Product
Keep the meaning
attached to the metric.
An observation, a buyer indicator, a portfolio share and a cost assumption answer different questions. These distinctions connect the product screens to decisions their evidence can support.
Trace listing facts to their actual source and capture date.
Distinguish price baselines, cohorts and indicator-based totals.
Leave unknown values unknown rather than creating false precision.
Buyer badges and indicator-based values are research signals, not verified unit sales or revenue.
Identify the source of an identifier.
A marketplace listing’s original typed UPC/EAN/GTIN, its normalized equivalent and a barcode used as a lookup input are different records of evidence. Normalization must not hide a known conflict.
Earlier accepted identifiers can remain useful after a price-only observation, but keep their original source and time. A later price does not refresh every field or independently confirm product equivalence.
Read buyer indicators and counters in their own scope.
Amazon purchased badges are approximate rolling product-level buyer indicators across an ASIN’s sellers. They are not one seller’s realized unit sales. Price multiplied by that indicator produces indicator-based value, not audited revenue.
eBay cumulative counters can span a listing’s lifetime and variations. Observed increases are not confirmed transaction records or monthly sales. Pooled or overlapping signals must not be summed into unique demand.
Keep time windows and cohorts separate.
Requested history is not guaranteed coverage. The qualified 30-day time-weighted price uses at most 48 hours of carry and needs 24 covered days. Watchlist seven-day mean changes and the 30/90/180-day Price Index have different calculations. Standard list-cost ranking separately requires eligible USD same-pack evidence and source prices no more than 48 hours old, with a valid fee profile; this is a gate, not a promise that every listing is always current.
A tracked portfolio, discovery catalog and index basket also have different boundaries. Concentration within one set is not whole-market share. Readiness and coverage scores are research heuristics rather than accuracy or success probabilities.
Separate models, messages and outcomes.
Contribution equals planned sale minus purchase reference minus the complete included cost total, before income tax. Unknown required costs are not replaced with zero; missing or invalid inputs prevent a complete total and contribution. Buyer sales tax is pass-through, excluded from revenue and costs, but can affect fee and processing bases. This is a cost scenario, not a realized net-profit statement. Inventory quantity, holding period and fee allocation remain assumptions unless supported by your own evidence.
Report selection and aggregation depend on period and scope. Channel readiness or transport acceptance is not proof of a human read. When a metric raises a new question, open the product, source date and calculation rather than treating the headline as a completed commercial conclusion.
Questions about this workflow
Why does a value stay unknown?
Because no eligible observation or complete input supports it. Substituting zero would change the result’s meaning.
What should I inspect when a total looks surprising?
The underlying records, source dates, period, cohort and calculation. Each feature page explains the relevant route back to that evidence.