Who do you trust?
Your trusted set is the validators you'd hand a transaction in the clear. Everyone else is untrusted. Every number on this page follows your choice.
TL;DR
Trusted set: . Collusion: . threshold, k = proposers per window, the 7 hubs from the NewCon talk.
1Ideally: one trusted proposer per window
In NewCon, time is cut into windows of 0.8 seconds (16 slots of 50 ms), so 75 a minute. Each window belongs to one hub, a region like Frankfurt or Tokyo, and gets proposers: validators from that hub, picked at random with odds that grow with stake (up to a cap). Each proposer can put transactions into that window's block. An app knows in advance which validators propose in each window.
The cleanest way to avoid leaks: send it, whole, to one proposer you trust in the next window, and to nobody else. Nobody you don't trust sees it before it's in a block.
2Choosing who to trust is the hard part
"Trusted" should mean something you can check, and every option trades coverage against how much you have to take on faith.
- Client networks. Most validators announce their software in gossip as a client ID (Agave, Firedancer, Jito-Solana, Rakurai, Harmonic, BAM…). It's self-reported, and nothing in the protocol stops an operator from forwarding or selling the flow it receives. Some networks forward flow by design: Harmonic validators to Harmonic's builders, GBX-connected validators to every GBX builder. Trusting one of those means trusting the whole network.
- Verifiable hardware. BAM runs its scheduler inside a TEE (trusted execution environment) that publishes a signed attestation of the code it runs. You can check it before sending, so the trust moves from each operator to the TEE hardware and BAM's code and operation.
- Specific validators. Operators you know, or the largest ones. Their identity is public but their behaviour isn't verifiable. A big validator is in more windows, though NewCon caps how much extra stake helps.
| If you trust… | Share of stake | Validators | Windows with one | How you'd check them |
|---|
Same hubs and windows for every row. "Windows with one" is the share of all windows with at least one such proposer. "Use" makes that row your trusted set.
3Trusted routing only
If the app sends only when someone in your trusted set () is one of the window's proposers, it simply waits through the other windows.
4No trusted proposer? Threshold encryption limits leakage
To avoid waiting, the app can still send in a window with nobody it trusts by using threshold encryption. It splits the encrypted transaction into n fragments, one per proposer, and any t of them are enough to decrypt it. With , no single proposer can read the transaction. Proposers only learn it together, once it's committed to the block.
5Collusion is what causes leaks
Splitting only helps if the proposers holding fragments don't pool them. Some validators structurally share their order flow: Harmonic validators forward to Harmonic's builders, and GBX-connected validators stream their flow to every GBX builder (Titan and others). If t of a window's proposers answer to the same party, that party can decrypt the transaction before it's in a block: a leak. How often that happens depends entirely on who you assume colludes. Validators in your trusted set never count as colluders.
Bars show the share of transactions that leak if the app sends in the next window: whole to a trusted proposer if one is there, otherwise as a split across the window's proposers. "Grouped by hub" is the same as one cartel here, because every window's proposers come from one hub.
6How low can leakage go?
The app knows each window's proposers and which ones it treats as colluding. That allows a stricter rule: send only in safe windows. A window is safe if it has a trusted proposer, or if the fragments can be placed so that no colluding group gets t of them. Skip every other window. Under the assumption you picked above (), that gives:
| Strategy | Windows used | Avg wait | p99 wait | Leak per tx |
|---|
The zero is only as good as the labels behind it. It assumes client IDs are honest, that no other validators secretly share flow (GBX was invisible in gossip and only showed up through Titan's API), and that fragments really reach the chosen proposers with no relay in between.
Advanced settings
All strategies
| Strategy | Windows it submits in | Mean wait | p99 wait | P(untrusted party sees plaintext before inclusion), per tx | How it sends |
|---|
Individually trusted validators
Exact under Alg. 3: a validator is one of its hub's k proposers with probability min(stake, c)/c, where c is the hub's stake cap.
Per hub
P(trusted) is the share of the hub's windows with at least one trusted proposer. The thin bar is the trusted share of hub stake. Drag a slider to override a group's or client's stake share inside the hub (the rest renormalize).