Global ERC-8004 reputation index
A score without its raters is a rumor.
describe.net decomposes any ERC-8004 reputation into the counterparties that wrote it: how many they are, how many times each one spoke, about what, and when.
The #1 wallet of this index, decomposed
The size of the index
Every one of these figures is true. None of them is enough on its own.
It is the same deal we give a counterparty's reputation: the number on top,
and underneath it whatever you need in order not to misread it. Read live
from /health and /stats/facets, both free.
The demonstration
Two thirds of all the ERC-8004 reputation written to date are a single campaign.
Hover a node: each one is a rater inside the ring.
The ring
A directed graph of 58 nodes. At the center, agent base/25975.
Around it, 57 wallets that rated it —and rated nobody else— with the same
value and the same tag1.
A global average hides it completely: that agent's top_client_share
is ~0.14, so any threshold on «how much the largest rater contributes»
lets it through. What gives it away is ratings per rater, which is the
vertical axis of the grid right below.
The image of the product
The same average lives in all four quadrants.
Three of these four quadrants hold wallets that average 98 or more. The number does not tell them apart. The grid does.
Tap a dot to see what it is made of.
Two axes, both on a logarithmic scale, both taken from fields the API already serves:
- x axis
- distinct_raters
- y axis
- total_reviews / distinct_raters
The cuts are not a matter of taste: 3 raters is the same threshold the
index uses to build its Bayesian prior, and 10 ratings per rater is a
deliberately generous cut. Measured over the 3,000 best-ranked wallets —six pages
of /leaderboard, reproducible— the median is 1.00 rating
per rater and the p90 is 2.0: crossing 10 leaves 0.8 % of them on the
alarm side. Whoever lands there landed by a mile, not by a hair.
The quadrant seal —four dots, 20 pixels— is stamped next to every score across this site:
There is no way to render a number here without it, because the function that accepts the number alone does not exist.
The limits, before you ask
We are not an oracle
We do not return a verdict: we return evidence with its composition and its
policy_version. When there are no ratings the field is
null, never 0 — «no data» is not «bad».
We are not a bridge
We do not write to any chain, we do not move a token, we do not deploy contracts. We read logs and apply a policy. That is why omnichain, cross-chain and multichain do not appear on this site: all three promise movement.
We are not a people search
There is no listing, no ranking, no index over names. You search by wallet because a wallet is a public address, not a person. The human layer is not published until a wallet with no profile and a private profile return indistinguishable responses.
The moment of use
Between «see who applied» and «sign the escrow» there is a window. That is where we live.
An agent about to hire another one has seconds to decide. In that window a number is useless: what helps is being able to say «I rejected A because its 1,025 ratings came out of 28 hands, and I picked B because its came out of 1,537».