Membership is per vault rather than per protocol. An address is registered against a vault's terms at an allocation window and holds a claim from that epoch. 1.24K member positions are held across the 10 vaults today — from 37 in Genesis to 284 in Anchor.

Yes, and it is not uniform. The minimum is a vault parameter, published with the vault's terms at the window and moved only by governance. A vault whose positions run to nine figures cannot usefully take a four-figure allocation; a vault built for same-epoch exit can.

Yes, and most members do. Three allocations produce three separate claims with three tiers and three lock-ups. Companies overlap across vaults — Anduril sits in four of them — but the claims do not: one vault cannot draw on another. Risk tiers covers what that isolation does and does not protect.

On an epoch boundary and never between them. The next window is Epoch 03, in Q4 2026. Compound opens to allocation for the first time in it. The mechanics of a window are in Allocation.

Fee terms are a vault parameter, published with the vault's terms and moved only by governance vote a full epoch in advance. They are not uniform across the 10: a vault with a 14-month median holding period and a vault that commits capital for 84 months do not cost the same to run, and Reserve — which owns no company — costs least of all.

The PnL figure is. +38.4% since genesis is net, and it is cumulative over less than six months rather than annualised. The 1.38× carrying mark is a different kind of number altogether: it is what the book is marked at, not what has come back. $412M has actually been realised. Marks & reporting sets out how to read each figure without being misled by it.

Principal returns to Reserve, settles at par in the same epoch, and is available to redeem or reallocate at the next window. The gain routes to Compound and is written into mid-stage companies at a maximum of $20.0M per name, unless the member opts out at the window. Opting out is an action taken each time rather than a standing setting.

Out of Reserve, yes — same epoch, at par, no queue and no gate. Out of any other vault, no. The lock-up runs from the entry epoch, there is no early redemption facility, no penalty exit, and no secondary market in which to sell a claim. If there is a real chance of needing the capital before the term, the allocation belongs in Reserve.

It is marked down as the evidence arrives and then written off. It returns nothing. The loss is contained by the vault holding the position and cannot reach any other vault.

Some vaults expect this. Genesis holds six positions and expects two of them to be worth zero — that is the stated base case, not a failure scenario. If you hold Genesis, that loss is yours in full, and the fact that it cannot reach Anchor is no comfort unless you are also in Anchor.

No. There is no cross-margining, no netting between books and no borrowing between vaults. Reserve funds redemptions and follow-ons out of cash it holds; it does not absorb another vault's write-down.

No. There is no leverage at the House level and no margin on a member position. The most a claim can lose is its principal. Two vaults carry hedged or momentum-led exposure — Shield and Momentum — and both are stated in their tier rather than hidden in a footnote.

At the level of a vault, which is the level that applies to a member: a total loss on Keystone's largest holding takes 11.4% off vault equity in one epoch. Momentum's maximum drawdown to date is 9.4% of vault equity. A Frontier position that misses a technical milestone can reprice 60% in a single round. Genesis can write two of six positions to zero.

At the level of the whole book, maximum drawdown is −7.2% — and that figure understates the risk, because marks move on epoch boundaries and only off priced rounds, so they move less often than a public price would.

Per vault, against a published mandate, and the mandates are deliberately narrow. Anchor requires more than $200M of annualised revenue, growth above 40% and an auditor the House will name. Perpetual requires three consecutive up rounds, gross margin above 60% and a market growing faster than the company — twenty-two names have been reviewed for six slots. Momentum enters only when a primary prices at least 60% above the last. Frontier has taken six rounds out of twenty-five reviewed since Epoch 02.

Structure is the clearest statement of the principle: it will pass on a clean primary at the same price every time. If there is no structure, there is no trade.

A committee decides per vault, and the decision is written to chain on the epoch it happens — the entry, the exit and the reason. Keystone carries the strictest version: to add an eleventh company it has to sell one of its ten first, and name which, in public, on the boundary.

Because a late-stage name can be a core holding in one mandate and a conviction slot in another, at different sizes and for different reasons. Each vault holds its own position and marks it on its own terms. The reverse index is built from each vault's published holdings, so a company can never claim a vault the vault does not claim back. The overlap table is in Vaults.

Every figure across this site — $1.94B deployed, the 1.38× carrying mark, $412M realised, the per-vault capital, marks, member counts and lock-ups — is derived from one internally consistent set of vault records. Nothing is typed twice, so nothing contradicts anything else.

Connect carries three addresses: founders raising, people who want to work on it, and institutions. They reach the people who built this, not a desk.