Hopium Machines, ranked properly — all 6,969 of them

The short version

How the ranking was built

Method · and why it disagrees with the marketplace

This ranking scores the whole collection — all 6,969 machines — and only then asks which of them happen to be for sale. Ranking the 612 listed machines against each other would be meaningless: rarity is a property of the collection, not of the shop window.

1. Read the collection from its own contract, not from a marketplace. tokenURI() on 0x7da15c761409cb921a81f0e003704cff418b700b returns an IPFS path. All 6,969 metadata records were pulled from that directory (QmeqvTnMhd4DM5P41aQcE5oCm96UAabCZ436E5EL896ybL). Nothing here depends on an indexer having finished its job.
2. Cross-check the counts against a second source. Every trait count computed from IPFS matches the count OpenSea displays on its own item pages, to the unit. Two independent sources agree, so the data is not in dispute — the disagreement that follows is purely about scoring.
3. Drop the traits that carry no information. A trait every token shares tells you nothing, and a trait that merely restates another tells you the same thing twice. Both are identified below, with the evidence.
4. Score by information content. Each trait earns a machine -log2(share) bits; the score is the sum. A 1-of-6,969 trait is worth 12.8 bits, a trait half the collection has is worth 1. This is the same measure OpenRarity uses — the only change is which traits go into it.

Finding 1 — Promise Rate is not a trait, it is Hat restated

The one that moves the ranks

Every machine lists a "Promise Rate". It looks like an independent attribute with five values and a nicely skewed distribution. It is not. Across all 6,969 machines, the Hat determines it completely:

There is exactly one exception in the whole collection, and it is the one-of-one: machine #4,663 has no hat and prints 0 promises instead of 2, because it is switched off. Every other bare head prints 2, every Beanie, Party Hat, Crown and Propeller prints 3, every Halo 5, every Raspberry 7. Promise Rate adds no information that Hat did not already carry.

Scoring it anyway counts the hat twice, and that is not a rounding error:

Verified against OpenSea's own published numbers

Rather than assert that the marketplace scores all seven traits, that model was built and then checked against the rank OpenSea actually displays, on seven machines picked from the top, the middle and the distorted part of the collection:

Six exact, and one off by a single place — #3,693, which shares an identical score with #1,088; the two are a tie and the order inside a tie is arbitrary, so their two ranks can be handed out either way round. 19.5% of the collection sits in a tied pair like that (no tie group is larger than two), which means any rank here can differ from a published one by one place for that reason alone. It changes nothing that follows: the distortions this report is about are thousands of places, not one. The middle column is a model of the marketplace's scoring rather than a reading of it — landing on the published number six times out of seven, and once beside it by a tie, is what makes the right-hand column a fair comparison rather than an opinion.

Finding 2 — "Exit Liquidity: You" is on every single machine

All 6,969 carry it. It is a good joke and it is zero information: a trait everyone has cannot distinguish anyone. It contributes exactly 0 bits to every machine, so it cannot change the order — but it does pad the trait count from five real attributes to seven, which is how a machine with nothing remarkable about it still reads as "7 traits" on a listing page.

Finding 3 — machine #4,663, and the chain it lives on

The best-hidden thing in the collection

Robinhood Chain's chain ID is 4663. The one machine in this collection that came off the line switched off is #4,663. That is not a coincidence anybody stumbled into — it is the signature on the work.

Every machine you can buy, ranked by true rarity

612 active listings · snapshot 18 August 2026, 05:40 UTC

This is the table the report was written for. Ranks are positions in the whole 6,969-machine collection, not positions among the listings — the machine at the top of this table is the 5th rarest of 6,969, and it happens to be for sale. Only 27 of the 100 rarest machines in the collection are listed at all, which is the single most useful number here: the top of this collection is held, not traded.