Methodology & Data Ethics

What we measure

Xerothread aggregates unit sell-through and revenue from independent impulse and specialty retailers (toy, novelty, collectibles, viral candy, and adjacent counter categories). Metrics are deterministic statistics: velocity, week-over-week change, OLS slopes on short windows, cohort percentiles, and street-price medians. We describe these as momentum and early signals. We do not claim predictive accuracy or forward-looking certainty.

Panel composition

Stores connect via POS OAuth (Square). Each connect may backfill historical tickets. Until a snapshot is published from real connected stores, public surfaces render labeled illustrative sample data only. We never invent a store count: the published payload carries the exact connected-store count, never rounded. Pre-launch: we are recruiting independent retailers.

Suppression (k-anonymity + dominance)

A buyer-visible aggregate cell is released only when both hold:

• k ≥ 5 — at least five independent stores contribute to the cell.
• Dominance ≤ 40% — no single store accounts for more than 40% of units or revenue in the cell.

Failed metro cells roll up to state, then panel; if they still fail, they are suppressed and omitted. Surfaces read only pre-flagged rows — they never re-derive suppression.

No consumer data

Ingest whitelists timestamps, money, quantities, item identifiers, and transaction refs. Customer objects, loyalty IDs, emails, phones, card details, and free-text notes are dropped before persistence. Nothing resembling shopper identity is stored or logged.

Store identity

Buyers never receive store names, addresses, or other store-identifying fields. A participating retailer always sees their own numbers inside their dashboard; panel views remain aggregated.

Provenance & permitted use

Every store carries a data license; every ingest batch carries a source. Aggregate resale of anonymized statistics is the product. Trained ML on POS-platform-derived rows is restricted by license flags (v1 ships no trained models). Classification uses LLM inference with provenance retained.

Merchant grant

No store enters the panel without a recorded data-rights grant covering use, aggregation, anonymization, and resale in aggregated form. Already-published aggregates may survive departure; future inclusion stops on termination. See the Merchant Data Agreement.

Trend Index construction

Weekly snapshot. Indexed series are base-100 unit paths over eight weeks for top risers and fallers that clear suppression. Direction labels (surging / rising / declining / dying) come from momentum scores on the same deterministic pipeline. Sample snapshots generated from synthetic data are hard-blocked from publishing as live.

All figures are aggregated and anonymized across a minimum of 5 independent retailers. No store-identifiable data is ever shared.