Gauntlet

The removal test leaves nothing standing. The firm was founded in 2018 as a simulation business and every product since has been an application of the same engine: agent based simulation run across each market a vault touches, financial simulation engines conducting real time stress testing, and automated parameter updates derived from the model output. Curation without the simulation would be a person reading yield rates, which is precisely the alternative the firm positions against.

Aera, the white label infrastructure, is sold explicitly as vaults backed by the risk engine, so even the infrastructure product is a wrapper around the models. The modelling is not a feature of the business, it is the business.

The best described control architecture found in this pocket, and it is notable that the controls are model derived rather than human. The published curation methodology names each constraint and its position in the execution path: allocation sizes are capped at what the simulations indicate a market can safely absorb, continuous liquidity checks confirm that decentralised exchanges could support potential liquidations without significant market disruption before exposure is taken, utilisation is monitored to prevent over leverage, and strategies are separated into stated risk tiers with each tranche held to its own risk target as conditions move.

Those are binding limits on autonomous action, published rather than asserted, which is a legitimate oversight design and better than most human gates. The limitation worth recording is what sits outside it: rebalancing runs continuously on capital with no described human override, no circuit breaker for anomalous or stale inputs, and no stated procedure for when simulation output and market conditions diverge.

The most complete process disclosure in this pocket and still not evidence about the models. The published VaultBook sets out curation frameworks, risk assessment processes, vault structures and the specific actions taken as curator across three venues.

A separate methodology paper names the working components: liquidity sufficiency checking before exposure, yield analysis across rate levels and curve shape, utilisation management against over leverage, and reallocation for risk adjusted yield. The application gives allocators live performance and risk metrics, monthly reporting sets out the rationale behind the largest positions, and each vault carries a published risk rating.

An institution can therefore follow the reasoning and watch the positions. What it cannot do is validate the engine: no back test, no calibration evidence, no statement of the simulation's assumptions or known failure modes, no comparison of modelled to realised outcomes, and no artificial intelligence management system certification.

A long track record, hard numbers, and a party with substantial money at stake having done the diligence. The firm modelled risk parameters for Aave, Compound and Maker from 2020, which is a seven year operating history in a market where most participants are younger than their first drawdown.

Current scale is stated consistently across the firm's own material and independent trade press at more than 1.5 billion dollars curated across more than thirty vaults, with 35 billion dollars monitored for partners. Deployments are named and described rather than listed: 500,000 users reached through KAST's earn programme, yield integrated inside the Uniswap app.

SBI Holdings, a listed Japanese financial group, was the sole investor in a 125 million dollar round in July 2026, following a Series B at a one billion dollar valuation led by Ribbit Capital. Independent coverage is broad. The gap is realised performance: no published record of vault returns against modelled expectations, and no disclosure of losses.

Two overlapping positions, and the first is the purest instance of this shape recorded anywhere in the index. The firm assigns an A plus to D risk rating to the vaults it itself curates, so the rater and the rated are one entity, and nothing describes what separates the rating judgement from the commercial interest in the rated product attracting deposits.

The second overlap is older and sharper in its implications: the firm built its reputation setting risk parameters for lending protocols and now runs vaults that allocate capital into markets on those same venues, so a party influencing the rules of a market can also be a participant exposed to it. Neither arrangement is denied and neither is described.

Nothing states what information barrier exists, whether parameter advisory work and curation are staffed separately, or how a conflict between the interest of a protocol being advised and the depositors in a vault would be resolved.

No privacy policy detail, data processing agreement, subprocessor list or retention schedule was located. The exposure is indirect but real: strategies are distributed through partners' consumer surfaces, including a neobank earn programme reaching 500,000 users and yield inside a widely used wallet, so end user data touches the arrangement even though the firm itself allocates rather than onboards.

Nothing describes what the firm receives from those distribution partners, what it retains, or whether allocation and performance data attributable to an institution's own book is segregated from the data informing strategies serving others.

A dedicated security page, a named and specific partner architecture, and no audited attestation anywhere. The published posture identifies each provider and what it does: zeroShadow for incident response, Hypernative for onchain threat prevention and detection, Chainalysis for tracing funds, and the Security Alliance intelligence group, with the scope of what is protected stated in dollar terms.

Naming the stack and its function is more than the silence that is the norm on this axis and it is verifiable by a buyer. Held at this grade because no service organisation control report, information security management system certification, penetration test attestation or trust centre document was located, and because the claim to be the only curator with an institutional grade security posture is a superlative rather than a credential and nothing published substantiates the comparison.

The firm holds no financial licence or registration. The observation worth recording, stated as fact rather than legal conclusion, is that the activity has moved a long way from where it started: selecting markets, sizing positions and rebalancing capital across stated risk tiers on behalf of allocators is, in traditional markets, discretionary investment management and generally authorised as such.

The structural reason the perimeter does not obviously attach is that the vaults are non custodial. The firm also publishes an A plus to D risk rating without any credit rating agency registration. That a listed and regulated Japanese financial group took the entire most recent funding round is a fact about the investor's appetite rather than about the firm's own supervisory standing.

No fairness testing, model bias assessment or concentration analysis is published, and the concentration question is the one that matters at this firm's size. As the largest curator on the venue where much of its capital sits, its models decide which markets receive liquidity across a substantial share of one lending ecosystem, so a systematic preference embedded in the simulation propagates as a funding advantage for the markets it favours and a funding drought for those it does not.

Nothing addresses whether that concentration has been examined, whether the models are tested for correlated behaviour with other curators running similar approaches, or what happens to a market the models consistently decline.

No liability position, loss sharing arrangement, error rate or recourse route is published. The economics run one way and the published material does not acknowledge it: the firm earns a fee for allocating capital it does not custody, while a depositor whose funds were placed by the models into a market that fails bears that loss directly.

Nothing states what the firm is responsible for if a simulation was wrong, if a cap was set too high, or if a rebalance executed into deteriorating conditions. No record of vault losses or drawdowns is published, and no correction or appeal path exists for an allocator disputing a curation decision after the fact.

No model provider, family, version or hosting arrangement is disclosed. The simulation engine appears to be built in house, which would be a meaningful supply chain fact in a market where most vendors assemble third party components, and the firm never states it, so a buyer cannot distinguish proprietary work from undisclosed dependency.

A recent reference to growing operations with artificial intelligence support is left entirely unspecified, including whether any external model provider touches allocation decisions or is confined to internal workflow.

The deepest integration position in this pocket, because the product is embedded inside other companies' products rather than consumed alongside them. Aera is white label vault infrastructure that institutions use to power their own branded yield offerings on the firm's risk engine. Curated yield is built directly into the Uniswap web application and wallet, curated vaults are reachable through Elwood's interface, and KAST's earn programme runs on one of the firm's stablecoin strategies.

Underneath, strategies operate across Morpho, Pendle, Drift and Symbiotic and across Ethereum mainnet, Base, Optimism, Arbitrum and Polygon, with the firm described as Morpho's largest curator. Named security providers are wired into operations alongside the allocation stack.

No hosting regions, residency commitments or tenancy model are published for the off chain infrastructure. The on chain half of the picture is unusually legible by construction, since strategies operate on named public networks and positions are inspectable, and the white label infrastructure is described as composable and customisable without stating what a deploying institution controls. An institution in a jurisdiction with data localisation obligations would find nothing addressing where the simulation infrastructure, the data pipeline or the application actually run.

No fee schedule for curation was located in the firm's published material, which is the most material commercial fact about a business whose revenue is a fee on capital it allocates. The honest qualification is that this information is not truly hidden: vault level fee parameters are set on chain and inspectable per vault by anyone who knows where to look, so the position is better described as undocumented rather than concealed.

It is still an omission, because an allocator comparing curators should be able to read the economics from published material rather than reconstruct them from contract state, and nothing states whether fees vary by strategy, by tier or by institutional relationship.

Breadth of buyer type here is genuine and unusually well evidenced. More than 150 fintechs and institutions are integrated, and the named ones span distinct categories rather than repeating one: the stablecoin neobank KAST, whose earn programme reaches 500,000 users, the decentralised exchange Uniswap, where curated yield sits inside the app and wallet, the institutional trading platform Elwood, the wallet infrastructure provider Privy, the node infrastructure firm Blockdaemon, and the stablecoin native network Tempo.

Scale is stated consistently at more than 1.5 billion dollars curated across more than thirty vaults on three networks, with a further 35 billion dollars monitored for partners. Geographic and currency expansion is under way beyond dollar and euro strategies into peso and yen denominated stablecoins, and a listed Japanese financial group took the entire most recent funding round.