# Sillage Index · Method and publication scope

**Published research summary · November 5, 2026.** This note describes the analyses and their evidence. Barbell reached a measurement checkpoint after all registered search captures were validated.

Sillage Index is Ryder’s quarterly fragrance-and-economy project: four distinct gauges and an essay series in the order **Lipstick Test → Bottle-Hours → Comfort Index → Barbell**. The project page presents their distinct evidence and observation dates. The fragrance pyramid groups readings by their intended pace of change; it is an organizing device, not a combined economic score or a validated forecasting model.

## Four questions, different evidence

| Gauge | Evidence | Permitted interpretation | Current status |
|---|---|---|---|
| Lipstick Test | National-accounts quantity indexes around nine US business-cycle peaks | Historical comparative resilience of a broad cosmetics category | Complete; registered primary hypothesis not established |
| Barbell | Registered tier search candidates, household wealth, and income-tercile sentiment | Planned associations in search interest and economic divisions | Collection complete; required mass basket unavailable, so tests were not run |
| Comfort Index | Annual fragrance catalog records and consumer sentiment | Exploratory associations in recorded launch composition | Revised exploratory study complete; original restricted test unavailable |
| Bottle-Hours | Evidence-reviewed archive list prices divided by snapshot-month average wages | Affordability of two fixed historical bottle benchmarks | Required basket complete; unequal, dated observation windows |

Searches, catalog launches, national-accounts consumption, and archive prices measure different things. Their agreement would not by itself establish causation, and they must not be described interchangeably as fragrance sales. A higher nominal price or revenue also does not establish greater consumption.

## Analysis and uncertainty

Machine-readable configurations and local preregistration notes record outcomes, transformations, test directions, lag choices, comparison families, and deviations. The plans were committed before the relevant outcome analyses. These are local, versioned analysis records; no external registration is claimed. Comfort Index’s revised scope and Bottle-Hours’ delegated Codex review remain explicit methodological deviations.

The research works in growth, differences, or deviations from trend rather than correlating trending levels. Registered regressions use HAC/Newey–West uncertainty. Directional tests retain their specified tail; an estimate in the opposite direction does not justify changing to a favorable test. Each gauge’s specified test family reports raw and Benjamini–Hochberg adjusted p-values at a nominal 5% threshold. Adjustments do not repair measurement limitations or guarantee validity of the underlying model and placebo assumptions. Null findings and separate robustness variants remain visible.

Effects and intervals take priority over a binary threshold. Lipstick Test’s nine-episode bootstrap intervals are rough. Catalog classification and sparse archive observations add uncertainty that model intervals cannot fully capture. Descriptive readings are distinct from hypothesis tests. The “rising,” “steady,” or “falling” label uses the registered one-year change relative to the historical standard deviation of such changes; an absolute standardized change below 0.5 is steady. A short-run direction label can differ from a fitted long-run trend.

## Dates, provenance, and replication

Every displayed reading needs its own observation date. A pull date records acquisition, and an update date records a source release; neither makes an older observation current. Bottle-Hours additionally distinguishes scheduled sampling labels from actual archive capture dates and matches macro values to the latter. The October 14 collection does not turn Dior’s August 2021 primary observation into a 2026 price.

Barbell has **12/12 verified exports as of October 19**: all four batches on October 14, 18 and 19 in America/Chicago, each with 273 registered months. The operational schedule was committed before the new queries. Capture receipts preserve exact source bytes and disclose the acquisition-time basis; October 18–19 receipts use observed filesystem modification times, with HTTP status unobserved and null. Original captures, the January 2004–September 2026 observation range, and all scientific parameters are preserved. No collection or unlock request remains pending.

The actual offline run stopped at the frozen resolution gate. All 15 candidate terms trigger review, but none has an active-period stitched median in [5, 30], so the registered rule skips Round 2. All five mass terms have fewer than 36 active months, leaving no retained mass basket. H2c and the other three primary tests, BH adjustment, holdout and robustness were not run; their effects, intervals and p-values remain unavailable. This is a measurement limitation, not a null result. A separate dated descriptive diagnostic reproduced byte for byte, and the pipeline reproduced the same checkpoint from cached raw files. These checks do not establish successful end-to-end analysis or results-JSON determinism. Any redesigned measurement plan must be registered separately.

Immutable source files have sanitized request metadata, UTC timestamps, status, and checksums. Gaps are retained rather than fabricated or interpolated. Saved verification reports document actual cached end-to-end replays for Gauges 1, 3, and 4, including deterministic output comparisons with timestamps excluded where specified. That verifies reproducibility from retained inputs, not independent validation of every source or access to the same future online vintage. This publication-text task does not rerun those pipelines.

## Sources, redistribution, and attribution

Public pages may use aggregate findings, approved figures, sanitized provenance, and publication-safe code. Community catalog downloads, archived retailer HTML, row-level price/review tables, private audit evidence, browser information, and credentials stay outside the public package. Raw BEA responses also remain excluded because their bodies can echo an API key. Gauge 1’s existing allowlisted replication package uses sanitized processed observations.

Source attribution does not assign a new license to third-party data. The materials retain their applicable source terms. Link external data through their source pages and preserve their own terms rather than bundling private raw caches.

Sources include BEA; FRED/BLS; NBER; Fragrantica via the named public Kaggle dataset; Parfumo via TidyTuesday; and Internet Archive captures of the registered retailers. Barbell's collected sources include Google Trends, Federal Reserve wealth series, and University of Michigan sentiment; source collection does not establish an analytic result. Preserve the existing notice: **“This product uses the FRED® API but is not endorsed or certified by the Federal Reserve Bank of St. Louis.”** Link the [FRED API Terms of Use](https://fred.stlouisfed.org/docs/api/terms_of_use.html).

## November research update

The [November research update](/work/sillage-index#research-update) connects evidence from Carnegie Mellon University Libraries’ business research guide with each study and records follow-up data needs.
