The same share, six issuers, six prices: measuring tokenised asset spreads

What was measured

Every token that CoinMarketCap files under the same real-world asset, for the top 53 assets by rank on 2026-09-10, out of 7,942 the endpoint tracks. For each asset: the price of every token, the issuer that minted it, its market cap and its 24-hour volume. Then the spread from the cheapest to the dearest, with the issuer named on both sides.

45 assets produced a number. 8 were refused, with the reason recorded — 3 of them because their tokens are not denominated in the same unit, which is the finding this piece ends on. The run cost 54 credits across 54 calls, and the raw result is one dated JSON file.

The number that matters, and the number that does not

A tokenised Tesla share is a claim on the same underlying whoever issued it, so its tokens should cost the same. On 2026-09-10 they did not — by 55.949% from cheapest to dearest.

That number is worthless. It comes from TSLA on Hyperliquid Assets, a token with $4,749 of market cap and $0 of trading in twenty-four hours. Nobody can transact at that price. Across the 5 Tesla tokens that clear both a $1,000,000 market cap floor and a $10,000 daily volume floor, the spread is 0.061%.

Both figures are true. One is a market and the other is a listing. Most published RWA spreads are the first kind, because the first kind is the one that makes a headline, and the payload gives you everything you need to tell them apart in the same response.

Result

Across the whole run, the median spread between the cheapest and dearest liquid token of an asset is 0.344%, over 22 assets, with 0 above one percent. Counting every comparable token, thin venues included, it is 0.864% over 45 assets with 18 above one percent.

So the honest answer to “does tokenised Tesla trade at one price?” is: nearly. Where there is a market on both sides, these things track each other to within a fraction of a percent. The wide numbers are venues, not disagreements about value.

The widest liquid spread in the run is 0.995% on Taiwan Semiconductor Manufacturing Co Ltd. Its all-token spread is 0.995%, so on that asset the filter barely moves anything — the spread is real.

How much of this is our judgement

The two liquidity floors are a choice, and the headline moves with them. Publishing one number without that range would be presenting a choice as a measurement, so the same measurement is repeated at five pairs of floors on every run:

From 0.864% with no floor at all down to 0.135% at $10,000,000 and $100,000, over 7 assets. Every step down the table is also a step down in coverage. The highlighted row is the one quoted above; we picked it because it keeps a usable number of assets while excluding venues with no market, and that is a judgement rather than a result.

The finding underneath, and it is about the data

Measuring the spread meant reconciling our arithmetic against the average CoinMarketCap publishes for each asset. That reconciliation found something better than the spread.

Netflix carries five tokens under one asset id. Four sit near $75.98. The fifth is 9.961× that:

Nothing in the payload states a unit or a quantity for any of them, so which of those two prices is “a Netflix share” is not something the data answers. The interesting part is that the same response resolves it two different ways:

The token is excluded from the asset’s average price and counted in the asset’s volume. Which means 61.447% of the tokenised 24-hour volume reported for Netflix comes from the token whose price the same payload declines to average.

Excluding it from the price is the right call — averaging across denominations produces a number that is not a price of anything. What costs the consumer is that neither decision is marked. There is no flag on the token, no count of what was dropped, and the weighting of the average is not documented, so someone summing the tokens gets $495.72 and has no way to learn from the response why the field says $76.04. We worked the weighting out by fitting candidates across the whole run: it tracks the volume-weighted price of the in-band tokens closely and the plain mean — which is what the field name suggests — badly.

Gold is the same shape, in the other direction

Gold carries tokens priced per gram beside tokens priced per troy ounce, under one asset id:

The ratio between the two groups is 31.11, and there are 31.1035 grams in a troy ounce. These issuers do not disagree about the price of gold; they are quoting different quantities of it. Silver does the same. We refuse to score either asset and print the reason instead, because a spread across mixed denominations is arithmetic rather than market.

One issuer keeps appearing

Ondo Assets is the dearer side in 12 of the 22 liquid pairs in this run. We name both issuers on every row so that pattern is visible rather than buried inside a median. We have not established why, and we are not asserting a cause: it could be fee structure, quote cadence, venue mix, or the composition of this particular sample. It is a thing to watch across the daily series, not a conclusion.

What this does not say

• A spread is not an arbitrage. Tokens from different issuers are not fungible with each other; redemption, custody, fees and settlement differ by issuer.
• Market cap and 24h volume are not depth. They are the only liquidity fields in the payload, and a token can clear both on one large trade with no order book behind it. The liquid spread is the more defensible of the two numbers, not a guarantee.
• Two spreads are published per asset: one across every comparable token, one across only those that clear both floors. The first is kept so the effect of thin venues is visible rather than quietly removed.
• The floors are a judgement call. method.sensitivity repeats the headline at five pairs of floors so the reader can see how much of the answer is that judgement.
• Prices are a snapshot at one instant, and the API timestamps the asset rather than the token, so a single stale quote inside a fresh asset cannot be detected.
• The underlying price is not in these endpoints. tradfi_markets names the venue but carries no price, so this compares issuers against each other and never against the real-world instrument.
• Which tokens belong to which asset, and who the issuer is, is CoinMarketCap's attribution. We report it and do not audit it.

Reproduce it

The dated receipt for this run is in the tool’s repository, and every figure above is generated from it rather than typed. To check the arithmetic without a key, without the network, and without trusting us:

git clone https://github.com/dyplux/assay.git && cd assay
python3 assay.py --offline

That recomputes every headline figure from the per-token rows in the newest receipt and prints whether the recomputation matches what was recorded. The measurement runs again once a day and appends a new dated file, so the series is exactly as long as the record and nothing is backfilled. The live tool, with one line per asset and every issuer placed by price, is at assay.dyplux.com.

The endpoints and the method are named in full on the tool’s own page, including the thresholds and what each one excludes. The collector’s internals — how the denomination inference is built, how issuers are reconciled across assets, and the parts of the pipeline we intend to keep building on — are not described here.

Disclosure

The author works at CoinMarketCap. This piece is independent, was built outside that role on a personal API key, was not reviewed by CoinMarketCap or Binance, and is not financial advice. A spread is not an arbitrage: tokens from different issuers are not fungible with each other. No positions are held in any asset or issuer named here.