Systematic liquidity provision against reference-price dislocations in tokenized equities

1 · The market microstructure premise

Tokenized equities on Robinhood Chain trade across a fragmented set of venues: a Liquidity Book DEX quoted in USDG and several thousand Uniswap v4 pools spanning arbitrary fee tiers. Fragmentation plus thin depth produces a recurring phenomenon: a single aggressive order in a shallow pool moves its marginal price far from the value of the underlying share: a dislocation. With no resident market maker, it persists for minutes to days.

A dislocated pool is not an anomaly to be lamented; it is an inventory-constrained profit opportunity. Whoever already holds the asset when the dislocation appears is the counterparty of record. mm's entire design reduces to one sentence: be the standing inventory in every instrument where dislocations occur, and be the fastest disciplined seller when they do.

2 · Reference pricing

For each instrument i the desk maintains a reference price P̂ᵢ(t) from the consolidated primary-market feed, refreshed on a sub-minute cycle, alongside a live ETH/USD conversion for ETH-quoted venues. For a pool p holding instrument i, the instantaneous deviation is

Outside the primary session P̂ is stale by construction; the desk widens every actionable threshold by a session premium θ rather than pretending the reference is live. Deviations in the parking band (below −95% or above +5,000%) are classified as structural artifacts (pools initialized at boundary prices) and discarded before they consume analysis.

3 · Fillability: the anti-mirage test

Displayed price is not evidence. Thousands of pools advertise spectacular deviations against which no trade can execute. The desk therefore admits a dislocation only after a fillability probe: a quoted liquidation of a fixed notional q₀ (at reference) through the pool's own execution engine, fees and impact included. Defining the probe yield

a pool qualifies as a real dislocation iff δₚ ≥ δ* and Φₚ ≥ φ_min. The economic logic is exact: a genuine pump deposits the aggressor's own quote asset into the pool, so a genuine pump is always fillable; a mirage quotes nothing. Probe results are cached per pool and invalidated on ±2% deviation movement, so the fleet of static artifacts is paid for once, ever.

4 · The liquidity survey and the eligible set

Daily, every instrument is classified by its impact function, the premium over reference paid by simulated aggressions of increasing notional:

Instruments where I(10⁴) ≈ 0 are deep: dislocations there are arbitraged away before inventory can monetize, so they are excluded from the program entirely. Instruments where I(10³) exceeds the thinness threshold are the eligible set: markets where a four-figure order visibly moves price, which is precisely where five-figure dislocations are born. Tokenized ETFs are excluded categorically, as are operator-blocked symbols; the operator may also pin instruments into the set on discretionary information. Classification is re-estimated every 24 hours because thinness is a state, not a property.

5 · Inventory construction

Capital enters as protocol fee flow. Deployment is a capacity-weighted round robin: given deployable budget C and a minimum viable clip c_min, a pass executes

rotating through the eligible set with a persistent cursor, so every instrument receives inventory before any receives twice. Each purchase routes through the best of all venues for that instrument and executes only at P_eff ≤ P̂ · (1 + α). The desk never pays a premium to acquire what it intends to sell at one. Entries run on a fixed cadence (T+5m, T+15m, then every 30 minutes) so fee flow converts to standing inventory within the hour it arrives.

6 · Execution against a dislocation

When a held instrument prints a real dislocation, the desk sizes the maximum liquidation S* whose effective price, after pool fee and self-impact as quoted by the venue itself, clears the floor:

where B̄ is the position's average cost basis. The desk sells strength; it does not realize losses into noise. Each order additionally requires an absolute profit increment π ≥ π_min over max(reference value, basis), carries a slippage-bounded minimum output so a moved market reverts rather than fills badly, and settles against the wallet delta: the quote predicts, the balance decides.

Tranche schedule

Exits are distributed, not slammed: at most a fraction τ of S* per order, one order per instrument per cooldown window Δt. A persistent dislocation is harvested across hours, each tranche re-sized against the pool's remaining depth, so extraction decelerates as the pool drains and the market is left visibly bid. A fading dislocation simply stops qualifying, and the remaining inventory is kept for the next event.

7 · Profit accounting and the daily distribution

Profit is a ledger identity, not an estimate. Every liquidation realizes proceeds minus the average cost basis of the size sold; lifetime realized profit Π(t) accumulates monotonically. With D(t) the cumulative amount already distributed, holders are owed

This balance carries forward and never resets. Every 24 hours the desk pays min(O, cash) in USDG, pro-rata over an eligibility-filtered holder snapshot: for holder h with balance bₕ,

The snapshot is reconstructed from the token's complete transfer history maintained locally, never an indexer, and spot-verified against chain state before any value moves. AMM reserves, protocol machinery and desk addresses are excluded from the eligible supply; sub-dust allocations remain in the pot.

8 · Custody and the execution agent

All capital (fee inflow, cash, inventory) is custodied by a single on-chain fund contract, publicly auditable in real time. An autonomous execution agent (the keeper) runs the entire operation against that contract: it surveys venues, routes orders, and triggers the daily distribution, while the contract constrains every action it can take. The agent itself holds nothing; every balance lives at the fund address. Properties, stated as held:

9 · Operating parameters

Parameters are operating policy, not physical constants; the desk publishes them so its behavior is predictable to the market it serves.