-lotion page compares two exact reviewed models by labeled volume, scent description and labeled hand or hand-and-body use. It replaces unsupported price-to-richness and botanical-benefit claims. Model rows and references disappear when an identity changes or publication filtering applies. Product cards, prices, links, ordering and landing-experiment allocation remain unchanged. Manufacturer descriptions are dated evidence, not clinical tests or current stock assurances.

Audited product guidance

shop-nobi-published-claim-copy.js corrects ten existing bedding, phone-case, children's apparel, jewelry, bowl, footwear and bath-soak pages. It replaces unsupported health, weather, durability and universal care claims with page-specific selection criteria. It adds no product specifications, ratings or inventory assertions. Normal empty/filtered publication rules still apply; product records, prices, links, order and landing assignment stay unchanged.

The separate shop-nobi-catalog-fit-guide-copy.js batch corrects ten more existing pages. It distinguishes replacement storage lids from complete units, Pearl names from decoration, selected colors and pack contents, and exact intended use from category labels. Children’s necklace guidance requires maker age and sizing information rather than a universal chain length. The shared rendered tests cover both batches, including filtered and empty publication; product cards, source data, feeds and landing assignment remain unchanged.

One-model Bodum product comparison

/shop/p/bodum-tribute-12076-57us-34oz compares the reviewed Bodum Tribute 12076-57US, 34 oz seller variants. The normal current page export supplies eligible identities and original outbound links. Its stored snapshot amounts are not considered fresh offer observations. Only this page makes four bounded, anonymous build-time GETs: each reviewed retailer's product JSON and currency JSON. Exact IDs, SKU, title, capacity, USD, positive integer cents and single-unit purchase terms must match. Errors and mismatches remove that seller option. No merchant reads happen when the export lacks its exact eligible row.

A successful observation lasts 24 hours. Public static HTML always contains the useful no-price fallback. Prerendering generates generated/bodum-document.js outside dist; the narrow Pages Function imports that private document and serves prices only within their observation window. Known aliases redirect to the canonical route, and Function responses use no-store without validators. Expired or invalid observations use the static fallback; browser display and JSON-LD also expire while a page remains open. An empty eligible catalog leaves a zero-seller noindex product page without sitemap or internal discovery links.

Two fresh matching offers produce one Product with AggregateOffer. There are no invented reviews, availability guarantees or delivered-price claims. The initial catalog links are direct tagged retailer links with paidLink:false; this test measures visits and merchant click-outs, not established commissions. Existing paid-link handling remains intact if future catalog records supply earning links. The French-press category retains its products, ordering and landing assignment.

Offline coverage in tests/shop-nobi-bodum-product.test.js exercises source validation, actual prerendering, removed identities, static/private separation, expiry, aliases, HEAD/conditional headers, hydration and outbound attribution. Run the full suite with the existing local page-data fixture server, never live model-backed search. A normal npm run build generates the private document; then WRANGLER_SEND_METRICS=false npx --no-install wrangler pages dev dist --ip 127.0.0.1 --port 4820 --compatibility-date 2026-10-28 previews the actual Function. Vite alone serves the intentionally unpriced static fallback.

Travel and gemstone guide corrections

Five existing guides (suitcases, hunting-shooting-jackets, 14k-yellow-gold-huggie-earrings, ruby-rings, and ruby-earrings) use reviewed copy from scripts/shop-nobi-travel-gemstone-guide-copy.js. The guides compare model measurements, intended use, package quantity and disclosed stone details without inferring weather performance, comfortable fit or gemstone authenticity. The existing publication filter still withholds guide and FAQ copy when products are removed. Product cards, ordering, prices, links and experiment assignments are unchanged. The consolidated published-claims matrices cover these five pages alongside the previous thirty, including filtered and empty states.

Apparel and pearl/silver guide corrections

shop-nobi-apparel-pearl-guide-copy.js supplies reviewed guidance for women's white sweatpants, sterling silver bracelets, pearl rings, men's parkas and men's blue T-shirts. It compares fit, disclosed construction, components and care instructions without universal performance or material assurances. The existing publication filter, product records, shopping links and experiment assignments remain unchanged. The consolidated published-claims matrices now cover forty pages, including filtered and empty states. Clear phone cases were excluded because existing publication filtering already withholds their guide and FAQ.

Wallet organization and carrying-clothing guides

shop-nobi-red-wallet-comparison.js adds a model-level wallet table for three exact reviewed catalog identities. It compares sourced card pockets and closures, omitting disputed dimensions and unverified Bosca/Zegna color mappings. Rows and references disappear when the corresponding identity changes or publication is filtered or empty. Product cards and tracked retailer destinations are unchanged. shop-nobi-bag-sweatpant-guide-copy.js adds practical shoulder-bag and sweatpant guidance without universal spray, durability, wicking or washing claims. The same three published-claims matrices cover all forty-three reviewed pages, including wallet identity mutations and filtered/empty publication.

Appliance, eyewear and family-accessory guides

shop-nobi-appliance-play-guide-copy.js supplies reviewed guidance for espresso accessories, men's black sunglasses, stroller accessories, water tables and rice cookers. The copy distinguishes model compatibility, condition, package contents and measurements without universal replacement, performance or certification claims. Water-play basins are explicitly separated from the retained wooden furniture listing. All product records, paid destinations, CUCKOO disclosures and image restrictions remain unchanged. The existing three published-claims matrices cover forty-eight pages, including filtered and empty publication.

Leather sneaker shopping answers

shop-nobi-leather-sneaker-answer.js answers where-to-find questions using the first displayed model from up to three distinct retailers, in existing card order. Examples refresh with the displayed records and remain subject to the normal filtered/empty publication rules. Size, upper, construction and care guidance replaces unsupported material-quality, rain and resoling assurances. The same three published-claims matrices now cover forty-nine pages, including renamed and removed retailer examples; product records and destinations stay intact.

Material and intended-use shopping guides

shop-nobi-material-use-guide-copy.js supplies eight existing guides with practical comparisons of bag compartments and finishes, jewelry components, bedding pieces, child sizing and office-mat uses. The copy avoids general skin-safety, environmental, weather-performance and treatment assurances. Product identities, order, prices, paid links, images, disclosures and landing assignments stay unchanged. The same three published-claims matrices cover these additions, including filtered and empty publication.

Fit, configuration and care shopping guides

shop-nobi-fit-care-guide-copy.js supplies five existing guides for pencil cases, red outerwear, red shoulder bags, black office furniture and blue athletic shorts. The copy compares actual measurements, formats, closures, configuration and garment care rather than inferring reliability or performance. All product records and publication gates remain unchanged. The same three published-claims matrices cover these additions alongside prior corrections.

Ecru jigsaw puzzle line

shop-nobi-ecru-jigsaw-copy.js explains Artifact Puzzles’ Ecru line using its official FAQ and collection references. It distinguishes the line’s matte images and closer piece cuts from a beige artwork filter, then helps shoppers compare piece count, difficulty, cut design and finished dimensions. No named product, stock, price or mystery-puzzle feature is asserted in the guide. All twenty retained cards, source records, destinations and landing assignment remain intact. The existing three published-claims matrices cover the page and its filtered and empty states; references are withheld with suppressed guidance.