The Reserve

The reserve is the balance sheet behind redemption. This page says what it holds, where it sits, why it holds what it holds, and how you verify every part of that yourself.

What it holds#

The working buffer exists so redemptions can be served immediately, independent of banking days and off-ramps. Its size follows a published rule: 1.5 times the trailing 8-week redemption run rate projected to 90 days, and never more than 20% of the reserve. It is the one part of the reserve you cannot check on chain, which is exactly why it is capped hard and attested monthly. We state that plainly rather than pretend the snapshot is proof.

Net, not gross#

The only number we call backing is net:

net reserve  = (liquid reserve + working buffer)
             - (unclaimed drip + unspent credits)

net backing  = net reserve / circulating supply, quoted per 10M tokens

Unclaimed drip and unspent credits are real liabilities: work the system already owes. A gross reserve figure with liabilities listed somewhere nearby is how a backing number gets restated later, and a restated backing number is what kills trust in one. The open stats endpoint, /api/stats, returns every input to this calculation separately, so you can recompute the net figure and disagree with our arithmetic if it is wrong. See Verify It Yourself.

Why stablecoins and not prepaid compute#

Prepaid compute sounds like the obvious reserve for a compute-redeemable token. It is the wrong one, for reasons that survive checking:

1. Prepaid balances freeze today's provider and today's capability level, and typically expire within 12 months.
2. The nominal price of a given model barely falls; what falls, by roughly 50 to 200 times per year, is the cost of a constant capability level. A dollar held as a dollar buys substantially more capable tokens next year. A dollar locked into a 2026 model buys 2026 capability forever.
3. Stablecoins keep provider optionality: price changes, quality changes, or terms-of-service changes at one provider do not strand the reserve.
4. The prepaid discounts that could compensate for any of this start at roughly $30,000 to $50,000 per month of spend (UNVERIFIED, secondary sources), far beyond this system's early scale.

Why the reserve earns no yield, on purpose#

This is a documented policy reversal, and we would rather show the work than hide it. The original design allocated up to 50% of the reserve to a curated USDC lending vault at an advertised ~7%. That plan died in two steps, both verifiable:

1. The advertised 7% did not survive an on-chain check. The relevant vaults on Robinhood Chain return maxDeposit() = 0 for every external address we tested: an outside Safe cannot deposit there at all today. Of the advertised rate, roughly 3.5 percentage points are a Robinhood-funded promotional subsidy, apparently tied to their own app; the native rate was 1.6% to 4.3% and falling.
2. The remaining upside does not pay for the tail risk. At the reserve sizes of the first half-year (tens of thousands of dollars), a ~4% position would earn roughly $1,800 per year. Against that stands a failure mode that actually happened, repeatedly, in the ecosystems we evaluated: the Stream Finance collapse (November 2025, $285M to $700M bad debt across the Morpho and Euler ecosystems) and the Aave Kelp/rsETH bridge exploit (April 2026, $196M to $290M). For a reserve whose entire function is to be a credibility anchor, one small incident costs more than $1,800 buys.

So the reserve is plain USDC, indefinitely. A future yield position has published preconditions, all cumulative: a reserve large enough that ~4% is operationally meaningful (low-to-mid six figures), diversification across at least two venues with no majority in one, a fresh maxDeposit check immediately before any allocation, and a collateral review per vault. Until all of them hold, there is no position, and nothing here should be read as a plan to open one.

What this leaves as residual risk on the reserve itself: USDC issuer risk and custody compromise. Both are listed in Risks, not mitigated away.

Custody#

• Two separate Gnosis Safes on the reserve path: one on Robinhood Chain (receives the reserve's two-thirds share directly from the immutable FeeRouter; nothing else can be routed to it and no person signs for the inflow), one on Base (holds the USDC reserve). Both 2-of-3 multisigs with three separate people on separate hardware devices.
• The team treasury that receives the other third is a separate address and is never counted in any reserve or backing figure. Fee income becomes "reserve" only after the contract-enforced split, not before.
• A 2-of-3 with three real signers, stated honestly, beats a 3-of-5 where one person holds three keys. An upgrade path to 3-of-5 is documented.
• Policy layer, published: 48-hour delay on any transfer above $10,000 equivalent.
• All addresses are published once deployed. Nothing is deployed yet; see Verify It Yourself for what will appear and where.

How the reserve can shrink, and how it cannot#

The absolute reserve shrinks when redemptions are served. That is the system working: outflows are capped at roughly 1% of the net reserve per day, every redemption burns more supply than it draws value, and the remaining holders end up with higher backing per token, not lower. Watch only the absolute reserve curve and you can misread a served claim as a leak, which is why the dashboard shows the absolute reserve and the per-token backing side by side on the reserve page.

Net backing per token cannot be reduced by any protocol operation. It can be reduced by exactly two things: losses on reserve assets (an issuer failure, a depeg, a bridge) and operational failure (custody compromise, accounting error). The yield-free policy and the custody setup above exist to keep those two channels as narrow as we can make them. They do not make them zero.