What belongs here, and what never will

Why we refuse to be a replica index

The compliance answer is short: selling counterfeits is illegal in every market Lantern operates in, purchasing agents refuse the parcels, customs seizes them, and a platform that indexes them is a party to the harm. That answer is true and it is not the interesting one.

The real reason is that a replica index cannot do the job Lantern exists to do. The value of this platform is judgement: telling you that one hoodie has a better-weighted fleece than another at the same price, that a lamp's diffuser is worth the extra fifty yuan, that a bag's hardware will fail in a year. A replica has no properties of its own to judge; its only claim is proximity to something else. The moment a catalogue starts ranking by proximity to a brand, it stops being able to rank by quality, and every editorial signal it publishes becomes noise. There is no version of Lantern that does both.

It is also a positioning decision we would make even if the law were silent. Replica traffic is enormous, low-trust and disloyal: it arrives from a search for a brand name, converts once and never forms a relationship with a catalogue. The audience we want is the person furnishing a flat, building a wardrobe, kitting out a desk: someone who will come back fifty times and eventually trust our verdicts. Those two audiences cannot share a homepage. We picked one.

Practically, this means Lantern indexes original goods and unbranded goods, and nothing that trades on a mark it does not own. The Chinese marketplaces are full of genuinely excellent original design: domestic labels, small workshops, factory-direct homeware, ceramics, tools, stationery. That is the catalogue. It is bigger than the replica one and nobody has indexed it properly.

We do not publish brand names as design references, we do not accept “looks-like” comparisons in reviews, and we do not run the search terms the replica market uses. If you arrived here looking for those, this is the wrong site and politely intends to stay that way.

Prohibited content

The following may not be submitted, published, linked to or described favourably anywhere on Lantern: not in product submissions, reviews, photographs, captions, collections, hauls, comments or messages.

Restricted, allowed with care

These are not prohibited, but they carry conditions. Where a condition is not met, the page comes down at review and nothing else follows from it: a missed condition is a removal, not a strike.

What we actively want

A policy that only lists prohibitions tells you nothing about what the catalogue is for. Everything published stays up unless there is a reason to remove it, and the submissions that survive review and end up in guides tend to look like this:

• Original design from a domestic label: a Chinese brand with its own aesthetic, its own store and its own following.
• Factory-direct goods with a real spec: a stated fabric weight, a measured lumen output, a named steel grade, a listed capacity. Numbers make a product reviewable.
• Unbranded staples that beat their price: the plain thing that is quietly better made than the expensive version.
• Evidence: quality-control photographs, owner photographs after months of use, measurement tables, a note about the seam that failed. Evidence is the scarcest thing in this market and the most valuable contribution you can make.
• Honest negatives: a review that says the colour is two shades off and the zip snags is worth more to us than five stars and no detail.

If you have something that fits, use submit a product. If you want a sense of the bar, read a finished guide such as Best Taobao Finds or browse the QC photo wall.

How detection works

Nothing here is magic, and we would rather describe it accurately than imply a capability we do not have. Four layers do the work, and the last one is people.

1. Submission screening

Every submission is checked against a maintained list of replica vocabulary (grade and batch terminology, tier language, factory shorthand and the deliberate misspellings used to evade filters) in English, romanised Chinese and simplified Chinese. It is also checked against a list of protected marks and model designations. This check runs before anything is written, and a match refuses the submission at the form rather than publishing it. The matched term is shown, because a false positive on a legitimate product is a real cost and you cannot argue with a rejection you cannot see. Nothing else about a submission is decided by a machine: pass this and the page is live.

2. Seller-level assessment

The stronger signal is almost never the individual listing: it is the shop around it. We look at the seller's wider catalogue, how long the store has operated, whether its product mix is coherent, and whether the same photographs appear across unrelated stores. A seller whose catalogue is predominantly replica is excluded wholesale, including the legitimate items, because indexing one product from that shop sends buyers into the rest of it.

3. Image and text similarity

Submitted and community photographs are compared against each other and against known marketing imagery to catch reused photos, stolen product shots and the same item resubmitted under new titles. Listing text is compared for the templated descriptions that replica sellers copy between shops.

4. Community reports

The most accurate detector we have is a person who owns the thing, and since nothing is held back before publication, this layer carries more of the weight than the three above it. Reports from buyers, designers and rights holders find what automation misses: the logo that appears only on the inner label, the pattern lifted from an independent illustrator, the listing whose photographs belong to somebody else. Reporting is one click from every product page, and reports carry weight in proportion to their track record.

Imagery, stated honestly: a product page shows generated artwork rendered from category, subcategory and colour metadata, photographs a member uploaded, or listing photographs resolved from the marketplace or from agent search services. Community and quality-control photographs are contributed by people who received the item and are labelled as such.

Publication and review

Submissions are published immediately. Nobody reads them first, a person reads them afterwards, and what breaks this policy is removed when we find it or when you tell us. The automated check in the section above runs before the page is written and refuses what it matches; past that point a submission is a live page in the feed, in search and in its category, the moment it is filed.

That is a real trade and it is worth stating which way it runs. An approval queue read by one person either moves slowly, which costs a contributor days before they can see their own work, or moves fast, which is approval in name only. Publishing on arrival moves the cost of a bad submission from “a contributor waits” to “a wrong page was up for a few hours”, and we pay for that in three ways. A page nobody has read yet says so on its own face, so nobody is told a listing has been checked when it has not. Reporting is one click from every product page and is now the main way a bad listing gets found. And a removal here is a live page disappearing, so the person who submitted it is told what came down and why.

Everything that needs a person (published submissions nobody has read yet, reports, appeals and automated flags) lands in one queue, triaged by severity rather than by arrival time.

• Priority 0: child safety. Handled immediately, removed on receipt, and referred to the appropriate authority. Never subject to a hold pending review.
• Priority 1: counterfeit, malware, fraud, weapons, drugs, hazardous goods. Content is unpublished as soon as the report is opened, usually within 1-2 days, before the review concludes. We would rather briefly remove a legitimate listing than leave a harmful one up while we deliberate.
• Priority 2: copyright, trademark and design-right complaints. Handled through the formal process in copyright and trademark, with removal on a valid notice and a counter-notice route for the other side.
• Priority 3: accuracy, dead listings, mislabelled specs, review-quality problems. Reviewed within five working days; usually corrected rather than removed.
• Priority 4: published submissions nobody has read yet. Already live, read in order of arrival, typically within five working days and longer for a category we have not covered before. What this review decides is whether the page stays, not whether it appears.

Decisions are made by a human. Automated flags queue work; they do not by themselves remove content or issue strikes. Every decision is recorded with the rule relied on, which is what makes the appeal in the next section possible.

Review is a policy and safety check made after the fact, nothing more. A page that has been read and left up means a person looked at it against this document and found no reason to take it down. It does not verify the accuracy, availability, quality or legality of the product, it is not an endorsement of the seller, and it does not make the submission Lantern's own statement. Responsibility for a submission stays with the person who made it, before review and after it.

Enforcement ladder

Enforcement runs against the contributor and, separately, against the seller or source. The steps escalate, and the intent is correction rather than punishment, except at the top, where it is exclusion.

Severe breaches skip the ladder entirely and go straight to a ban: counterfeit promotion at any scale, malware or phishing, coordinated review manipulation, and anything in the child safety category. Everything else starts at removal.

Appeals

Every enforcement decision can be appealed, and we treat appeals as a quality-control mechanism on our own moderation rather than as a complaints department.

• How. Email trust@lantern-finds.com with a link to the removed content or a description of the restriction, and anything that supports your case: an authorisation letter, a licence, a store certificate, photographs of the item in hand.
• When. Within 30 days of the decision. Later appeals are considered where the evidence is genuinely new.
• Who reviews it. Someone who was not involved in the original decision.
• How long. Ten working days, and we tell you the outcome with the reasoning, not just the result.
• What a win means. The content is restored, the strike is removed from the twelve-month record, and any restriction that followed from it is lifted. If a pattern of similar overturns emerges, the rule or the detection list gets changed, which is the actual point.

Bans for child safety and for malware are not appealable. Bans for counterfeit promotion are appealable once, and only on evidence that the goods were authorised.

Repeat infringers

Strikes expire twelve months after they are issued. Three live strikes result in a ban. Attempting to evade a ban (new submissions from the same seller under a different store name, or contributions from an obviously linked source) resets the clock and makes the ban permanent.

Exclusions are maintained at the seller and shop level as well as the contributor level, and the list is checked on every submission. This is the mechanism that keeps a removed counterfeit from reappearing next month under a new title, and it is the reason a single replica listing costs a seller their whole presence on Lantern rather than one page.

Rights holders who repeatedly file notices that turn out to be invalid are also handled under this section: the notice route stays open, but the accelerated removal that a trusted-reporter status buys is withdrawn.

How to report something

Use the report control on any product page, or the report a product form. Tell us what is wrong in one sentence and give us whatever evidence you have: a link to the original design, a photograph of the mark on the item, the name of the shop selling the same photographs.

Rights holders can use the same route, but the formal notice in copyright and trademark gets a faster and more structured response. Safety issues, scams and anything involving a minor should go straight to trust@lantern-finds.com.

You do not need an account to report, and you do not need to leave a contact address, though we cannot tell you the outcome if you do not.

Transparency

We publish what we remove in aggregate: the number of submissions refused at the form, pages removed after review, pages removed on a report, strikes issued, bans applied, appeals received and appeals upheld, broken down by policy category. Because nothing is held back before publication, the removal figures are the whole story rather than half of it, and anything less makes a document like this unfalsifiable. Where a removal was required by a legal notice, the notice type is recorded too.

This policy is part of the terms of service, and breaching it is a breach of those terms. It is reviewed twice a year and whenever a new evasion pattern shows up in the queue.