How This Site Verifies (and Doesn't)

Almost every number on this site carries two words: "as reported." That phrase is doing real work, and a site that asks you to trust receipts should say exactly what it means. This page is our receipts methodology, the house rules behind the directory of builds made with Jev, the decision model from TypeSafe AI. We are not affiliated with TypeSafe AI.

Short version of the "as reported" meaning: the number came from the builder, in public, at a link we give you. We copied it carefully. We did not re-run it.

What "as reported" means here

Every build page has a receipts panel: author, source link, reported cost, reported latency, category, source type and the date it was filed. When a field is empty, it shows a dash rather than a guess.

The source link is the receipt. It points to where the author made the claim: an X post, a GitHub repository, a video, a write-up. The submission form says it plainly: "The X post, repo, video or site. This is the receipt." If you want to check us, one click takes you to the original.

"As reported" therefore means three things at once. The figure is the author's claim, not ours. It is quoted in the author's own unit (per run, per item, per hour). And any caveat the author gave travels with it.

What we actually do

Keep the author's caveats. When David Madurzak says he paused Ms. Pac-Man between moves and removed the pauses from the footage, the entry says so. When Hassan says the fraud detection video isn't sped up, the entry records that as his statement.

Flag what we haven't checked. The SuperX post scoring entry says, in so many words, that we have not checked its accuracy figures. That sentence is the house style, not an exception.

Keep units honest. "$0.03 per 1,000 posts" and "$0.13 for 3,282 posts" are different kinds of claim. We don't normalize them into one column that pretends they're comparable.

Refuse to invent benchmarks. No official Jev benchmarks exist. We never add up builder numbers into a leaderboard, and we say so on the accuracy page.

Sort by hand. The catalog is hand-sorted. Of the 551 entries at the time of writing, 132 carry at least one reported number and 45 report a cost. The rest are listed because the build itself is the evidence: the repo exists, the demo runs.

What we don't do

Being clear about this matters more than anything above.

• We don't re-run builds or reproduce their numbers.
• We don't see anyone's invoices, so reported costs are not audited.
• We don't validate accuracy claims against held-out data.
• We don't speak for TypeSafe AI. Pricing, limits and API details come from docs.typesafe.ai, which outranks us on anything official.
• We don't claim the numbers generalize. A receipt tells you what happened once, for one builder, on one workload.

How the site is funded is disclosed on the sponsors page.

How to read a receipt

Four quick checks before you quote a number from here:

1. Unit. Total, per item or per hour? The cost ladder mixes all three and says which is which.
2. Source type. A repository you can run is stronger than a post with a screenshot. Both are listed; weigh them differently.
3. Caveats. Read the entry note. Authors are often more candid than their headline.
4. Reproduction. The only number that counts for your project is the one you measure. The accuracy page shows how to do it in about an hour with your own labeled cases.

Our overall judgment on the model, built from these receipts, is in the Jev review. This page is about the plumbing, not the verdict.

Frequently asked questions

What does "as reported" mean on shipwithjev?

It means the number is the builder's own public claim, copied from the source linked on the entry, in the builder's units and with their caveats. We did not re-run or audit it.

Does shipwithjev verify build numbers?

No. Entries are transcribed by hand from the linked source, and we flag figures we haven't checked; we don't reproduce builds or audit costs.

Why trust unverified numbers at all?

Because they're public, linked and convergent: many unrelated builders reporting similar cost shapes is weak evidence that still beats marketing. Treat them as leads to test, as the review does.

How do I add a build with receipts?

Submit it with the source link and any numbers you measured. Real numbers move a submission up the queue.