AI is the
X-axis.

“AI is the X-axis. Other channels are reference frames. We don't predict — we observe what AI is doing and which channel shaped it. Anyone can compute embedding distances; only we have the cohort substrate to attribute AI's vocabulary to a specific reference frame across 109+ categories.”

When AI doesn't anchor to the OEM, it usually anchors to buyer search.

57% of non-OEM-anchored slugs in 2026-W19 are SEARCH-anchored — 39 of 69 (39 SEARCH + 24 DIST + 6 AGG closest, of 174 cohort-stable categories). The remaining 30 are anchored to distributor or aggregator language.

Replayable arithmetic: 39 / (39 + 24 + 6) = 39/69 = 56.5%. Reproducible from our weekly per-frame attribution counts for 2026-W19. Alternative framing for a larger denominator: 22% of all cohort-stable categories (39 of 174) — “roughly 1 in 4 categories AI hears the buyer first.”

This is the substrate's first cohort-stable week. Prior weeks (W17/W18) had cohort composition still shifting; the methodology lock-in date is 2026-W19. We'll surface the W19 → W20 trend on the next refresh.

The weekly record.

One category, five vocabularies.

For each of 253 categories in the CategoryRank Ontology, we collect the vocabulary five distinct vantages use to describe it:

• AItop-alias forms emitted by 10+ LLMs (US, EU, CN-domiciled) when prompted for the category. This is what AI remembers — the object of measurement.
• OEMthe brand's own product-page taxonomy, Wayback-validated weekly. Reference frame: how the maker speaks.
• DISTparametric tree forms from authorized distributors — Mouser, DigiKey, Arrow, Avnet, Newark, LCSC. Reference frame: how the channel speaks.
• AGGcategory labels from aggregator / search-surface layers — Octopart, Findchips, Oemsecrets. Reference frame: how the discovery layer speaks.
• SEARCHtop-10 buyer queries per slug from a leading search-query data provider. Reference frame: how buyers speak.

AI is the subject of the sentence. The other four are reference frames we compare AI against — not products we sell.

Continuously refreshed. OEM-anchored.

The CategoryRank Ontology is the taxonomy underneath every verdict on this site. We optimize it to maximize semantic closeness to OEM canonical names — the anchor — while amplifying coverage through distributor parametric trees and aggregator discovery-layer labels.

New categories emerge in the wild. We observe and integrate them. Vocabulary drift gets tracked week by week, never smoothed into a single snapshot. The result: a living standard that stays accurate to how the supply chain actually talks, instead of locking the map at the moment of publication.

Every category number on this site cites a specific ontology version. As the universe shifts, the version increments and the delta is replayable.

Anchoring × manufacturing risk.

Crossed with the manufacturing-risk composite (a separate scorecard covering geo concentration, brand concentration, substitute scarcity, naming divergence, and inventory thinness), each category lands in one of four quadrants:

Anchoring and fragmentation thresholds are heuristic and currently uncalibrated; they will be re-tuned with a held-out validation set after the demo cycle.

Same measurement, four operating brains.

The same quadrant verdict drives different action verbs depending on whose book is reading the data:

What this measurement is and isn't.

The substrate is honest about its own limits. When it updates, this list updates.

• Centroid is an average across all AI/OEM/DIST/AGG forms in our W19 cohort observation; smooths over within-frame variance. Cohort disagreement score reported separately to surface AI internal fragmentation.
• SEARCH centroid uses top-10 buyer keywords per slug from a leading buyer-query data provider; short-form text systematically embeds at lower cosine to long-form OEM text, hence quantile-normalized distances reported alongside raw cosines.
• Anchoring thresholds are heuristic and uncalibrated — should be re-tuned with held-out validation set post-demo.
• Slugs with missing vantages (e.g. no DIST coverage in cohort) are flagged as data-incomplete and the missing frames listed; they cannot be cleanly classified.
• Quadrant assignment uses scorecard manufacturing-risk band as orthogonal axis; scorecard composite weights are heuristic (per our manufacturing-risk scorecard).
• Brand-level drill-down (per-brand AI-vs-frame attribution) deferred to v2 of this artifact — would power /lens/oem customer-specific 'you are anchored to DIST' callouts.
• This is a DESCRIPTIVE per-slug attribution. Not causal. Useful for identifying ARBITRAGE candidates (high-risk + fragmented), not for forecasting.
• Buyer-search coverage is partial: 112 of 253 slugs (44%) have a buyer-keyword pull. The remaining slugs have no Buyer-search field — this is absent-not-zero.

Every number cites its source.

Every number on every lens page is reproducible from our weekly substrate — SHA-pinned per artifact, with a replay manifest committed to source control. Derived rollups regenerate weekly on the same lock; nothing on a customer-facing page is hand-edited.

We render Expanding coverage categories honestly rather than hiding them. Cohort substrate composition is disclosed when it changes between weeks. AI's internal disagreement (the variance across model cohort) is surfaced as a separate signal on every drill page, never smoothed into the centroid.