Methodology
What decides the order programs show up in the feed.
Every program is scored on seven things, and the feed shows the highest scores first. Open a row to see how that part is measured.
disclosureHow much the program says about who built it30%
Six things, and it scores the fraction it has:
- a recovered name
- a public source repo
- a website or social link
- a published IDL
- a security.txt in the binary
- a verified build
It carries the most weight because it is the sharpest bot filter available — six things a sniper never bothers with.
noveltyHow unlike anything already on record the code is20%
Every new program is compared against everything on record using a fuzzy fingerprint of the compiled bytecode, which survives recompiling where a plain hash would not. The distance to its closest relative puts it in one of three groups, and the radar shows those as separate lists:
- novel — distance 150 or more. Nothing on record looks like it.
- variant — 50 to 149. Loosely similar to something, not a copy.
- clone — under 50. A fork or near-identical redeploy. Scores zero here.
The match is not always trustworthy, and the dossier says so when it isn't — a nearest relative found only because the two are a similar size, or one of dozens equally close, is flagged rather than presented as fact.
primitivesHow rare the syscalls it imports are15%
A program scores on its peak tier — the single rarest syscall it imports. Not a sum, so twelve ordinary imports never beat one rare one. Legendary is worth +0.09 of the final score, epic +0.06, uncommon +0.015; common earns nothing.
Two rules are held below what their rarity would earn. Memory operations and basic logging never distinguish a program whatever their share. And sol_alloc_free_ and sol_get_fees_sysvar are as rare as anything legendary but worth nothing — rare because deprecated, not because hard.
Shares measured over 3,497 mainnet programs on 2026-08-05, and they drift as the corpus grows · syscall list
newnessHow recently it deployed10%
Decays fast, on purpose: a program keeps about a third of this a day after deploying and almost none of it after a week. Fresh deploys surface, then have to earn the place on something other than the hour they landed.
momentumTransactions in the last 24 hours10%
Log-scaled and capped at 10,000, so a giant cannot bury an emerging program on traffic alone.
convictionSOL locked as rent by the deploy10%
Log-scaled, capped at 100 SOL. A large program costs real money to put on chain and keep there — a cost a throwaway deploy does not pay.
adoptionHow much of the last 48 hours saw any activity5%
Counted in hourly buckets, so steady use scores and a single spike does not.
What pushes a program down
Applied after scoring and multiplied together. This is what sinks the churn to the bottom of the feed.