WHAT IT IS
A catalog of specs is a list. A knowledge graph is a set of relationships.
Most product sites store a spec sheet per model and stop there. We store two things and the link between them: a product-model record — one durable entry per model, its identity, protocols, subscription reality, and lifecycle — and a compatibility graph that connects each model to the ecosystems and hardware it actually works with.
That link is the asset. It's what lets the checker answer "will these work together in my home" instead of reciting what each box claims on its own. Every product is a node; every "works with Alexa," every "needs a Thread border router," every "cloud link removed" is an edge — and every edge carries a source.
It's built to be read by people and by machines — the same graph powers the human checker and a public feed for AI agents.
Nodes — the models
292 product records across 17 categories, each with a stable identity that survives renames, generations, and supersessions.
Edges — the compatibility
1460 integration cells linking models to 5 ecosystems, plus protocol and hub-dependency edges — the "picks that work together" relationships.
Provenance — the trust
5165 source entries. A claim without a citation isn't in the graph; an unknown we haven't confirmed says so plainly.
THE GRAPH TODAY · LIVE FIGURES
The graph, by the numbers.
Not a marketing snapshot — the actual shape of the catalog right now, including the parts we haven't finished confirming. These numbers are generated from the live data on every build.
Why we show the grey. "Unknown" and "confirmed none" are separate states here on purpose. A confident-looking guess is worse than an honest gap — so an edge we haven't verified is never dressed up as a "no."
Each record has 25 attributes it can hold — protocols, hub requirement, subscription reality, whether it works offline, the security-update window, and so on. Not every question applies to every product: a plug-in device has no batteries to maintain and a light bulb provides no hub capability, so those cells are not counted against us. Across 292 records that leaves 6,560 facts the graph is built to carry. Here is how many it actually does, and how many of those point at a source you could open.
Where the biggest gaps are right now
- Security-update window26.4% established
- identifiers28.4% established
- Vendor account needed36.3% established
- App platforms45.9% established
- Works without the internet46.6% established
Read the gaps honestly. Some attributes are empty because the fact is rare — most products aren't hubs, and most vendors publish no security-update end date — and some because we haven't researched them yet. Either way the cell says "not established" rather than guessing, and the list above is regenerated from the live data on every build.
PROVENANCE
Why every claim in the graph holds up.
Open any record and each fact traces to where it came from. Here's a real one from the catalog, annotated.
A source, or it doesn't ship
Every edge cites where it came from — a vendor spec, an ecosystem developer doc, a standards registry — not "our research."
Honest unknowns
We distinguish "we checked, it's a no" from "we haven't checked." The graph carries both, labeled.
Negatives are claims too
"Needs the Hue Bridge" or "no HomeKit support" is evidence we record and cite, not silence.
Made for agents
90.3% of our claim-bearing citations now carry a fetchable source URL — so an AI agent, and our own scheduled watcher, can re-check our work. The rest are being converted from prose, and the number above is live.
WHAT'S LIVE · WHAT'S COMING
The catalog's promise is shifting from complete to true.
We're honest about what the machinery does today versus what it's built to do next.
Working today
- ✓Source-tracked records — 292 models, every claim cited
- ✓The compatibility graph — 1460 model-to-ecosystem edges powering the checker
- ✓Honest unknowns & evidenced negatives — gaps and "no"s are labeled, not hidden
- ✓Machine-readable feed — the same graph, published for AI agents
- ✓Source watching — 1,198 cited URLs re-fetched on a schedule and fingerprinted; a change is flagged for a person to re-verify, never written into a record automatically
- ✓Graph health in the open — the coverage figures above are recomputed from the live data on every build
Being built next
- →Filling the gaps named above — offline operation, account requirements, and Wi-Fi bands are the next research passes
- →Per-record change history — what moved, when, and which source said so
- →"Compatibility changed" alerts — the day a firmware update breaks a pairing, you'd hear it
Straight answer: the graph is kept current by scheduled research and by a watcher that notices when a cited page changes — but every correction is made by a person, the catalog does not maintain itself, and a "verified" date is a record of when we last looked, not a guarantee nothing has changed since. Change detection is our stated direction. The day it goes live, this page will say so.
Put a question to the graph.
Pick what you own, or what you're about to buy, and see whether it all works together — sourced, not guessed.
Want the methodology — what we borrowed from the formal provenance standards, and what we had to add for facts that change? Read how we're building the knowledge graph →