EXP-01 / Ionize / Verifier nodes and bonds

Verifiers

To ionize a track is to run the inference again and watch whether the same condensation forms. Nodes do that work, and none of them is trusted to do it honestly -- each one is bonded, so being wrong costs it. This is the page where $KORTX does its job: it is the collateral behind every claim the rest of this site makes, and the currency its fees settle in.

Where the bonds sit

Total bonded

66.29MKORTX

Capital at risk across every role. Not revenue, not a reserve.

Behind plates

2MKORTX

Providers' stake against their own model claims.

Behind nodes

64.29MKORTX

The checking side of the network.

Slashed to date

0KORTX

Taken from a bond after a ruling. This figure going up is the system working.

Provider plates

2,000,000KORTX3.0%

Staked by providers behind a registered model fingerprint. This is what an upheld challenge reaches.

Verifier nodes

64,285,710KORTX97.0%

Staked by the nodes that rerun inferences. It is what makes a careless rerun expensive to submit.

Open challenges

0KORTX0.0%

Locked by challengers in incidents that have not been ruled on. It moves the moment a verdict lands.

The split is the point. A network where almost everything sits behind providers and almost nothing behind the nodes that check them has a large total and very little checking, and the total alone cannot tell you which one you are looking at. The bars above are read from the same accounts the incident pages rule on.

Reading the program config from the cluster...

What $KORTX is for

A provider stakes behind a model

register_platecommit_trace

SecuresThe plate: a model id, a weights hash, a version and the sampling settings a rerun has to reproduce. Traces can only be committed against a plate that carries a bond.

Slashed whenA challenge against one of that plate's traces is upheld -- the rerun did not reproduce the committed output hash.

A node stakes for the right to rerun

register_verifierverifier_stakeverifier_unstake

SecuresThe right to submit rerun results that rulings are decided on, weighted by the bond behind each one. Without a bond, a node's rerun is an assertion; with one, it is a position. Registration can be curated by a switch on the program config; while it is on, the protocol authority co-signs every registration.

Slashed whenThe node's submitted rerun is contradicted by the ruling it fed. Stake is withdrawn through verifier_unstake, which is why it cannot be pulled the moment a challenge turns against it.

A challenger stakes to force a rerun

open_incidentsubmit_rerun

SecuresOne challenge against one trace, and the deadline the rerun has to land inside. This is what makes the incident log cost something to fill.

Slashed whenThe rerun reproduces the committed output and the challenge is rejected. The bond is forfeit -- being wrong in public is not free on either side.

Where the rules areThe minimum bonds, the commit and reveal windows, the withdrawal cooldown and the share slashed per lost vote are parameters on the program config account. The verifier panel below reads that account from the chain when it loads and prints them under "Terms, read from the program". The figures above come from the indexer and show what is bonded now, not what is required.

DenominationEvery amount on this page is $KORTX, the SPL token the program takes bonds in, as the base-unit integer recorded in the program accounts. The indexer reads the token's decimals from its mint account on chain and the figures are converted with that value; until it has been read, amounts print as a dash rather than a guess. Nothing here is converted to SOL or USD. No price, market capitalisation or yield appears anywhere on this site, and none is implied here.

Contract address

ProgramSolana mainnet

Explorer

Verifier nodes

ScopeAll Active only

Registered nodes

Every node the indexer has read.

Active

Currently accepting rerun work.

Bonded by nodes

64.29MKORTX

Stake held by the nodes listed below.

Slashed from nodes

0KORTX

Taken after a ruling went against a submitted rerun.

Reruns completed

Inferences re-executed and submitted for comparison.

Mismatches found

Reruns that disagreed with the committed output hash.

Network mismatch rate

No rerun has been completed, so there is no rate to state.

Idle nodes

Bonded and have never submitted a rerun. Stake without evidence.

These aggregates cover the 5 nodes listed under the current filter, of which 0 have submitted at least one rerun. Switching the filter on drops inactive nodes from these totals as well as from the table, so the two never describe different populations.

Node	Authority	Bond	Mismatch rate	Slashed	Status	Registered
4fcRun...cJncfd	HSA7d6...9sUaay	12,857,142KORTX	No reruns yet	0KORTX	ActiveCurated	2026-10-06
B524mz...z8pvkY	5Z43ao...gaYQMQ	12,857,142KORTX	No reruns yet	0KORTX	ActiveCurated	2026-10-06
EiJRYC...sksbnp	BchWFZ...Dv1SLf	12,857,142KORTX	No reruns yet	0KORTX	ActiveCurated	2026-10-06
Hr6bzJ...uX9kvD	E6b7gJ...M8CPY7	12,857,142KORTX	No reruns yet	0KORTX	ActiveCurated	2026-10-06
HUxnsa...HaVgdg	J7zTrY...cxpSuu	12,857,142KORTX	No reruns yet	0KORTX	ActiveCurated	2026-10-06

Mismatch rateA node that has never rerun anything reads "No reruns yet", not 0.00%. A clean rate and an absent one are opposite findings, and a node with no evidence behind it is the one you should trust least, not most.

Mismatches foundThis counts disagreements a node reported, not mistakes it made. A high number means the node caught things; what costs a node its stake is a rerun the ruling later contradicts, and that shows up in the slashed column.

Indexer ok / read through slot 453,936,211 / 2026-10-06 15:02 UTC