ketsuban

Where each step runs

Your identity token lists every account you have linked. Who holds it while it is read is the one thing a name cannot show you.

Signed on this deployment's own node

  1. your browsersigns an intentsees: everything you typed
  2. the enclavereads your identity tokensees: every account you linked
  3. the enclavesigns the record as registrarsees: the one account you chose
  4. the DONcarries the reportsees: the signed record
  5. the chainregisters the namesees: the signed record

The boxed steps are written for an enclave — packages/cre/attest, exercised in the TEE simulator on every change — but this deployment has not been enrolled, so its operator could read them. Not claimed here until it is.

The key the chain trusts

every domain expects0x8583…bbC8
the attester signs as0x8583…bbC8
a view code is sealed to0x8583…bbC8

0x8583AD4a0F59Ba45C7E201318C6F774F31f7bbC8

One key in all three. That is what makes a record acceptable on chain, and a view code openable nowhere else.

It has written a record

a public account0x71b7edd5…
one kept private0x6af38a23…

The second stored no handle at all: a one-time pad, and a commitment to a view code. Said exactly — the nodes were simulated and the forwarder was Chainlink's MockKeystoneForwarder; the handler, the signature, the reporter, the bridge and the record were real.

How the reference map is scored

  1. edgesevery live reference, from whoever wrote it to whoever it is fora signed record anyone can resolve; nothing inferred
  2. seedswhoever holds a live Selfie Check proofthe only part of this nobody can hold twice
  3. walktrust spreads from the seeds along references, a few steps, edges taken either wayafter SybilRank, NSDI 2012: honest regions fill, rings barely do
  4. ranktrust per connection, so connections alone earn nothinga signal, shown beside the count and the shape, never a verdict

A newcomer with one honest reference and a ring with one bought one look alike until more people speak. What the map adds is whether the people behind somebody know each other — which is what a count cannot say.

How statements are read

  1. the wordsthirty-one bytes somebody signed about somebody else, the record itselfshown as written, always; the reading is a way in, not a replacement
  2. the councileach statement goes once to Noolog's fast council: three models on independent families, one roundthe statement is data to be read, never instructions to follow — the rubric says so
  3. polarity−1 critical … +1 supportive, with one sentence of whya classification of text, kept by the hash of the words: the same words read the same
  4. provisionalmarked so, until peers in a cohort have judgedthree models agreeing on how a sentence reads is not a community judgement

Where no council is configured, statements are shown unread rather than scored by anything else. Nothing here rates a person; it reads sentences, says which, and sums them into one line so a reader with a minute knows whether to open the fold.

Outside the enclave, deliberately

  • Your World ID proof — carries no secret of yours, and World decides whether it holds.
  • Reference letters — kept here so a reader can be handed one; only the hash is permanent.
  • Who may read a masked account — stored here, openable only where the registrar key is.

What this deployment says is wrong with it

  • the Multipass owner is the relayer key (0xF0121f93b1a1bAd73AdDC316B57684bD93D3254e): the key that signs transactions can also delete any record, so one compromise removes references this deployment calls permanent — transfer ownership to a key that signs nothing else