Decentralised News · Agentic Finance · Decision research
By Heath Muchena · Source review 2026-11-14 · Instrument version 1.0.0
What Matters
Two different checkout totals establish a difference between two observations. They do not establish that a merchant charged more because one shopper used an agent. Freeze the product, seller, market, account benefits, timing and purchase terms before investigating the access channel. DN’s paired-price log excludes visibly unmatched observations and keeps causal attribution unresolved.
An agent returns a price of 105 currency units. A person opens the store and sees 100. It is tempting to call the difference an agent surcharge. It may instead reflect a different variant, seller, shipping option, account discount, market or moment in time.
This is a protocol for investigating the discrepancy. It can be used during a major sale or an ordinary shopping session. DN has not run a Black Friday comparison or established that a named merchant discriminates against agents. The contribution is an auditable observation method, a local paired-price instrument and clear boundaries around the conclusions.
DN Evidence Block
Sources reviewed: 2026-11-14 · Evidence type: Primary pricing documentation, bounded regulator findings and an original DN controlled-observation protocol.
Scope: Permitted comparison of displayed offers or checkout totals for the same intended purchase, without requiring a completed order.
- The FTC’s January 2025 preliminary surveillance-pricing findings describe capabilities and hypothetical public examples; they do not establish that a named merchant charges agents more.
- Google Merchant Center documentation distinguishes feed, landing-page, variant, currency and checkout-price consistency. Shopify documents market-specific pricing and currency conversion.
- DN’s seven controls define what must match before an observed price difference is worth investigating. Passing them is not a causal finding.
Not established by this work: No merchant, customer account, agent session or purchase was tested. The examples are synthetic. The tool does not prove price discrimination, infer a pricing algorithm or calculate statistical significance.
Original DN instrument · version 1.0.0
DN Paired Price Observation Log
Summarize permitted paired observations only when the entered conditions match and the time gap falls within your declared window. Every result remains descriptive; no price-cause verdict is produced.
Enter your assumptions or load a labelled example. Required blank numbers stay missing; unknown evidence remains unknown. Use nonpersonal labels only. This instrument makes no network requests and uses no persistent browser storage.
Use your own inputs
No current result. Complete inputs or load an illustrative example.
One currency and one price stage per run. No statistical test, live fetching, account access or causal identification. A yes flag is the user’s assertion; the tool does not inspect screenshots or verify conditions. Synthetic examples are not market evidence.
1. Freeze the exact product and quantity
Start with the item that would actually be delivered. Keep the product identifier, variant, size, quantity, condition and included accessories fixed. A product name is not a sufficient match. A bundle with a charger, a refurbished unit and a new unit can share a search phrase while carrying different prices and obligations.
Google’s inaccurate-price guidance specifically points to default-variant mismatches. That is a concrete reason to record the selected variant at the time of each observation rather than relying on the page title.
For an experiment, write the purchase specification before opening either channel. If the agent substitutes an alternative, preserve that as a separate finding about product selection. Do not place the two prices in the matched-price sample. An apparently cheaper result that fails the specification is not a discount on the same purchase.
2. Match the seller and the complete offer
A marketplace can show the same product from different sellers. Freeze seller identity, delivery service, warranty, return conditions, subscription commitments and any required membership. Compare the complete offer the buyer would accept, not just the most prominent number.
Record the item amount and each additional charge separately, then compare the final payable total at the same stage. A product-page estimate and a checkout total with delivery are different observations. Google’s feed and landing-page guidance addresses price consistency and the treatment of handling charges; it is not a universal guarantee that every displayed offer is complete.
Set the experiment’s stopping point. Usually it can stop before placing an order. If a question requires an actual purchase, that needs an approved budget and clear cancellation or return terms. Buying and returning items is not a cost-free measurement method.
3. Keep market, currency and destination constant
Use the same declared market, currency and delivery region. A location difference can affect tax, shipping, availability or the commercial price. Converting two totals with an external spot rate does not necessarily reconstruct what either shopper was offered.
Shopify’s international-pricing documentation describes automatic currency conversion, manual exchange rates and market-level price adjustments. These are documented mechanisms by which different market contexts can produce different prices without any agent-specific rule.
Record only the nonpersonal region needed to assess comparability in the public research log. Keep any necessary full address out of shared exports. If the delivery context is missing, mark the pair unmatched or unresolved. Do not assume two sessions share a market because the browser interface uses the same language.
4. Separate account benefits from the access channel
A signed-in customer may have a loyalty benefit, saved coupon, membership price or negotiated business rate. A fresh agent session may not have the same entitlement. The reverse can also occur if an agent discovers a publicly available offer the manual path did not apply.
Shopify’s discount documentation provides a bounded example of percentage and fixed-amount discounts interacting with local currencies. The experiment should retain the actual accepted discount and its eligibility conditions rather than treating every visible code as usable.
Choose the comparison before collecting observations: equal entitlements across channels, or the realistic experience each channel provides. Both can be useful. They answer different questions. A realistic-experience study may show a customer loses a benefit when using an agent, but that does not by itself identify a discriminatory pricing rule.
5. Bound the time gap and record changing inventory
A price observed at 09:00 and another observed at 17:00 can reflect a repricing event. Promotional eligibility, available stock and delivery capacity can also change. Predeclare the largest time gap you will accept for a pair, retain actual timestamps and record the order in which the two paths were observed.
A narrow window reduces one source of ambiguity; it does not eliminate it. Alternate the order of observations across repeated pairs where practical. If the manual path always runs first, a systematic time trend can look like a channel effect. Never silently remove the inconvenient observations after seeing their prices.
The DN instrument accepts the elapsed minutes and a user-chosen maximum. Pairs outside that window are excluded from its matched summary. The threshold is a study choice, not a scientifically established safe interval. Inventory or promotion changes can still invalidate a pair inside the chosen window.
6. Treat agent identity as the variable under investigation
Only after the other conditions are controlled should you examine the channel: browser, referral, session history, device presentation or an explicitly identified agent request. Change one declared factor at a time where the service permits it. Do not bypass access controls or impersonate a service to make a comparison work.
The FTC’s preliminary findings describe intermediaries capable of using consumer and contextual information in pricing and promotions. The public examples are hypothetical. This supports investigating context carefully, not accusing an identified merchant on the basis of one screenshot.
Keep a rival-explanation column: cached offer, changed seller, different campaign, unsupported coupon, wrong destination or unresolved session state. If you cannot isolate the factor, report the observed channel difference and the unresolved explanations. That result is still useful for a customer deciding whether to review an agent’s proposed total.
7. Report matched observations before making a causal claim
Repeat the comparison with a written inclusion rule. Preserve failed attempts, missing totals and exclusions, because a study containing only successful checkouts can hide an important part of the experience. Predeclare the outcome: total payable amount, price at a specified stage, or availability of an eligible offer.
Synthetic example: pairA records 100 versus 105 within one minute and matched conditions; pairB records 100 versus 98 within two minutes; pairC records 100 versus 110 but has unresolved conditions. With a two-minute limit, two pairs enter the summary and the unresolved pair does not. The median relative difference is 1.5%. These invented values do not describe a store or prove a cause.
A credible causal study needs more than this descriptive log: a defensible assignment design, replication, adequate sample planning and a method that addresses confounders. The instrument intentionally does not attach a confidence score or discrimination verdict to a few user-entered prices. Its export preserves the observations, exclusion reasons and the attribution boundary.
Methodology, limitations and corrections
DN reviewed the primary sources on 14 November 2026 and designed this protocol for an unresolved price-difference question. It consolidates the proposed seasonal shopping tests into the same controlled-observation method. No sale-period result is claimed.
| Record | Retain privately or redact for publication | Decision effect |
|---|---|---|
| Purchase specification | SKU, variant, quantity, seller and terms | Different offers are excluded from matched-price claims |
| Session context | Market, currency, benefits and declared channel | Unresolved context remains a rival explanation |
| Observation timing | Both timestamps, order and elapsed minutes | Out-of-window pairs are excluded by a predeclared rule |
| Outcome | Same-stage total, missing result or failure | Failed attempts remain in the study ledger |
For accepted pair i, the amount difference is agent total minus control total. Relative difference is 100×difference/control total; the control total must be positive. The tool reports the mean amount difference and median relative difference across accepted pairs. It does not combine currencies, weight by transaction volume or estimate a population effect. The worked sample’s median 1.5% is an illustrative arithmetic result.
A report should disclose the selection frame, attempt count, exclusions, observation dates and permitted access method. Publishing only the largest price gap would undermine the experiment. A claim about why a price changed needs a stronger design and supporting evidence than this descriptive log provides.
For corrections, identify the claim, source and date when contacting Decentralised News. A new record-creation timestamp does not refresh its underlying evidence.
Change log: version 1.0.0, 2026-11-14. Initial documentation review, original instrument and labelled examples.
Frequently asked questions
Does a higher agent price prove discrimination?
No. It establishes a difference between observations. Product, terms, market, benefits and time may differ, and an observational comparison can leave other causes unresolved.
Do I need to buy the item to run this protocol?
Usually no. Declare a price-observation stage and stop there. Any experiment requiring a purchase needs its own budget and accepted terms.
Why exclude an unresolved pair?
Its price may be useful as a diagnostic observation, but it cannot support the same matched-price summary. The exclusion remains visible rather than being silently discarded.
Is a two-minute window always valid?
No. The example window is a study assumption. Fast repricing or stock changes can invalidate a pair even inside that window.
Can I compare prices in different currencies?
Not in one run. Currency conversion, market pricing and charge composition introduce additional assumptions that this tool does not normalize.
Primary sources
- FTC: preliminary surveillance-pricing findings. Contextual pricing capabilities and the hypothetical nature of public examples; no named-merchant agent surcharge finding. Reviewed 2026-11-14.
- Google Merchant Center: inaccurate prices. Variant and price-consistency issues relevant to matching observations. Reviewed 2026-11-14.
- Google Merchant Center: feed and landing-page inconsistency. Product, price and additional-charge consistency requirements in this specific service. Reviewed 2026-11-14.
- Shopify: pricing in local currencies. Market-specific pricing and currency-conversion choices. Reviewed 2026-11-14.
- Shopify: discounts in local currencies. Discount behavior relevant to account and currency comparability. Reviewed 2026-11-14.
Turn the finding into a decision
A merchant investigating inconsistent offers can ask the DN Research Desk to scope a permitted comparison study for a named product and market. Provide redacted observations and the decision you need to make. Feasibility, access and a written quote come first; no discrimination finding or price advantage is promised.
Related DN research
Disclosure: Analysis and instrument results are editorial. Where labelled affiliate links appear, DN may receive compensation when you sign up or buy through them. Commercial relationships do not determine findings. No tested integration or guaranteed outcome is implied.
