How it works

You choose one photo of one piece. A few seconds later you are reading a written identification of what is in the frame, with the honest part — how usable the photograph was and what it cannot settle — visible before you pay for anything.

Upload a photo

What happens to your photo

You pick a photo. Camera roll, drag and drop, or the camera on your phone. One piece per photo; a close-up of a mark on a piece counts as a valid photo on its own.
Your browser prepares it before anything leaves the device. The file is decoded, converted to JPEG, PNG or WebP and shrunk so its longest edge is no more than 2048 pixels and the file fits under 10 MB. HEIC shots from an iPhone are converted here too, when the browser can decode them. A picture smaller than 320 pixels on its shortest side is refused at this point, because it was never going to carry a reading.
The prepared photo is sent once. The server checks the file type, size and dimensions again, holds the image in memory for that single request, and passes it to the vision model. It is not written to disk, to object storage or to logs. If the frame does not show a piece of jewelry or a mark on one, the request is refused instead of written up.
The model writes the whole report in one pass. Every section and every analytical block is produced together, in a fixed structure, under instructions that keep observation and interpretation apart.
The preview appears immediately. Part of the report is held back until you unlock it; nothing is re-analysed when you do.

More on the photo itself in the identification guide and in how to identify jewelry from a picture.

What the free preview shows

The free part is the part that tells you how much to trust the rest. You get:

What this piece is — the identification section in full: object type, how many pieces are in the frame, the visible construction, the fastening or setting, the apparent metal colour, the count and arrangement of stones, and the most likely reading with the reasons behind it.
Photo quality — a score for how usable the photograph itself was, with the specific problems named: glare, blur, shadow, colour cast, crop, distance, a busy background.
Limits of this reading — what this particular report could not judge, such as a face of the piece that never appeared in the frame.
What a photo cannot settle — the things no photograph settles, however good it is.

The headline confidence score sits in the preview too. Hiding it would read as a missing number rather than a locked one, and it is the number that tells you whether to keep reading.

What one analysis credit unlocks

The rest of the same report opens in place:

Visible material clues — metal colour and lustre, plating wear at edges and high points, patina, solder lines, two-tone construction, and how the stones read optically. Each clue comes with what it is consistent with and what else looks identical.
Markings and hallmarks — whether a mark is in the frame, a character-by-character transcription of what can genuinely be resolved, the punch outline, where it sits, and what such marks carry. Characters that cannot be read stay unread.
Style and period clues — the design vocabulary, the style families it belongs to, and what argues against each suggestion.
What to check next — which photo to retake and of exactly which part, what a loupe and a gram scale would show, what a magnet test does and does not prove, and which question belongs to a jeweller, a gemmologist, an assay office or an appraiser.
Most likely identification and What it could be instead — the ranked candidates with their confidence, and the readings that were considered and set aside with the reason.

One credit unlocks one full report. Packs and what they cost are on the pricing page, and you can see how far the free preview gets you before deciding — start with a photo.

The two scores, and what neither of them is

Confidence is how strongly the visible evidence supports the identification as a whole. Photo quality is a separate judgement of the picture. They move independently: a sharp, well-lit photograph of an unmarked plain band earns a high quality score and a modest confidence, because there is little in the frame to reason from. A blurred photograph of a heavily marked brooch can run the other way.

Neither number is an accuracy guarantee, and neither is a grade you can appeal. A confidence of 60 next to “the underside was never in the frame” means one more photograph would probably settle it. The same 60 next to “this cannot be distinguished from a photograph” means the next step is a person holding the piece.

Report links, retries and errors

When storage is enabled, the report is saved under an unguessable link. On this deployment a report that has not been unlocked is kept for around three days and an unlocked one for a year; the link arrives in your receipt email and report pages are excluded from search engines. Credits you have not spent stay on that same link.

If the analysis fails for a reason that could pass next time — a timeout, a provider hiccup, a malformed answer — a retry button appears and re-sends the photo you already chose. If the failure is one that would repeat, such as a photo of something that is not jewelry or a picture too small to read, no retry is offered and the message says what to change instead. Uploads are rate limited, so a long burst in one sitting can hit a daily limit; that message says so plainly rather than failing silently.

What this is not

This is a reading of one photograph, and it is not an appraisal, an authentication, a laboratory report or a certificate. It quotes no price, no value range and no insurance figure. It asserts no purity, no gemstone species and no maker as fact, and it will not place a piece in a particular year — style and construction give a broad range at best. Composition, authenticity and value all need the piece in hand.

That restraint is the product. The useful output is not a verdict but a short list of the right next questions, so the money you spend on a professional opinion goes on the question that actually needs one.

Questions

Is the free preview a shorter version of the full report?

No. It is a different part of the same report. The preview gives you the identification section in full, the photo-quality judgement, the limits of this particular reading and the list of what a photograph cannot settle. The credit unlocks the analytical half: the material clues, the markings, the style and period reading, the ranked identification, the alternatives that were set aside, and the checks to do next.

Does unlocking run the analysis again?

No. The whole report is written in one pass and the preview simply holds part of it back. Unlocking reveals what was already written, on the same page, without a second analysis and without re-uploading the photo.

What does the confidence score mean?

It is how strongly the visible evidence supports the identification, shown as a percentage. It is not a probability of being right, not a grade for the photo and not a guarantee. Read it next to the limits of the reading, which name what could not be judged from the picture.

How long does my report link stay open?

On this deployment a report that has not been unlocked is kept for about three days, and an unlocked report is kept for a year. The link is private to whoever holds it and report pages are excluded from search engines. Keep the receipt email, because that is how you reopen the report.

Can I get a valuation or a certificate?

Not here. The report states no price, no value range, no purity and no guarantee of authenticity, and it cannot place a piece in a specific year. What it does is tell you which question to take to a jeweller, a gemmologist, an assay office or an independent appraiser, and what to show them.