The rules Coreplex follows
How Coreplex labels where information comes from, which source wins when two disagree, what counts as evidence, and which reviews qualify. Written so people and AI systems can check Coreplex against its own rules.
Principles
- AbstainIn effect
Coreplex makes no statement it cannot identify, scope and source. When it cannot, it names the gap or says nothing, and the rest of the page still works.
- Source is not truthPolicy
A source saying something does not make it true. A seller field, brand copy, review or social post is first an observation, and only becomes usable evidence after it is classified.
- Three layersIn effect
Facts, Coreplex-derived intelligence, and narrative (seller marketing, reviews, creator content) are kept apart. Narrative stays visibly attributed and is never presented as a verified fact.
- Seller of recordIn effect
The seller stays the seller. Checkout, payment, tax, shipping, returns, refunds and warranty are the seller's. Coreplex adds no markup or fees and does not take payment.
- Payment never buys rankIn effect
Payment, commission or sponsorship never buys eligibility, completeness, labels or rank. Today Coreplex has no sponsored placements.
- Observation is not specificationPolicy
Views, likes, trends, popularity and review volume describe attention. They do not prove quality, efficacy, suitability, safety or superiority. Coreplex may say people mention something; it does not turn that into a product fact.
- Ai limitsPolicy
AI may help extract, normalize and write, but it is never the sole authority for a critical field, a claim or a disclosure. Generated text cannot introduce a fact that is not in the governed record.
Where each value comes from (provenance classes)
Every value Coreplex shows has a source class. Coreplex's evidence contract defines five roles; the product also uses the specific labels below.
Roles
| Role | What it is | How it is treated | Status |
|---|---|---|---|
authoritative_primary | Manufacturer technical specification, official product label, verified regulatory source. | Used as evidence once classified | Policy |
merchant_operational | What the seller reports about now: current price, availability, shipping policy. | Used as evidence once classified | In effect |
brand_supplied_marketing | Selling points, promotional copy, benefit language. | narrative, attributed to the seller or brand | In effect |
observational | Customer reviews, social posts, editorial discussion. | attributed opinion, never a specification | In effect |
unclassified | Kept, but never used for a strong conclusion. | Used as evidence once classified | Policy |
Labels used in the product
| Label | Meaning | Status |
|---|---|---|
SELLER_STATED | Read from the seller's own title, option values and description. It is the seller's statement, not a verified fact. If the text is ambiguous (two different counts or strengths), nothing is stated. | In effect |
MERCHANT_STATED | A specification shown with the seller's figure because the catalog's inferred figure disagreed with it. | In effect |
SELLER_PAGE_OBSERVED | Content Coreplex read from the seller's own product page. | In effect |
SHOPIFY_NORMALIZED | A field taken from Shopify's catalog and put in a common shape, such as the brand or vendor name on a listing. A vendor field that only repeats the storefront name is not promoted to a brand. | In effect |
SELLER_DOMAIN_MATCH | A brand inferred because a store sells under its own name on its own domain. It stays inferred until the brand confirms it, and is never applied to known multi-brand retailers. | In effect |
SHOPIFY_INFERRED | Features and specifications that Shopify's catalog infers. They are discovery signals, not text the seller wrote. | In effect |
Computed by Coreplex
Price comparisons, match labels and unit math that Coreplex computes. Each carries its inputs, rule, freshness and scope, and the passport names the matcher version and the evidence for every match. In effect
Marks on listings
| Mark | Meaning | Status |
|---|---|---|
official_store | The listing is sold by the brand's own store, one confirmed as the seller of its own listings in the live Shopify catalog. | In effect |
clean_label_project | The listing is matched to a published Clean Label Project certificate for that exact product. | In effect |
frontrowmd | The seller's pages carry FrontrowMD's clinician badge. | In effect |
store_lists_was_price | The store's own compare-at price, shown as the store's statement. | In effect |
When sources disagree (what wins)
What wins when two sources disagree.
- Rule 1In effect
The seller's own statement on its listing or product page wins over data Shopify's catalog infers. Price and stock are always the seller's own.
- Rule 2In effect
If a catalog-inferred specification disagrees with a figure the seller states, the seller's figure is shown and labeled as the seller's. If an inferred feature disagrees, it is withheld. The disagreement is recorded.
- Rule 3In effect
An inferred number is shown only when the seller states the same number with the same unit. Otherwise it is withheld. If the seller's page could not be read on that request, inferred numeric claims are withheld rather than shown unchecked.
- Rule 4In effect
Inferred text that is not numeric (a material, a connection type) can be shown, labeled as catalog-inferred.
- Rule 5In effect
In supplements, catalog-inferred data is not shown to shoppers. Claims such as certifications or nutrient amounts go through Coreplex's governed evidence process, and what a seller says is shown as seller-stated.
- Rule 6In effect
A price is called lowest only among sellers Coreplex matched to the exact same item. Similar listings, other versions and bundles are shown separately and are never counted as savings.
- Rule 7In effect
Commercial relationships are not an input to how results are ordered or labeled. Sponsored items, if they ever exist, are labeled on the card, never first, at most one per rail, and never count toward price claims or the seller comparison.
What counts as evidence
- TraceablePolicy
Anything used in a decision must trace to a source assertion, a normalization, a deterministic or statistical derivation, a governed model-assisted derivation, or a human-reviewed determination. If Coreplex cannot explain where an important conclusion came from, it does not present it as strongly supported.
- Support statesPolicy
The evidence contract defines five support states: verified, strongly supported, supported, conflicted, insufficient.
Today: Listing facts are shown as supported with the basis "seller stated". Coreplex does not independently verify them and does not lab-test products.
- Missing is not falseIn effect
A missing value is not read as zero, false or an average. When text is ambiguous, nothing is stated.
- Grounded numbersIn effect
A number from an inferred source is shown only when the seller states it too (see source precedence).
- Same itemIn effect
Two listings are the same item when Shopify lists them under one catalog product or they share a barcode. Passports also use brand, product line, formula and the size or color named. A formula difference in the title (for example caffeine-free or a PM formula) makes a listing a different product even when the catalog id matches.
- Match labelsIn effect
Match labels, strongest to weakest: same item; likely same item; other version; customized or company merch; bundle or multipack; accessory or part; used; uncertain. Only same item is price-compared.
- CoverageIn effect
Coverage is partial. Only Shopify stores in Shopify's catalog are checked, and a missing seller is not proof it does not sell the item. Tools say when coverage is incomplete.
- FreshnessIn effect
Prices and stock are read from the live catalog when a tool is called. The product feed can be cached up to 5 minutes. A passport shown on coreplex.ai can be up to 6 hours old.
- Health claimsIn effect
Health and certification claims (HSA/FSA eligibility, USP, NSF, Informed Sport, GMP, ConsumerLab) are what the seller states, not verified by Coreplex. Health and supplement content is general information, not medical advice. The label on the product you receive is the final word.
- Version historyPolicy
Material changes to identity, ingredients, allergens, certifications or destination create a new product version. They never erase the earlier context.
- ScopeIn effect
US shoppers and USD prices. Search results include only listings that are in stock.
Reviews: what qualifies
Today In effect
Coreplex does not collect reviews of its own. It shows ratings and reviews that sellers publish on their own pages, labeled with the seller and linked to the source.
A review is shown only if:
- The review has text.
- It is about this product: the exact variant, the product, or its product family. Reviews about several products, a brand in general, a category, or an unrelated item do not qualify.
- If it names an exact variant, that variant match must be resolved.
- The match is not low confidence.
- Review summaries written by a source are not shown. Coreplex shows individual reviews, labeled as the reviewer's opinion.
How it is treated. Attributed opinion. A review is never turned into a product fact. A rating is the seller's, and review volume is not evidence of quality.
Derived statements. Policy Any statement derived from many reviews (for example, share of reviews that mention fit) must keep its cohort, sample size, time window, method and source references, and does not become a product specification.
Coreplex's own reviews Not live
Coreplex does not host or collect its own reviews today. If that changes, the rules for which reviews qualify will be published here before they take effect.
Corrections
If a match, price, label or claim looks wrong, report it with the product URL. Contact Coreplex.
What Coreplex does not do
- Take payment or hold a card number.
- Hold inventory or ship anything.
- Lab-test products or verify NSF or USP status unless a page says so.
- Rank sellers for money.
- Check out on behalf of an agent. Agents send the shopper to the seller's product URL.