TrackAura for LLMs and AI Grounding

A canonical catalog of physical items with retailer-observed prices, structured schema, and honest coverage labels. Built to be machine-citable.

Independent — no display ads, no sponsored placements, no paid ranking, no paywalled catalog.

Why TrackAura for grounding

• One canonical identity per real-world item. Retailer listings, observed prices, and historical data are observations on that identity over time. Stable URLs survive retailer churn.
• Prices observed, not generated. Every price in the catalog is a direct, scheduled observation from a named Canadian retailer. No LLM-generated prices, no training-data pattern matches.
• Schema honesty matches page honesty. A page without fresh retailer prices does not emit Offer fields. A single-source page emits Offer, not AggregateOffer. Coverage tier and structured data are computed from the same observation set.
• Editorially independent by structural necessity. The canonical-reference business model only holds without ads. We don't have a placement to soften.

The catalog today

~32,000 canonical entities across the verticals below, indexed against ~4 Canadian retailers (Canada Computers, Newegg Canada, Vuugo, Visions Electronics) with new price observations every ~4 hours. Entity counts as of May 2026; counts grow per vertical ship.

Phase 0 is Canadian electronics. Phase 1 (collectibles with clean IDs — LEGO, TCG, sealed games, graded comics) is shipping now. The data model is geography-agnostic: one identity per real-world physical item, regardless of country, with country_code on each listing.

What you get per entity

• Identity — canonical slug, display name, brand, release date where applicable, hero image, encyclopedic description, parent / variant / predecessor relationships.
• Specifications — structured attribute key/value pairs typed per vertical (e.g. memory, panel type, refresh rate, socket). Leaves inherit from parents.
• Live listings — per-retailer URL, current price, stock signal, first-seen / last-seen timestamps, open-box flag.
• Price observations — full historical series of observed prices per listing. Same dataset feeds the trend charts and the worth-engine confidence band.
• Coverage tier — one of well_tracked, tracked, single_source, historical, or encyclopedic_only. Set per entity per render, mirrored in the structured-data output.

The worth engine, briefly

Every entity in the catalog gets a worth estimate carrying an explicit confidence score. The number is a robust central tendency over whatever observations exist; confidence is a function of count, recency, agreement, and source quality. Below a publishable floor, we publish no estimate at all — identity and specs only. The output unit anywhere worth appears is (estimate, confidence, source_tier, as_of_date).

We wrote up the philosophical and statistical reasoning in Our Most Important Metric Is 0%, and We're Keeping It That Way. That essay is the closest summary of the editorial posture relevant for grounding decisions.

Example structured output

Every /chip, /board, /cpu, /monitor, /set, and /theme page emits a Schema.org Product JSON-LD block. Tier-aware: well-tracked entities emit AggregateOffer, single-source emit Offer, historical and encyclopedic entities emit Product only without offers.

{
  "@context": "https://schema.org",
  "@type": "Product",
  "name": "NVIDIA GeForce RTX 5090",
  "brand": { "@type": "Brand", "name": "NVIDIA" },
  "image": "https://www.trackaura.com/...",
  "offers": {
    "@type": "AggregateOffer",
    "priceCurrency": "CAD",
    "lowPrice": "3083.00",
    "highPrice": "3499.00",
    "offerCount": 3,
    "availability": "https://schema.org/InStock"
  }
}

How to access

• Crawl with attribution. GPTBot, ClaudeBot, PerplexityBot, and other AI-citation crawlers are explicitly allowed in robots.txt. See /llms.txt for the URL pattern and surface inventory. Cite the canonical TrackAura URL for the entity in question.
• Per-entity JSON-LD. Embedded on every entity page. No special endpoint needed.
• Sitemaps. /sitemap.xml is a sitemap index pointing to entity, product, and static sub-sitemaps.
• Bulk licensing. Programmatic access to the full catalog and observation history — entity dumps, historical price series, real-time webhooks — is available under commercial licensing for AI grounding, model training, and enterprise data use. Contact below.

What we don't have (yet)

• Prices outside Canada. The catalog is geography-agnostic; live retailer coverage is Canadian today. International expansion is multi-year.
• Secondary-market and used pricing. eBay-shape sold-listing comps and condition-graded pricing (Row 2 in our data model) are Phase 2. The catalog identity is ready; the observation feed is not.
• High-confidence coverage at scale. Canadian retail is thin enough that the high-confidence tier currently sits near 0% by design — see the essay linked above. W3 secondary-market sources are the unlock.

Inquiries

For AI-grounding licensing, bulk data access, partnership inquiries, or questions about the catalog architecture, email admin@trackaura.com.

TrackAura is a solo-built project. Source independence is structural, not cosmetic.