What makes two prices comparable.

The only claim shape

On this date, for this product, observers saw prices differing by X%. Fixed in advance, method version N

• Must rule outcurrency, delivery, tax, region, stock, A/B test
• Never published"prices vary"
• Never publishedany claim about a company's intent
• Never publisheda figure without its method version

The crowd is a sensor, never the proof

• Tier oneobservations from the extension
• Tier one mayrank what is worth measuring next
• Tier one may neverbecome a published number
• Tier twoa controlled run we perform ourselves
• Tier two needsone window, one product, several vantages
• Cost of a false submissionone wasted verification run

The item is the comparable number

• Compared on the total$251.50
• Compared on the item$229.00
• Every rule below applies tothe item, never the total
• A raw spread on totalsis mostly delivery

The seven rules, in this order

Applied in sequence · The rule that removed an observation is recorded with it

• 1. Not the same offerA different size, colour or seller is a different thing
• 2. The gap was delivery, tax or a feeCompare the item alone, never the total
• 3. A different currencyInside the published spread between two currencies
• 4. A published regional priceThe retailer already states it prices that region differently
• 5. A different stock stateBackordered and in stock are not the same offer
• 6. An A/B test or a live repricingOne profile shown two prices means testing, not targeting
• 7. Outside the comparison windowA gap across hours is a price change, not a difference

• Survives all sevena candidate, still not a finding
• A candidate needsreplication on different days

Thresholds, fixed in advance

• Minimum publishable difference5% of the base price
• Minimum vantage points6
• Before naming a retailer3 runs on 3 different days
• Premise testunder 1 in 10 products, the premise fails
• If the premise failswe say so publicly
• A threshold changes only bya new version of this method
• Published findings keepthe version they were made under

What this method forbids

• Publishing a crowd average as a finding
• Publishing any figure without its method version
• Changing a threshold after seeing a result
• Characterising a company's intent or conduct in our own voice
• Any affiliate or referral link inside a comparison
• Charging researchers or academics for access

Version history

• v1.0 · 19 August 2026initial pre-registration
• Written beforethe extension's schema
• Every figure on this sitecomputed, never typed