Method
Show the work.
Every number here comes from free, public data. This page says exactly what is measured, how it is turned into a reading, and — the part most dashboards skip — what it cannot tell you.
Three rules
Hemindex measures signals, not sales. It has no access to revenue, units or transactions, and it does not estimate them. What it measures is attention: how many people searched for something or looked it up. A brand in a scandal gets enormous attention; a quiet, profitable label gets very little. Reading one as the other is the fastest way to be wrong in public, so every figure on this site is labelled with what it actually counts.
Every basket is public. Each index lists every constituent, its weight, and its contribution this month, one click from the number.
Indices are constructed, not measured. There is no “boldness of fashion” in the world waiting to be observed; there is a set of editorial choices about which trends count as bold, applied consistently over time. That is how the CPI and every consumer-confidence index work too. What separates a credible index from astrology is that the basket, the weights and the formula are public and never change silently. So the site says “the Boldness Index reads 61”, never “fashion is 61% bold”.
Where the numbers come from
Google Trends
Search interest, the workhorse: 78,897 weekly observations across 289 tickers, 2021-08-22 to 2026-11-08. It arrives as a relative 0–100 scale rather than a count, set per request — so two terms pulled separately are not comparable, and a big term in the same request crushes a small one to zero. Every ticker is therefore permanently assigned to a small group of similar size with a fixed, dull reference term (words like “dentist” and “password manager”), and that assignment never changes: moving it would silently rescale the ticker’s whole history.
The five-year window slides, so the whole window is pulled again every week and replaced. Every stored week always comes from one request. The visible consequence is that last week’s figure can move by a point when this week’s pull lands — that is Google’s sampling, and it is why the terminal shows one decimal.
Wikipedia pageviews
Absolute counts of how many people opened an encyclopedia article, automated traffic excluded, back to 2015: 26,270 weekly observations across 54 brands, 2015-06-28 to 2026-11-08. Used for brands, where it is the more trustworthy of the two sources.
When an article is renamed, its history splits between the old title and the new one, and asking for the current title alone returns a series that jumps on the rename date and looks exactly like a surge. Every article’s former titles are resolved and summed, and the resolved list is fixed once recorded.
Not used, and worth saying: TikTok and Instagram (no public data), retail and resale sites (would require scraping), and point-of-sale data (does not exist for free — which is the root of the never-say-sales rule).
From a ticker to an index
A ticker is one tracked trend, with a permanent four-letter code. Each week it gets a reading.
The calendar is removed first. Almost every ticker is seasonal — puffer jackets rise every autumn, and so does Hermès, a December gift — so “unusual” has to mean unusual for the time of year. Each week is compared with the ticker’s own norm for that week of the year: the average of the same week in every other year on record. What is left over is the residual.
The residual is turned into a z-score: how unusual is this week against this ticker’s own trailing two years? That is what stops loud tickers from drowning out quiet ones, and it is clipped at three standard deviations so one viral week cannot dominate. A ticker needs a year of history before it counts toward anything.
A basket is a group of tickers assigned to one pole of an index — bold against quiet, small indulgence against big ticket, high effort against low. These are editorial choices. Somebody decided which trends count as bold. The constituents and their weights are published so the choice can be argued with, and a change to a basket is a versioned commit with a stated reason (currently basket v2).
An index is the ratio between the two poles, scaled 0–100 with 50 as neutral: each pole is the weighted mean of its constituents’ shifted z-scores, and the index is 100 × A ÷ (A + B). Poles are averaged rather than summed, so the pole with more tickers cannot win on headcount. Indices are monthly because a composite of twenty-odd tickers cannot meaningfully move in a week; a single ticker can.
- Boldness Index — Confidence versus anxiety. Does the moment want to be seen, or to disappear?
- Lipstick Index 2.0 — Trading down versus trading up. The classic recession signal, computed live instead of asserted.
- Effort Index — Social re-engagement versus retreat. How much labour are people putting into appearance?
On the Lipstick Index: the original claim — that consumers substitute small indulgences for large ones under economic pressure — is contested in the economics literature. Hemindex reports the ratio. It does not assert that the ratio predicts recessions.
Coverage
Coverage is the share of a basket’s tickers that reported this month. Above 0.80 a reading publishes normally. Between 0.60 and 0.80 it publishes with a visible warning. Below 0.60 it does not publish at all, and the gap is shown instead — a reading from half a basket is not a weaker version of the real reading, it is a different number wearing the real number’s name.
A source that returns nothing is stored as unknown, never as zero. Zero means “nobody searched this”; unknown means “we did not find out”. Conflating them would drag readings down while coverage still read 100%.
What these indices cannot tell you
- They measure attention, not behaviour. Searching for something is not buying it, wearing it, or wanting it.
- They are relative, not absolute. Every reading is a comparison between two baskets, both of which move.
- Basket membership is a judgement call. Reasonable people would sort some tickers differently.
- They cannot establish causation. A rising Boldness reading alongside a falling stock market is a coincidence until proven otherwise, and proving it is outside this product’s scope.
- Search interest reflects who searches. The data skews toward the online, the young and the trend-aware. This is not the general population.
- Readings are seasonally adjusted, and adjustment is an estimate. The seasonal norm is an average of about five prior years, so it is approximate, it shifts a little as each new year is added, and one freak week in a past year nudges it. Adjusted, not immune.
Why France appears, and why there is no French index
France is a comparison lens for an American audience, not a second market. Brand names survive the crossing — 29 of them carry enough French data to compare, drawn as a dashed line on the ticker page — but English aesthetic labels do not, so no French index is published as a headline number.
Versions
Every stored reading records the basket version and the method version that produced it (currently basket v2, method v2-seasonal), so a number is never separated from how it was made. The full internal methodology, including every measured correction to earlier reasoning, is in the project docs.