Capital flows and token valuation, measured on the same axis.

Follow the capital. TVL is a level: what sits in DeFi contracts on the chain today. Bridged value is the part of a chain's value secured that arrived from another chain; the "7d vs L2 median" column is its weekly move minus the median move of every L2Beat project above $200M, with both numbers under it, so a chain gaining ground on its peers stands out from a market-wide move. Stablecoin net flow is the 30-day change in pegged dollars circulating on the chain: dollars appear on a chain when somebody mints or bridges them there, so it is the closest public reading of where capital is actually moving. A muted value is a chain outside the bench's ranked cohort, shown from the daily history and left out of the leaders and counts. Open interest shows where leveraged positions and prediction bets sit.

Valuation divergences. Price to fees is market cap over the last 30 days of fees annualized; price to sales uses the protocol's own share of those fees. Each token is read against the fee-weighted median of its category, and the "fees up, token down" badge marks the rows where four things hold at once: fees grew more than 20% against the prior 30 days, the token fell more than 10% over the same days, the P/F sits below the category median, and the category holds at least 5 ranked protocols. The block at the top of the tab lists the five with the largest fee growth; the mirror badge marks the opposite. Neither is a recommendation: fees can grow for one month, a token can fall for reasons the fee line does not see, and a low float means most of the supply is still to come.

Why those numbers. Both legs have to clear the cohort's own noise or the badge means nothing. The median 30-day fee growth across the ranked protocols was +21%, so a row at or under 20% grew no faster than the median protocol that month. Twenty of the ranked tokens, the flat middle of the field, moved less than 10% in either direction over 30 days, so a move inside that band is not a direction. And a median over two protocols is a comparison with one other protocol, so a category under 5 ranked members has its vs-category cell marked not applicable instead of carrying a ratio. Every threshold is checkable from the Fees MoM, Token 30d and Category columns in the table itself.

Reading each signal. Under the divergence table, each of the two signals says in three lines what it means on the numbers in the row, what would falsify it, and what to check next. The falsifier is the part that matters: a month of fee growth can be one incentive program and a month of fee decline can be a quiet category, so the reading names what to look at before treating either as a trend. None of it is a recommendation, and every claim in it is checkable from a column on this page or a page it names.

Empty cells say which kind of empty they are. A dash carrying a small letter is a column that does not apply to that row, and the legend under the table spells out why: Ethereum is the chain the bridges start from rather than a destination, a sovereign L1 has no host chain to bridge from, Circle runs no CCTP domain there, DeFiLlama publishes no fee adapter for the chain. A plain "n/a" is the other thing entirely, and it means here what it means on every other page of this site: the feed should carry a value for that row and does not. A muted value is a chain outside a bench's ranked cohort, read from the daily history and left out of the leaders and the counts.

Under each table, three lines say how to read its main column, where it misleads, and what it does not say.

DeFiLlama for protocol fees and revenue (dailyFees, dailyRevenue per adapter), chain TVL, DEX volume and stablecoin circulating per chain. Windows are 30 closed UTC days, annualized as 30-day sum times 365/30.

L2Beat for value secured per scaling chain, split into native, canonical and external, and its 7-day change; the cohort is the OCB chain registry intersected with L2Beat's scaling summary.

CoinGecko for market cap, fully diluted valuation, circulating and total supply and the 30-day price change. Polymarket and Kalshi APIs for prediction-market open interest, DeFiLlama TVL for venues that publish none.

Each perp venue's own API through the perp cohort harness, for open interest and 24-hour traded notional both. That column used to read DefiLlama's open-interest overview through bench 265, and it was changed on 29 September 2026 for two reasons: the two disagreed by half on the largest row (Hyperliquid at $8.6B against the venue's own $12.8B) and bench 265's cohort is token-gated, so venues without a ranked token had no row at all. Turnover is bench 271, which divides 24-hour averages of the same two gauges, so it does not divide exactly into columns showing the latest read; it is read from the bench rather than recomputed here because one ratio must not have two values on this site.

Not measured here. Token unlock schedules (no free source publishes them; float and its 90-day change are the closest public proxy), bridge volumes other than the ones OpenChainBench reads itself (DeFiLlama's bridge endpoints are paid), and any figure a venue reports about itself without a public trail. Every number on this page is reproducible from the URLs above and released under CC BY 4.0; the rows of this page are one JSON payload at /api/capital (cached five minutes, the same gating as the page), and the daily history is at valuation/history.json and chains/history.json.