We measured Robinhood Chain before we built on it

Robinhood put real US stocks onchain as ERC-20 tokens — NVDA, AAPL, SPY, 96 of them — with live Chainlink prices, on its own Arbitrum Orbit chain. We are building Basquets on top of them: an open protocol where anyone can create a basket of stock tokens, invest in one with a single transaction, and automate it with a recurring savings plan.

Before writing a line of Solidity, we spent time answering three questions the documentation does not: where does the liquidity actually live, how big is this market really, and can you trust the prices you can most easily get? Everything below is measured directly from chain 4663 on 21 July 2026, against the public RPC (https://rpc.mainnet.chain.robinhood.com) and the public Blockscout API. All of it is reproducible, and every finding shaped a design decision in the protocol. This article replaces our earlier separate notes and is the single reference for both.

Finding 1 — the tokens live in Uniswap, not RFQ#

Robinhood’s developer documentation says tokenized stocks “trade via RFQ at launch.” That is true, and it is also not where the tokens are.

Uniswap v4 keeps every pool’s tokens in one singleton contract — PoolManager at 0x8366a39CC670B4001A1121B8F6A443A643e40951, verified on Robinhood Chain — so “how much of this stock token sits in Uniswap” is a single balanceOf read. We ran it for all 96 registry assets, alongside totalSupply():

94 of the 96 stock tokens hold a non-zero balance in the Uniswap v4 PoolManager. Not a handful of blue chips — effectively the entire registry. The concentrated names are striking:

Token	In Uniswap v4	Total supply	Share of supply
NU	17.27	19.96	86.5%
NFLX	395.38	491.94	80.4%
RIVN	13.60	18.41	73.9%
NVDA	6,267.50	13,963.61	44.9%
F	21.81	49.34	44.2%
SPCX	3,408.82	8,313.42	41.0%
MSFT	361.85	1,732.82	20.9%
TSLA	520.57	3,004.27	17.3%
AAPL	373.56	2,625.54	14.2%

Nearly half of every NVDA token in existence sits in one contract, with roughly $8.3M of USDG on the other side of those pools.

This matters more than it sounds. RFQ is a fine mechanism, but RFQ liquidity is not composable: you cannot route through it, and another protocol cannot build on a quote. An AMM pool is a public object — its state is readable by any contract, its price derivable by anyone, its depth a number rather than a relationship. The fact that stock token liquidity landed overwhelmingly in Uniswap pools is the difference between tokenized equities being tradeable and being composable. It is why an onchain index protocol is possible on this chain at all.

Finding 2 — the market is real, early, and brutally concentrated#

The same sweep gives the market’s true shape, and it is worth stating plainly.

Onchain market capitalisation across all 96 tokens is roughly $17M. That is the entire tokenized float of every Robinhood stock token combined — less than a rounding error against the companies they track.

24h volume is about $6.7M, and heavily concentrated. NVDA alone is around 60% of it; NVDA, TSLA and SPCX together are roughly 86% of all trading.

Holder counts, indexed from Transfer events by Blockscout, tell the same story from the other end:

Token	Holders
NVDA	23,624
AAPL	22,562
SPCX	18,782
TSLA	17,043
SGOV	114
TSM	58
BABA	42
NOW	15

Fifteen holders. The long tail is not thin — it is close to empty. If you are buying, depth is the number that matters, not price: on the concentrated names you will trade close to the quote; on a token with 42 holders, the gap between the price you see and the price you get is the whole cost of the trade.

Finding 3 — the convenient price is frequently wrong#

Anything built on these tokens needs prices, and there are two obvious sources. The explorer API returns a price in one HTTP call. A Chainlink feed takes an ABI, an address and a decimals field. The convenient one is tempting, so we checked whether they agree.

We compared both sources at the same moment for every token that has both: 30 tokens in total, since only about a third of the 96-token registry has a Chainlink feed at all. For the oracle we called latestRoundData() on each aggregator; for the explorer we read exchange_rate from the Blockscout token API.

17 of 30 differed by more than 2%. The worst cases:

Token	Chainlink	Explorer	Difference
CRWV	$78.64	$11.24	−85.7%
IONQ	$35.35	$45.77	+29.5%
COIN	$176.81	$210.42	+19.0%
MU	$973.79	$885.93	−9.0%
CRCL	$71.72	$65.29	−9.0%
MSTR	$101.15	$92.32	−8.7%

CRWV is the one to sit with. An eighty-six percent error is not drift or staleness — it is a different number about a different thing. Anything that priced a position or computed a portfolio value from that field got an answer wrong by a factor of seven. And since an explorer’s market-cap column is supply times its own price, the market caps inherit the same error: CRWV’s real onchain capitalisation was around $417k, not the $59.6k reported.

To be fair to the convenient source: on well-covered tokens it was fine — AMD, SNDK, ASML and ORCL matched Chainlink to within a fraction of a percent. The failure is not uniform, which is exactly what makes it dangerous. Spot-checking three tokens would have told you everything was fine.

The oracle side has its own honest caveat: staleness is normal. At a single instant, feed ages ranged from 0.0 to 6.3 hours (GOOGL 6.3h, SLV 5.5h, AMZN 5.3h). This is not a malfunction — the stock feeds update 24/5 with a deviation threshold and a long heartbeat, so a quiet stock simply does not tick, and outside US market hours they freeze entirely. The mistake is not the staleness; the mistake is displaying a stale price as current, or transacting on one.

What the measurements dictated#

Basquets was designed backwards from these three findings. Each one maps to a decision you can verify in the docs and the contracts.

Liquidity is composable but thin → the core must not depend on trading. A basket is a fixed recipe of components, and minting or redeeming it is in-kind: you deposit or withdraw the underlying tokens themselves. No swap happens in the core protocol, ever. Swaps live only in the periphery — the zap that converts USDG into a whole basket in one transaction, and the savings plans that automate it — and the periphery is replaceable. If liquidity vanishes tomorrow, existing baskets still mint, redeem and hold. On a $17M market, designing for thin pools is not pessimism, it is the job.

Prices disagree → one source of truth, and it is the one settlement uses. Every value-bearing path in Basquets reads the same Chainlink feeds through one public contract, the NAV oracle. There is no “display price” from a different source than the “settlement price” — the stock token dashboard shows the same oracle data the contracts read, tags every figure with its source, and a token we cannot price from a source we trust shows no price at all. A second source that disagrees is not redundancy; it is ambiguity.

Staleness is normal → freshness is a first-class condition, and we fail closed. The NAV oracle’s nav() returns the last known value with a fresh flag for display; navStrict(), which every value-bearing code path uses, reverts unless every component passes every check — staleness against the feed’s heartbeat, sequencer uptime, price positivity, corporate-action pause. When US markets close, minting pauses and the UI says so. That is not a limitation, it is what makes a stock product behave like one — and what makes basket tokens safe for other protocols to build on.

Only ~a third of tokens have a feed → components are allowlisted. A basket may only contain tokens registered with a verified Chainlink feed and its heartbeat. This bounds the component universe today, and it also means every basket in existence can always be valued by the same public oracle — no basket of unpriceable tokens, no fake “NVDA” component.

Everything is measurable → so are we. Fees are enumerable onchain in a single controller contract, rate-capped and timelocked; the UI reads them rather than hardcoding them. Holdings, recipes and NAV are public state. Publishing this research is the same posture applied to the market we build on: the check is cheaper than discovering the failure in production.

Method and reproducibility#

Supply figures and Uniswap balances are exact onchain reads — balanceOf, totalSupply, latestRoundData — with no indexer or API in between. Holder counts, 24h volume and the aggregate market capitalisation are explorer-reported and indicative only; as Finding 3 shows, explorer-derived dollar figures deserve skepticism, so treat their order of magnitude as the claim. All figures are a snapshot of 21 July 2026. For current numbers rather than this snapshot, the stock token dashboard reads the chain live, with every price tagged with its source.