I have a way to manually identify yourself as a decoy but I haven't automated that yet. What I want to do is use known "view keys" to identify decoys, because if you have the view key you know the "true" tx in which you spent your own coin, do if you've been used as a decoy you can automatically filter that out.
Right now it uses two heuristics: merge analysis and recency bias. If you received coins in two very close blocks and then spend them together, your ring, which should contain keys from random blocks, will have two suspiciously close blocks where you held both inputs. So you can identify those as the real spender's coins.
Recency bias takes advantage of the fact that most decoy selection algorithms are biased toward selecting keys from recently created utxos. Ones that are significantly older stick out like a sore thumb.
Other metrics I want to add include bot detection (which is based on identifying txs with many outputs, since these tend to be created by automated software rather than a real person) and taint tree analysis (which creates trees of possible senders fanning out backward in time from a known destination).
Super Testnet
npub1yx…c399s
2024-08-04 19:21:10
in reply to nevent1q…8fzw
Author Public Key
npub1yxp7j36cfqws7yj0hkfu2mx25308u4zua6ud22zglxp98ayhh96s8c399sPublished at
2024-08-04 19:21:10Event JSON
{
"id": "e9690a70b7d7ac29821511fbfb0f382d789e447860b4f7e8b3eddffce985f6ea",
"pubkey": "2183e94758481d0f124fbd93c56ccaa45e7e545ceeb8d52848f98253f497b975",
"created_at": 1722799270,
"kind": 1,
"tags": [
[
"e",
"e42b104f8d68b307db0e9c97737ca0ce10244cd9fbd2fc09b92b65b26407f34f",
"",
"root"
],
[
"e",
"e27f35db6902aa98dd8e4bedf4a6459fd9d8281263b0b4edf7bfbc86ea2e7a28",
"",
"reply"
],
[
"p",
"2183e94758481d0f124fbd93c56ccaa45e7e545ceeb8d52848f98253f497b975"
],
[
"p",
"b9e76546ba06456ed301d9e52bc49fa48e70a6bf2282be7a1ae72947612023dc"
]
],
"content": "I have a way to manually identify yourself as a decoy but I haven't automated that yet. What I want to do is use known \"view keys\" to identify decoys, because if you have the view key you know the \"true\" tx in which you spent your own coin, do if you've been used as a decoy you can automatically filter that out.\n\nRight now it uses two heuristics: merge analysis and recency bias. If you received coins in two very close blocks and then spend them together, your ring, which should contain keys from random blocks, will have two suspiciously close blocks where you held both inputs. So you can identify those as the real spender's coins.\n\nRecency bias takes advantage of the fact that most decoy selection algorithms are biased toward selecting keys from recently created utxos. Ones that are significantly older stick out like a sore thumb.\n\nOther metrics I want to add include bot detection (which is based on identifying txs with many outputs, since these tend to be created by automated software rather than a real person) and taint tree analysis (which creates trees of possible senders fanning out backward in time from a known destination).",
"sig": "dc3991ed8d3d5ab22cebfd57d5ceae8c6571773afe8fe853572fd594637534518029101043ab82cc1c513f62c7bf942e6e6dac45cf9239068b98343e289c16e2"
}