Methodology
How the platform is built, what it measures, what it ignores, and where it breaks down. Every design decision is documented here because opacity in methodology produces opacity in interpretation.
No black boxes. Six layers, transparent weights, every assumption documented.
How the Intelligence Is Built
Six layers. One score.
The transparency argument
Most fashion intelligence products treat their methodology as proprietary. This one does the opposite, and the reasoning is structural, not altruistic. If you cannot see what drives a ranking, you cannot know when to trust it and when to discount it. Every signal score on this platform shows its component breakdown because that is the only way the ranking is useful rather than merely authoritative.
Signal taxonomy
The platform tracks 139 canonical signals across silhouettes, garments, accessories, materials, colors, and styling constructions. These are not trend labels invented for marketing — they are specific enough to be observed on a runway, searched for on Google, and listed on a resale marketplace. Each signal has a canonical label, a set of commercial translation terms (used for search and resale matching), and a category classification.
The taxonomy was designed to be exhaustive for luxury womenswear while remaining precise enough for computational matching. Signals like “Relaxed Leather” or “Bar Jacket” map directly to observable runway moments, searchable consumer terms, and resale listing categories.
Six scoring components
Each signal is scored across six independent components. The components are designed to capture both creative authority (runway, editorial, cultural) and commercial validation (search, retail, resale). The separation is deliberate — it prevents the platform from collapsing into a marketplace popularity index.
Runway authority (27%)
Derived from runway archive records. A signal scores high when multiple houses show it across multiple collections, weighted by the house's editorial intensity and confidence. Currently sourced from 195+ runway records covering Paris, Milan, London, New York, Copenhagen, Seoul, Tokyo, and Rome fashion weeks.
Editorial proxy (22%)
Derived from the editorial intensity field in runway archive records. This captures how much critical attention a collection received — a proxy for how strongly the fashion press endorsed the creative direction. Values are normalized from a mix of numeric scores and qualitative labels (high, medium, low).
Cultural diffusion (14%)
Measures cross-city and cross-house spread. A signal that appears in only one house scores low; a signal observed across Paris, Milan, and London from multiple houses scores high. This component rewards breadth of creative consensus.
Search intent (14%)
Real consumer demand data from Google Trends via the pytrends API. Each signal's canonical label and commercial translation terms are queried for recent search volume and acceleration. Currently 139 of 139 signals have live search data. The score captures both absolute volume and recent momentum.
Retail adoption (14%)
Covers approximately 50 signals across 8 retailers. Weight is partially redistributed to other components for signals without retail data. Expanding as additional retailer integrations go live.
Resale durability (9%)
Sourced from Vestiaire Collective via automated Playwright scraping. Captures listing count and median price for each signal's commercial translation terms. High listing counts with strong pricing indicate sustained consumer demand beyond the initial retail cycle. Currently 164 signals have live resale data.
The math
Signal heat scores are not simple averages. The scoring engine uses a statistically rigorous pipeline designed to handle missing data, outliers, and uneven distributions.
Step 1 — Robust z-scores
zrobust = (x − median) / (MAD × 1.4826)
Where MAD is the median absolute deviation. The 1.4826 constant makes the denominator a consistent estimator of the standard deviation for normal distributions. This is more resistant to outliers than mean/standard deviation.
Step 2 — CDF mapping
score = 50 × (1 + erf(zrobust / √2))
The robust z-score is mapped to a 0–100 scale using the error function (erf), which produces the cumulative distribution function of the standard normal. This gives a percentile-like score that is smooth and bounded.
Step 3 — Proportional weight redistribution
When a signal is missing data for one or more components (e.g., no runway records, no resale data), the missing component's weight is redistributed proportionally across the available components. This means a signal with only search and resale data is scored only on those two dimensions, not penalized for missing runway coverage. The available component count is tracked and surfaced as the confidence indicator.
Step 4 — Empirical Bayes house shrinkage
House-level scores use empirical Bayes shrinkage to prevent small-sample inflation. A house with only one signal match is pulled toward the global mean; a house with many matches retains its raw score. The shrinkage factor is: shrunk = global + (raw − global) × n / (n + k), where k is a smoothing constant.
Derived metrics
Several display metrics are derived from the six component scores rather than scored independently.
Directional strength averages the non-zero creative components (runway, editorial, cultural diffusion). Market intent averages the non-zero commercial components (search, retail, resale). Movement is derived from search acceleration and volume momentum. Durability is the resale durability score. Category (halo-led, accelerating, commercial breakout, selective, emerging) is derived from the gap between directional strength and market intent, with special handling for signals that lack runway data.
Data sources
Weight profile
Runway authority and editorial proxy carry the largest combined weight (49%) so the model remains couture-to-commerce, not marketplace-first. This is a design choice, not a limitation. Marketplace-only indexes over-rank volume spikes. Signal heat requires creative authority before market validation can elevate rank.
Component weights
Runway 27% · Editorial 22% · Cultural 14% · Search 14% · Retail 14% · Resale 9%
When components are unavailable, their weights are redistributed proportionally. A signal with only search and resale data is effectively scored 61% search / 39% resale.
Known limitations
Runway coverage is expanding. With 195+ records across the archive covering FW2026, SS2026, and Spring 2026 Couture, many signals now have direct runway attribution. The phrase-matching fallback has been disabled to prevent false positives, which means signals without explicit runway matches score zero on runway authority. Coverage continues to improve as the archive expands.
Retail adoption is partial. The retail component (14% weight) covers approximately 50 signals across 8 retailers. For signals without retail data, the weight is redistributed proportionally across available components. Coverage is expanding as additional retailer integrations go live.
Resale data reflects Vestiaire Collective only. The RealReal integration was blocked by bot protection. Vestiaire skews European and may under-represent US market dynamics.
Search data is point-in-time. Google Trends scores are relative, not absolute. A signal's search score reflects its position within the current query batch, not historical demand levels. Refresh cadence matters.
Editorial proxy is a rough instrument. The editorialIntensity field in runway records captures broad editorial attention, not specific signal-level endorsement. A house with high editorial intensity boosts all its attributed signals equally.
A Note on Scope
Rankings are directional intelligence, not commercial endorsements. The dataset is continuously evolving as additional source packs are validated, corrected, and expanded. Signal scores reflect structured hypotheses about creative authority and commercial momentum — informed by real evidence, refined with every data merge. Treat them as decision-support tools calibrated by methodology, not as buy recommendations.
Platform Glossary
Terms used across this platform
Definitions for readers outside fashion merchandising, analytics, or luxury strategy.
Signal Classification
Accelerating
A signal gaining momentum across multiple scoring layers at the same time. Both creative authority and market interest are increasing, suggesting the signal is moving from directional interest toward genuine commercial viability.
Commercial Breakout
A signal where consumer demand (search volume, resale activity) is running ahead of runway endorsement. The market wants it before the creative system has formally endorsed it — which means it may lack the longevity that runway-backed signals carry.
Emerging
A signal with early-stage evidence — it's been spotted but there isn't enough data across enough layers to score it with high confidence. Worth tracking, not yet worth buying into.
Halo-Led
A signal with strong creative authority (runway evidence, editorial endorsement) but weak commercial translation (low search interest, little retail pickup). Halo signals drive brand perception and press coverage rather than direct sales — they're the runway pieces that make headlines, not the ones that fill shopping bags.
Strategic Language
Commercial Conviction
The combined strength of market-facing evidence: consumer search interest, retail adoption, and resale durability. High commercial conviction means consumers are actively searching for it, retailers are stocking it, and it holds value on the secondary market.
Conviction Tier
Signals scoring 75 or above on the composite signal heat scale. These carry the strongest combined evidence across creative and commercial layers — they're the signals a buying team can allocate depth behind with confidence.
Creative Authority
The degree to which the fashion system (designers, editors, critics) has endorsed a signal through runway presence, editorial coverage, and cross-house recurrence. A signal with high creative authority appeared in multiple collections across multiple cities and was discussed in independent reviews — not just shown once.
Diffusion
How a signal travels from its origin (typically haute couture or luxury runway) through progressively broader market tiers (premium designer, contemporary, accessible luxury, high street). A signal that “diffuses” successfully maintains creative authority as it moves down-market. One that doesn't either stays niche or loses its identity.
Noise Floor
Signals scoring below 45 on the composite signal heat scale. They're tracked for completeness but lack sufficient evidence across enough layers to inform any commercial decision. Most of the 139 tracked signals fall here — that's normal and expected.
Signal Heat
The platform's composite score combining creative authority and commercial conviction across six weighted layers. Heat is not popularity — a signal can have high heat with zero search interest if its runway authority and editorial intensity are strong enough. Creative authority is weighted more heavily than market data by design.
Translation
The commercial conversion of a runway signal into buyable product categories. Translation answers: which product categories will this signal enter first, at what price tier, and with what buy depth? A trench signal might translate into outerwear (obvious) but also into leather goods and accessories (less obvious, higher margin).
Scoring & Methodology
Cultural Diffusion
The third scoring layer (14%). Measures cross-house and cross-city spread. A signal confined to one design house or one city scores low regardless of how strong the individual showing was — breadth of adoption across the creative system matters.
Editorial Intensity
The second scoring layer (22%). Measures how frequently and prominently a signal was discussed in independent fashion media — reviews, commentary, analysis. A signal that multiple critics independently highlighted scores higher than one that was merely visible on the runway.
Empirical Bayes Scoring
The statistical method used to compute house-level scores. It prevents houses with limited data from being ranked unfairly high or low by pulling their scores toward the overall average. A house with only 2 signals in the archive gets pulled toward the mean; a house with 15 signals is trusted more at face value.
Evidence Depth
The number of scoring layers (out of 6) that have active data for a given signal. A signal with 5/6 active layers has deep evidence — its signal heat score is well-supported. A signal with 2/6 active layers is scoring on thin evidence, and its rank should be interpreted with caution.
Resale Durability
The sixth scoring layer (9%). Measures whether a signal holds value on secondary markets. High resale durability suggests staying power — consumers aren't just buying it, they're keeping it or it's maintaining price. Low durability suggests a flash trend.
Retail Adoption
The fifth scoring layer (14%). Measures whether major retailers are stocking products aligned with the signal. Currently covers approximately 50 signals across 8 retailers. Signals without retail data have this weight redistributed proportionally.
Runway Authority
The first and most heavily weighted scoring layer (27%). Measures how many houses showed a signal, across how many cities, with what evidence depth. A signal shown by 12 houses across 4 cities scores near 100. A signal shown by 1 house in 1 city scores low.
Search Intent
The fourth scoring layer (14%). Measures consumer search interest via Google Trends data. A signal that consumers are actively searching for carries commercial proof that purely editorial signals lack.
Weight Redistribution
When a signal is missing data for one or more scoring layers (for example, no retail data), the weight that layer would have carried is redistributed proportionally across the layers that do have data. This prevents signals from being penalized for data gaps while maintaining the relative importance of each layer.