Public provenance
Every source class answers a different question.
Manufacturer pages establish identity and stated specifications. Marketplaces reveal listing sentiment and owner wording. Long-form owners reveal tenure and mechanisms. App reviews expose the software layer. Editorials add structured testing when the method is real. One household record exposes physical behavior under a declared scope. None deserves to impersonate the others.
This page owns source eligibility and provenance. The dated listing, storefront, brand, and app census stays separate because those public counts change.
Source eligibility rules
- The exact model must be identifiable from a model number, SKU, ASIN, listing, image, mechanism, or publication context.
- Generic “MeoWant” complaints remain brand-level unless the model can be attributed reliably.
- Manufacturer and seller claims stay labeled and are not rewritten as tested facts.
- Marketplace ratings remain listing-specific and dated; they are never averaged or summed into unique owners.
- Early reviews and invited reviews are useful for setup and first impressions, not long-term reliability.
- Repeated mechanisms receive more editorial weight than dramatic isolated anecdotes.
- Conflicting values, commercial relationships, tenure, variants, exclusions, and missing evidence stay visible.
Current public-source census
Listing counts, ratings, distributions, and app-store snapshots belong to the dated census rather than this methodology page.
What each source class can support
Manufacturer pages and catalog feeds
Canonical product identity, SKU, official dimensions, compatibility, setup, warranty language, seller claims, manuals, accessories, and launch timing.
Cannot establish: independent reliability, actual noise, real drawer interval, owner satisfaction, or large-cat comfort.
Retailers and marketplaces
Dated rating distribution, demand signals, verified-purchase wording, setup impressions, recurring complaints, model variants, and seller-channel differences.
Cannot establish: comparable defect rates, representative tenure, or one universal score across duplicate and imported listings.
App stores
Companion-app ratings, version history, network, notifications, pet-profile, placement, weight, and multi-device complaints across compatible boxes.
Cannot establish: exact-model hardware reliability when the app review does not name the box.
Long-form owner reports
Cat behavior, urination posture, cleaning burden, maintenance cadence, faults, replacements, support outcomes, and changes after months of ownership.
Cannot establish: prevalence from a convenience sample.
Independent editorial and creator reviews
Structured setup, comparison, dimensions, demonstrations, and observed fit when the exact model, method, tenure, and commercial relationship are disclosed.
Cannot establish: independent testing when the article merely repeats maker copy or the product was gifted without meaningful method.
Brand-level review platforms
Delivery, support, replacement, return, accessory, and broad satisfaction themes across the company.
Cannot establish: model performance unless the review explicitly names the model.
Household telemetry and incident records
Observed device state, physical behavior, events, faults, drawer cycles, and bounded incidents under a declared test scope.
Cannot establish: population reliability or behavior outside that household and window.
Curated exact-model ledgers
The complete SC01, SC02, SC03, and SC09 review pages publish qualifying manufacturer, retailer, owner, social, editorial, and video source tables. SC10 publishes the household claim and evidence records alongside its current external marketplace and official-hosted snapshots.
Editorial research artifacts
The repository includes the full model synthesis and a separate expanded size-and-source correction record. These artifacts drive model writeups, improvement requests, traffic clusters, source adapters, duplicate-listing logic, and the unknown-unknown queue.
Raw copyrighted review text is not republished as a corpus. Public pages use concise paraphrases, dated counts, source links, and bounded claims.
Corrections and freshness
Marketplace counts, ratings, prices, policies, stock, app behavior, official-hosted review systems, and product revisions can change. Corrections should identify the exact route, statement, model, source URL, source date, and requested change.
Read the methodology · See update history · Submit a correction