How we compute this

ARKOV diagnoses the lifecycle phase of fashion trends. It does not just chart popularity. The core distinction: hype ≠ adoption ≠ normalization. A trend can lose search interest because it is dying, or because it became so normal nobody searches for it. Telling those apart is the product.

The 7 stages

Discovery. Waking from a low base, and growing much faster than the market. Early, and easy to over-read.

Rising. Attention is climbing meaningfully faster than the fashion market as a whole.

Peak. Attention near its top after a fast run-up, with the climb stalling.

Cooling. Attention is losing ground against the fashion market: its share is receding year over year, which can happen even while raw searches hold steady.

Normalized. The excitement is gone but the baseline holds: a staple, not a fad. The Cooling/Normalized line is drawn by the same year-over-year threshold; the staple tests (baseline, persistence, stability) are what separate Normalized from Rising and Dormant.

Dormant. Far below its former peak with no sign of waking; residual niche presence.

Revival. A documented earlier cycle, a quiet spell, and now a fresh climb. We never call revival without prior-cycle evidence.

Sources and their roles

Google Trends (search attention, weekly, 5 years, US + state-level geography) and Wikipedia pageviews (a second, independent attention signal) power v1. Growth is always measured relative to the median of all 100 tracked trends, so a platform-wide shift in search behavior cannot masquerade as a fashion trend.

The Momentum Index breakdown names its axes after what they actually measure. The wikipediaaxis is Wikipedia pageviews against that article's own peak. It is not a social-platform signal, because we have none. Trends without a Wikipedia article show four axes instead of five, and the index is renormalized over what is present rather than filling the gap with a guess.

About trend imagery:photos illustrate what a trend looks like. They are not a data source. Where no suitably licensed photograph exists, we use an AI-generated illustration instead; these are always labeled "AI-generated illustration" in the image credit. Every real photograph is credited to its photographer and source.

The pipeline

Raw observations → seasonal adjustment → canonical features (level vs own peak, year-over-year growth, persistence, stability, geographic breadth) → calibrated phase rules → mechanism scoring → this page's diagnosis. The narrative you read is generated fromthe structured diagnosis, so it can never invent a conclusion the numbers don't hold.

What we don't know (yet)

We have no retail sell-through, no resale-sold prices, no street-style imagery, and no age demographics yet. Every diagnosis lists these as explicit evidence gaps rather than guessing. Probabilities and forecasts are deliberately absent until our calibration standard says they can be trusted. Evidence strength is shown in words, not percentages.

A note on phase history

The 5-year phase strips are retrospective diagnoses: today's methodology applied to backfilled data, not what we would have said at the time. When the methodology changes, history is recomputed and versioned; it never changes silently.

Methodology v1 · monthly cadence · US market · Data source: Google Trends