Skip to content
SCAMWIZ / PUBLIC BETA

How ScamWiz researches and scores a merchant

Updated November 6, 2026

Current rubric: merchant_heuristics_v1. Deterministic checks inspect first-party evidence. Scores are experimental, not certifications, fraud probabilities or validated safety ratings.

1. Retrieve public pages

The scanner reads the submitted HTTPS page and up to six linked policy pages from the same origin, prioritizing returns, shipping, billing, contact, terms and privacy. Requests have size and time limits and public-network restrictions. No JavaScript, login, checkout or purchase is executed. Bot blocking, region-specific content and JavaScript-only pages can reduce coverage.

2. Apply published rules

HTML is parsed into readable text. Valid JSON-LD is parsed separately. English-language patterns look for concrete statements rather than navigation labels. Each assessed finding carries a literal excerpt matched to its retrieved source. This establishes provenance, not the truth or applicability of merchant claims. No Gemini request, Google Search grounding or external reputation lookup is performed.

The fifteen checks

Contact channel

Found only when a non-example email address is published. A navigation label alone is insufficient.

Named business

Found when an explicit company-name suffix or parsed Organization legalName/name is present. Presence does not verify the entity.

Concrete return terms

Found when a return-window or concrete fee/exclusion/refund rule is detected.

Concrete shipping terms

Found when a shipping/delivery statement states a fee, free-shipping condition or delivery duration.

Billing/cancellation terms

Found only from explicit paid recurring terms or customer order-cancellation instructions. Newsletter unsubscribe is excluded.

Substantive privacy statement

Found when privacy wording occurs with a concrete data collection, processing or sharing statement.

Return window

At least 30 days = more favorable; 7–29 days = mixed; fewer than 7 days = concern. Multiple differing windows stay unknown because eligibility can differ.

Who pays for return shipping

Unconditional merchant-paid/free return shipping = more favorable; customer-paid or conditional coverage = mixed. Customer-paid wording takes precedence when optional coverage is also mentioned.

Restocking fees

Explicit no restocking fee = more favorable; positive fee up to 15% = mixed; above 15% = concern. Absence of a fee mention stays unknown.

Non-returnable / final-sale restrictions

Explicit non-returnable or non-refundable final-sale restrictions = mixed. Generic final-sale navigation and negated restrictions do not count.

Recurring billing

Explicit automatic renewal/recurring charges = mixed. Explicit one-time/non-recurring purchase wording = more favorable. Silence or a product price is insufficient.

Customer cancellation steps

Explicit self-service customer cancellation = more favorable; required phone/email/support contact = mixed. Merchant rights to cancel and newsletter unsubscribe are excluded.

Valid structured product offer

Parsed Product JSON-LD with a named product and numeric offer price/currency = more favorable. Invalid JSON or loose Product markers do not qualify.

Product price information

A valid product offer, or a currency/price in a bounded product-card block with an add-to-cart/quick-add action = more favorable. Shipping prices do not qualify.

Readable policy detail

At least two concrete return/shipping/billing/privacy categories found = more favorable; one = mixed. Mere links are insufficient.

3. Calculate scores and show coverage

Disclosure completeness is the percentage of six fixed categories detected. Missing categories may be inapplicable and are not proof of fraud. Purchase-term favourability and agent information access use more favorable = 100, mixed = 50 and concern = 0, averaged over assessed checks only. At least two assessed checks are required in each of these dimensions. Unknown checks are excluded; every score displays its coverage. No overall score is assigned. The rubric has not been calibrated against real-world outcomes.

Ambiguity and failures

Different detected return windows remain unknown because eligibility may differ. Optional coverage does not override explicit customer-paid return shipping. Newsletter unsubscribe and merchant rights to cancel are excluded from customer cancellation findings. Static English patterns can still misinterpret nuanced wording; inspect the citations. Failed retrieval produces unscored dimensions. The same retrieved pages produce the same findings; merchant content can change between scans.

Owner decisions and corrections

Merchant identity, independent reputation and fulfillment remain unverified. The owner decides whether to share personal information or purchase. Question a finding with the domain, scan time, source and disputed interpretation. Recorded reports retain their original date and method and are not live endorsements.