Methodology and data limits
RWA Markets Methodology (https://rwa-markets.xyz/methodology) explains how volume, open interest, velocity, trader, DeFi, and equity figures are collected, what they can and cannot claim, provenance tiers, permanent gaps, and the incident log. Coverage windows are queried live from the database. Cite this page whenever quoting dashboard metrics.
Generated 2026-11-22 16:31 UTC.
Per-venue collection methods
A venue's method class determines what its numbers can honestly claim. Wallet-attributed counts on sampled venues are always labeled as samples next to the venue's own totals.
| Venue | Type | Method | Direction | Coverage (live) | Notes |
|---|---|---|---|---|---|
| Hyperliquid | DEX | Observed fills | observed | Venue totals: 2026-04-25 โ 2026-11-22 Wallet fills: 2026-05-01 โ 2026-09-28 Taker stream: 2026-07-09 โ 2026-11-22 Net flows: 2026-07-04 โ 2026-11-22 Complete wallet flows: 2026-07-09 โ 2026-11-22 | Three layers: complete per-wallet WS aggregates (hl_wallet_flows_1h, hourly, since 2026-07-09), a per-fill sample (recentTrades poll + one-time 0xArchive HIP-3 deep backfill with real Open/Close direction and taker flags), and candle-authoritative venue totals. users[] = [buyer, seller]; taker mapped side-dependently. Both legs of sampled fills are stored (maker direction derived from the taker pair, 2026-07-10); venue-level skew counts strict taker legs only, wallet-level exposure counts both. |
| Extended Exchange | DEX | Observed fills | unavailable | Wallet fills: 2026-04-30 โ 2026-10-22 Taker stream: 2026-07-09 โ 2026-11-22 Net flows: 2026-07-03 โ 2026-11-22 | Starknet Trade-event decode (both legs, 50/50 by construction) to 2026-10-22 13:41 UTC; from 14:00 UTC that day, when Extended moved settlement to Arc, its public taker trade feed on the same both-legs basis (measured 2.00-2.04x on shared markets); venue-level long/short re-split by the observed taker ratio. Deposits and withdrawals from Arc contract events, reconciled daily to the custody contract's USDC. |
| Lighter | DEX | Observed fills | observed | Venue totals: 2025-09-01 โ 2026-11-22 Wallet fills: 2025-09-18 โ 2026-11-22 Net flows: 2026-05-01 โ 2026-11-22 | Two deployments, shown as one venue: core (Ethereum, mainnet.zklighter.elliot.ai) and Robinhood Chain (api.rh.lighter.xyz, from 2026-06-26). Volume, OI and trades are summed per ticker; hovering a Lighter row shows each deployment's share. Closed days come from Lighter's own trade and hourly OI files on AWS Open Data (s3://aws-public-blockchain/v1.1/lighter/), from 2025-09-01 (the first RWA listing); OI there counts both sides and is halved to our one-sided convention (matched our hourly OI at median 1.0000 over 146,110 symbol-hours). The open day is live: candles and snapshots for venue totals, the trade socket for wallets. Wallet flows are a complete hourly per-account aggregate from the same files (Hyperliquid's census model); a closed day replaces the live hours at about 07:30 UTC the next morning. The per-fill leg table (core only) is a sample: 34-93% of AWS fills on days measured May to September 2026. Wallet skill PnL is the venue's own realized PnL (average entry), not a FIFO reconstruction. Trades are UTC-day fills on closed days; the open day carries the venue's rolling 24h count. Decisions (2026-10-27): Lighter here is PERPS ONLY, as Hyperliquid is; Lighter's tokenized-stock spot books (core XAUT/USDC, rhSPY/USDC, rhQQQ/USDC, about $20k a day, previously counted here; Robinhood Chain */USDG books, about $254k a day) belong to the spot side and were moved out from 2026-10-27 forward, with earlier rows left as written. Closed-day volume before 2026-04-25 and OI before 2026-07-12 come from the AWS files; closed-day trade counts moved from the rolling 24h count to UTC-day fills (old values kept); the wallet skill uses the venue's own realized PnL instead of a FIFO reconstruction. |
| Nado | DEX | Observed fills | observed | Venue totals: 2026-04-25 โ 2026-11-22 Wallet fills: 2026-04-02 โ 2026-11-22 Net flows: 2026-04-30 โ 2026-11-22 | Archive indexer matches (both legs; 92.5% carry is_taker from raw tx payloads - the NULL residual is excluded from skew). History deep-backfilled to 2026-04-30. |
| Phoenix | DEX | Position deltas | observed | Venue totals: 2026-05-23 โ 2026-11-22 Wallet fills: 2026-05-01 โ 2026-11-22 Taker stream: 2026-07-13 โ 2026-11-22 | Volume, trade counts and wallet identity come from the full seat-history fill walk (the fill instruction carries the taker's seat + signing key). DIRECTION comes from the position snapshot-diff (from 2026-07-13): each per-trader Trader account is decoded with the SDK-exact struct (Ellipsis-Labs/rise-public phoenix_rise; base_lot_position sign = long/short, len-bounded packed array, is_position filter), and the hourly signed-position change per wallet is the directional pressure - a positioning signal, not taker fill flow. The washed byte-parity per-trade direction it replaces is excluded. One onchain program (EtrnLzgb...) carries the whole venue, crypto and equity; the seat walk covers ~41.5K seats and the position-diff decodes ~3.1K open positions. Trader positions DO reconcile to the venue's reported open interest once the basis is matched: the venue publishes ONE side, so the larger side of the trader book is the like-for-like figure (0.99x on 2026-08-06), not the sum of both sides. |
| GM Trade | DEX | Position deltas | observed | Venue totals: 2026-07-06 โ 2026-11-22 Wallet fills: 2026-05-01 โ 2026-11-22 Taker stream: 2026-07-09 โ 2026-11-22 Position deltas: 2026-07-09 โ 2026-11-22 | Hourly onchain position snapshots (gmsol Position accounts, IDL-verified) diffed per wallet; direction = sign of the signed-notional change; open interest reconciled against the venue API each hour. |
| Pacifica | DEX | REST aggregate | observed | Venue totals: 2026-04-25 โ 2026-11-22 Wallet fills: 2026-06-27 โ 2026-07-09 | REST fills feed (taker side), live-only (40-row API cap, no pagination). Onchain ledger is a single opaque program account (no per-wallet state to decode); wallet rows only. |
| Flash Trade | DEX | REST aggregate | unavailable | Venue totals: 2026-05-23 โ 2026-09-03 Wallet fills: 2026-06-28 โ 2026-06-30 Net flows: 2026-06-27 โ 2026-09-03 | Per-market volume, trades and long/short flow from fstats.io, a third-party Flash analytics site (disclosed; Flash matches orders offchain, so there are no onchain fills to decode ourselves). Open interest and per-side OI positioning from Flash's own /dune/openIntrest endpoint, a Dune-query republish served on Flash's domain. Deposits/withdrawals decoded onchain. A first-party Flash volume endpoint was identified 2026-07-13; migration is pending a parallel diff-window (fstats reads cumulative, Flash 24h rolling). |
| Variational | DEX | Per-side OI | unavailable | Venue totals: 2026-04-25 โ 2026-11-22 Net flows: 2025-06-07 โ 2026-11-22 | RFQ venue with no public fills or book; per-side open interest (a positioning signal, not taker flow) + size-tiered firm quotes + settlement-pool net flows. |
| EdgeX | DEX | WS taker stream | observed | Venue totals: 2026-01-22 โ 2026-11-22 Taker stream: 2026-07-08 โ 2026-11-22 Net flows: 2026-05-01 โ 2026-11-22 | WS trade stream with isBuyerMaker (taker side); deposits/withdrawals via StarkEx + eth_getLogs. |
| Aster | DEX | WS taker stream | observed | Venue totals: 2026-04-25 โ 2026-11-22 Taker stream: 2026-07-08 โ 2026-11-22 Net flows: 2026-07-05 โ 2026-11-22 | WS trade stream with isBuyerMaker (taker side). Deposits via bridge-transfer tracking on Ethereum, Arbitrum, BSC and Solana (the Solana leg parses USDC/USDT deposits into the bridge program, signer-attributed; since 2026-07-10). Aster Chain itself is private - per-trade wallet attribution is impossible. |
| GRVT | DEX | REST taker feed | observed | Venue totals: 2026-04-25 โ 2026-11-22 Taker stream: 2026-07-08 โ 2026-11-22 | REST trade feed with is_taker_buyer (taker side). Own hyperchain; no onchain wallet attribution. Untracked: GRVT's Stable Funding Perps (the USDC-margined commodity markets listed in November 2026: BZ, CL, NATGAS, XAG, XAU, XPD and XPT). GRVT publishes no trade feed or candles for them, so their volume (about $0.1M a day at the 2026-11-13 close) is not counted, never shown as zero. GRVT's USDT-margined perps on the same commodities are collected as usual. |
| Ondo Perps | DEX | REST taker feed | observed | Venue totals: 2026-04-25 โ 2026-11-22 Taker stream: 2026-07-08 โ 2026-11-22 Net flows: 2026-05-01 โ 2026-11-22 | REST trade feed with aggressor_side (taker side); net flows via eth_getLogs. Ondo Perps also lists spot pairs (23 on 2026-11-19, ANT-USDC among them); spot is not collected because the venue publishes no spot volume, so every figure here is perpetuals only. |
| Carbon | DEX | Net flows only | observed | Venue totals: 2026-06-18 โ 2026-11-22 Wallet fills: 2026-05-01 โ 2026-11-20 Net flows: 2026-05-01 โ 2026-11-19 | Deposit/withdrawal discovery via public Arbitrum RPC eth_getLogs; no per-trade data. |
| QFEX | DEX | REST taker feed | observed | Venue totals: 2025-12-14 โ 2026-11-22 Taker stream: 2026-02-26 โ 2026-11-22 Net flows: 2026-06-11 โ 2026-11-22 | Offchain central limit order book (USDC margin on Arbitrum, BitGo custody); RWA-only book (equities, ETFs and commodities by the venue's own category). Per-market UTC-day volume from the venue's 1DAY candles, venue total cross-checked hourly against QFEX's own daily aggregate; open interest from its contracts feed. Taker direction (long/short) is the venue's own hourly taker buy and sell notional from its /taker-volume endpoint, read every hour and rewritten as each hour completes (history from the same endpoint): aggregated by the venue, so it carries no wallets and no trade counts, and buy plus sell equals the venue's hourly candle volume. Eight names quote in KRW/JPY/CNY/SEK and are converted to USD at spot FX (QFEX's own fields carry them in quote currency), taker notional included. QFEX names four ETFs after the index they track (KOSPI=EWY, TAIEX=EWT, NIKKEI=EWJ, HSI=EWH); they are stored as the ETF. No trade count is published, so trade-based metrics are blank rather than estimated. Prices off its own index (in-house oracle). Net flows cover the Arbitrum USDC deposit rail only (per-user BitGo forwarders); fiat (ACH/wire/FedNow) and USDT-on-TRON rails are invisible, so the net figure is that rail's, not the venue's. |
| Decibel | DEX | Aggregate stats | observed | Venue totals: 2026-06-15 โ 2026-11-22 Net flows: 2026-07-06 โ 2026-11-22 | Onchain perp exchange on Aptos with equity, commodity and crypto perps plus a small crypto spot book. Oracle: Chainlink. RWA = the venue's equity and commodity perps; crypto perps stay in the dominance denominator; spot is not collected. Per-market UTC-day volume from hourly candles (base volume x OHLC average), reconciled to the venue's own daily_stats to 0.03%; crypto volume is that venue total minus the RWA sum. Net flows are the collateral deposit and withdrawal entry functions, signer-owned rows only (internal store hops excluded), from the Aptos indexer. No trade count is published. |
| Veranta | DEX | Aggregate stats | observed | Venue totals: 2025-10-08 โ 2026-11-22 Net flows: 2025-10-08 โ 2026-11-22 | Perpetuals on Base (formerly Avantis; renamed Veranta on 2026-09-14), USDC margin, no spot book. Volume is the venue's daily pair array, checksummed against its own venue total to $1 on a closed day. RWA is the EQUITIES, FOREX and COMMODITIES groups when the ticker passes is_rwa; CRYPTO1-4, including the five Upside markets, stay in the dominance denominator and are not asset rows. Per-name open interest is the live catalog and is forward-only. Oracle label is Pyth Network (Lazer on every listed pair; Hermes on some; Chainlink armed on seven crypto pairs only). No order book, so spread and depth are unsampled. Net flows are ERC-4626 Deposit and Withdraw on the avUSDC tranche, wallet = owner; trader settlement and share transfers are not flows, and the vault TVL also moves with trader PnL so the flows do not reconcile to TVL. No funding rows: the unit was not confirmed. |
| Bulk | DEX | WS taker stream | observed | Venue totals: 2026-09-24 โ 2026-11-22 Net flows: 2026-06-01 โ 2026-11-22 | Perpetuals whose matching runs on Solana validators (bulk-agave); USDC custody in one Solana program. Mixed book: RWA is the TradFi markets that pass is_rwa (US500 at S&P 500 index level, single stocks including SPACEX = SpaceX, Nasdaq-listed as SPCX since 2026-06-12, SOXL, PAXG as gold); crypto perps stay in the dominance denominator. Equities trade 24/7 on this venue, so weekend volume is real. UTC-day volume from hourly candles (base volume x OHLC average). Trade count, taker direction, wallets, fees and liquidations come from the venue's public WebSocket trade tape, captured live since the integration (the venue serves no trade history); a day whose tape covered under 98% of candle volume publishes no trade count. Open interest is the ticker's one-sided figure x mark (the venue's /stats OI is two-sided). Funding is hourly; history before the integration is rebuilt from per-wallet funding payments. Oracle: Pyth (venue docs); prices match Pyth's 24/7 equity feeds within 1 bp, which keep publishing through US post-market, overnight and weekend sessions (Bulk does not publish its feed ids). Net flows are USDC deposits to and withdrawals from the venue's Solana vault (the Solana USDC rail), from the vault's first transaction on 2026-06-01, so the June to August pre-deposit campaign is included (the venue's own API starts at mainnet, 2026-09-05, and shows carried-in balances as one deposit that day; it is not used). A withdrawal is booked at the payout transaction; the user's request moves nothing. Every row's tx is checked against the vault's balance chain, and a sample matched the venue's account history to the micro-USDC. Transfers between Bulk accounts never touch the chain and are not flows. |
| HelloTrade | DEX | Aggregate stats | observed | Venue totals: 2026-03-10 โ 2026-11-22 | Perpetuals on Monad, USDC margin. Mixed book: RWA is the markets that pass is_rwa (US equities AMZN, NVDA, TSLA; ETFs EWY, QQQ, SPY; gold XAU and silver XAG); BTC, ETH and SOL stay in the dominance denominator. Market data is the venue's public WebSocket (the REST API is wallet-scoped). UTC-day volume is the sum of hourly candles (base volume x the bar's volume-weighted price); a closed day matches the venue's own day bar to the dollar. The venue publishes no trade count and no open interest, so both are absent (never zero) and HelloTrade carries no OI share. Funding is the ticker's hourly rate. Oracle: Stork, a per-market mark and index feed (HELLO_<SYM>_M / _I). Volume history is rebuilt from the venue's hourly candles; book depth exists from the integration on. |
| Imperial Perps | DEX | Aggregate stats | observed | Venue totals: 2026-05-23 โ 2026-11-22 | Aggregator routing through Phoenix/Flash/GM Trade; venue-level volume/OI only, deducted from the underlying venues to prevent double-counting. |
| Backpack | CEX | WS taker stream | observed | Venue totals: 2026-04-25 โ 2026-11-22 Taker stream: 2026-07-09 โ 2026-11-22 | WS taker-side trade stream + venue stats API. |
| Binance | CEX | WS taker stream | observed | Venue totals: 2026-04-25 โ 2026-11-22 Taker stream: 2026-07-01 โ 2026-11-22 | WS taker-side trade stream (spot + futures) + venue stats API. |
| OKX | CEX | WS taker stream | observed | Venue totals: 2026-03-23 โ 2026-11-22 Taker stream: 2026-07-01 โ 2026-11-22 | WS taker-side trade stream + venue stats API. Which contracts count as RWA is the venue's own declaration (instCategory stock or commodity on its swaps; its xStocks spot book is not counted, so spot is never booked as perp), read daily since 2026-11-11, ahead of our ticker lists; contracts declared before that date carry history from the venue's daily candles. |
| Bybit | CEX | WS taker stream | observed | Venue totals: 2026-04-25 โ 2026-11-22 Taker stream: 2026-07-01 โ 2026-11-22 | WS taker-side trade stream (spot + futures) + venue stats API. Which contracts count as RWA is the venue's own declaration (symbolType stock, ETF, commodity or forex on its linear contracts), read daily since 2026-11-11, ahead of our ticker lists; contracts declared before that date carry history from the venue's daily candles. |
| Bitget | CEX | WS taker stream | observed | Venue totals: 2026-04-14 โ 2026-11-22 Taker stream: 2026-07-01 โ 2026-11-22 | WS taker-side trade stream (spot + futures) + venue stats API. Which contracts count as RWA is the venue's own declaration (isRwa on its USDT futures), read daily since 2026-11-11, ahead of our ticker lists; contracts declared before that date carry history from the venue's daily candles. |
| Kraken | CEX | WS taker stream | observed | Venue totals: 2025-05-18 โ 2026-11-22 Taker stream: 2026-07-01 โ 2026-11-22 | WS taker-side trade stream (spot + futures) + venue stats API. |
| Coinbase | CEX | WS taker stream | observed | Venue totals: 2026-04-25 โ 2026-11-22 Taker stream: 2026-07-01 โ 2026-11-22 | WS taker-side trade stream + venue stats API. |
Metrics & cards
What each dashboard card computes and the caveats that used to live in its footer. This list grows tab-by-tab as each card's claims are re-verified against the data.
Oracle consensus
Oracle consensus on RWA Markets is a cross-venue ranking score for a tokenized asset, not a price feed and not a venue's settlement oracle. On each tracked venue, assets are ranked by that venue's volume plus open interest; the rank becomes a 0 to 10 contribution (zero when volume and OI are negligible), then those contributions are averaged across the tracked venue universe. A score of 0 means the asset had no 24h volume or under $5,000 of open interest. The Pricing Oracle column is separate: it reports which mark each venue uses (Pyth, Chainlink, in-house, order book, and others), weighted by that venue's volume and OI. Venue oracles set contract settlement on that venue; they are not RWA Markets's consensus score. Live values are on Top Traded Assets and the Asset Analyzer; the Partner API returns the same fields.
- A venue is listed only where its fills and books are captured; every other venue is shown as not yet measured or untracked with the reason, never as zero. Oracle-priced and RFQ venues (GM Trade, Veranta, Variational, Carbon) have no book mid and are untracked.
- A fill is scored only when the book stream was live at the fill and at each markout instant; socket outages and restarts (exec_marks_gaps, bulk_tape_gaps) leave fills counted but unscored, never estimated. Fills still waiting on their markouts when the collector stops keep their score at the fill and carry no 60 s or 300 s figure.
- Taker fee: realized per fill on Bulk and Lighter (each Lighter trade carries its own taker fee; Standard accounts pay 0, so Lighter's figure is what its takers actually paid); the venue's published base-tier schedule elsewhere: Hyperliquid (without the frontend's builder fee), Phoenix (its per-market fee: 3.5 bps, 0.35 bps on FX), Extended (1.0 bps on markets it files as RWA, 2.5 bps on the rest) and Pacifica (tier 1, 4.0 bps). Large takers on volume tiers pay less, so schedule-basis all-in cost is an upper bound for them. Frontend and builder fees are left out everywhere. Where no fee is published or verified, all-in cost is n/a; Lighter days before 2026-11-14 carry no per-fill fee, so its all-in cost appears once the window holds only measured days.
- Liquidation and ADL fills are excluded where the venue flags them (Bulk, Lighter). Hyperliquid's trade stream carries no liquidation flag, so its liquidation fills are counted in the averages.
- Mid timing differs by venue: Hyperliquid is exact to the block; Lighter's book stream arrives about once a second. The card shows the share of fills whose book was under 1 second old.
- Concentration: one taker can carry most of a venue's notional. The fill-weighted mean, the largest taker's daily share and each day recomputed without that taker are shown on venues whose tape carries wallets.
- Size buckets are per fill: a taker order filled in several pieces lands in a smaller bucket than the order. History starts at each venue's first measured day; the 30-day view fills in as days accrue.
- Today is the UTC day so far, as of the time shown; 7D and 30D are the last 7 and 30 closed UTC days, today excluded; a date shows one UTC day. Days measured are counted per venue and asset, and a market first captured inside the window says from when (a listing the stream picked up after it began trading).
- Inside the quote: the share of fills that printed strictly inside the spread standing before them. Where it is above half (GRVT and EdgeX at launch: orders posted milliseconds earlier at exactly the fill size, the pattern of self-crossed or internalised flow), the line is marked with an (i) and its headline uses only the fills that traded against the public book; all fills stay in the detail.
- Venues with no stored fill tape (the centralized exchanges, Aster, Nado, QFEX, Ondo Perps, GRVT, EdgeX, Backpack and Lighter's Robinhood Chain books) are measured into 30-minute totals as fills arrive; their spot and perpetual books are separate sub-lines and are never merged into one number.
- Exchange markets that trade a different token under an RWA ticker are left out: a crypto token that shares an equity ticker on that exchange (for example the CAT memecoin on OKX and Kraken, which is not Caterpillar), the RLUSD stablecoin, and currency pairs such as USD/CAD. Their fills are kept but not counted; from 2026-11-21 this removes about 0.4% of the centralized-exchange notional measured since 2026-11-12.
- Deltas need a prior-window baseline: assets without history in the chosen window show no delta rather than a fabricated 0%.
- Volume is summed across all tracked venues; a venue outage understates the total rather than being estimated through.
- Intraday partial days are trimmed from growth math - including today would read as a fake collapse.
- Today's end-of-day volume figure is a projection from hourly pace until the day closes; it is labeled as such.
- Historical OI is point-in-time and never restated after the fact.
- All OI is counted one-sided (each contract once, never once per leg), so venues stay comparable; platforms and venue dashboards that count both legs show up to 2x our figures. The OI Dominance methodology entry has the per-venue convention.
- The ORACLE row covers perpetual markets only, on the last complete UTC day, so its total is slightly below COMBINED (which includes spot and can be the running day). Its shares are never rescaled onto the COMBINED total.
- Each market is attributed to the oracle it uses today; past days in the 30-day view carry today's attribution.
- Markets with no verified oracle are shown as Unverified with their share, never folded into another oracle. Order Book and RFQ / Solver venues use no oracle and are labelled as such.
- Hyperliquid open interest is split between its builder exchanges by each one's share of that day's volume.
- A venue whose direction is not observed shows a dash, never a placeholder split.
- Where a venue publishes no taker side, the long share comes from per-side open interest and is marked as positioning, not flow.
- A 24x5 venue shows its last trading session outside market hours; its positions stay open through the weekend.
- Volume is each venue's whole RWA total, spot included, on the same basis as the Combined bar above.
- The funding extreme is capped at 500% APR: uncapped, a thin book's per-hour print (observed: 6,727% and 12,183%) wins the headline over anything tradeable.
- Carbon prices each side of a market independently (asymmetric legs); when a Carbon market wins the funding extreme the figure is its declared long-side leg, labeled as such with both legs on hover.
- Net-flow coverage spans the venues with onchain deposit/withdrawal tracking, not all venues - see the per-venue table above.
- Survival curves need the asset to age: day-30 retention only exists for listings at least 30 days old.
- Dominance % is share-of-venue (RWA over that venue's RWA+crypto volume), not share of the whole market; a tiny venue can be 90% RWA on negligible flow.
- The legend's RWA:crypto ratio counts listed markets (breadth), not volume; a venue can list many RWA markets and still trade few.
- Today's point is the venue's in-progress 24h figure and moves until the day closes; the midnight pipeline run skips it rather than writing a premature near-zero.
- A collector outage holds the last good hourly print instead of writing a fake 0% dominance datapoint (zero-guard).
- The volume KPI's day-over-day delta compares that in-progress figure with yesterday's completed day: read it as pace, not a final print.
- OKX and Bitget daily volume before 2026-08-05 is aligned to those venues' default daily candle, which opens at 16:00 UTC (Hong Kong), not 00:00 UTC. Rows dated D in that period therefore span D 16:00Z to D+1 16:00Z and are shifted ~8 hours against every other venue in the same table. From 2026-08-05 both are read from the UTC-aligned bar for the day being written; the earlier rows are left as the venues natively reported them rather than restated.
- Imperial Perps is a routing aggregator: its trades execute on Flash Trade, Phoenix, GM Trade and Jupiter, and already appear in those venues' own volume. It keeps its own row and card, but is excluded from cross-venue totals and share denominators so no routed trade is counted twice. Its figure is therefore NOT additive with the venues it routes to. Jupiter has no standalone collector, so its routed slice (~0.3% of Imperial) leaves the totals with it.
- Flash Trade's volume comes from fstats.io, a third-party Flash analytics site with an opaque methodology; Flash matches orders offchain, so there are no onchain fills to decode ourselves. That endpoint's undated response silently changed from a day-shaped figure to a trailing 30-day total on 2026-06-22; the call now pins an explicit start/end date to the UTC day being written, and 2026-05-23 to 2026-08-04 was rebuilt from the same date-filtered source.
- Ondo Perps listed BTC, ETH and SOL as perps on 2026-07-06 and HYPE on 2026-08-14 (ONDO is on the same crypto book). It is on the dominance % view from 2026-07-06; Carbon still lists no crypto and stays off that view. Crypto volume is the venue's own UTC-day candle, same basis as the RWA numerator. A later crypto listing enters that denom the hour it appears and never goes into asset_velocity.
- A venue whose every point in the window is 100% RWA has no crypto book to compare against, so it is left off the dominance % view (Carbon; QFEX, which listed only equities, indices and commodities at launch on 2026-10-05). The first day a crypto market prints there, the venue appears on the view with no change on our side.
- Decibel (Aptos) lists crypto perps alongside its equity and commodity perps; its crypto volume is the venue's own daily total minus the RWA sum. Its five spot markets are crypto and are not collected.
- Veranta's crypto denominator includes its five Upside markets (BTC, ETH, SOL, XRP, HYPE). Those are separate symbols from the plain crypto perps, and they are large enough to swing the venue's RWA share from day to day. The share is RWA over that denominator, not a claim that the Upside book is equity volume.
- OI is a point-in-time stock, not a flow: each date stores the last snapshot taken that day, and historical OI is never restated after the fact.
- Only today's row receives OI updates; a venue's OI history therefore starts the day its collector went live (e.g. GM Trade 2026-07-06), even where volume history reaches further back.
- Dominance % is share-of-venue (RWA over that venue's own RWA+crypto OI), not share of the whole market.
- Open interest is counted one-sided: every contract is counted once, never once per leg. Order-book venues report this convention natively (a matched book's longs equal its shorts); for pool-counterparty venues (Variational, GM Trade, Flash Trade) we sum user longs plus user shorts, which still counts each user-vs-pool contract exactly once.
- Some venue dashboards and analytics platforms sum both legs of the same contract and publish roughly double these figures; Variational's own headline OI is exactly 2x the per-market long-plus-short figures we ingest (verified live 2026-07-13), so our Variational number reads at half the venue's banner. One-sided counting is what keeps venues comparable on the same axis.
- Flash Trade's open interest comes from Flash's own /dune/openIntrest endpoint; the /dune/ path means the figure is a Dune-query republish (Dune-derived), served on Flash's domain.
- Ondo Perps crypto OI (BTC/ETH/SOL from 2026-07-06; HYPE from 2026-08-14; ONDO on the same book) is the same end-of-day snapshot as its RWA OI. Historical OI is not restated; the series is what the collector wrote that day.
- Veranta per-name open interest starts the day the collector is live. Backfilled dominance rows carry zero OI on purpose: the venue publishes a per-pair OI only on the live catalog, and the historical open-interest endpoint is a venue total, not a per-name series.
- Range averages include a venue's pre-capture days as zeros: for venues whose OI collection started recently, a windowed average understates the true level (158 volume-bearing rows carry zero OI).
- 'Latest' is a single day's snapshot; picking a past date shows what was recorded then, never a reconstruction.
- Coverage-change markers flag dates where a venue or protocol started reporting; growth across a marker is a coverage artifact, not market movement, and the insight sentence says so when markers are in the window.
- DeFi TVL history starts later than OI history; missing TVL days render as gaps, not zeros.
- The growth insight compares the window's first and last points only; it is a net change, not a trend fit.
- Assets with zero global volume or under $5,000 global OI are floored to a consensus score of 0, so dust listings cannot ride a single venue's rank into the table.
- A venue position under $100 of volume+OI earns no points on that venue.
- The Pricing Oracle score is the top oracle's share of the asset's volume+OI across venues; it measures oracle concentration, not oracle quality.
- For centralized venues, 'Pyth' means Pyth is the primary underlying source the RWA contract tracks; each venue wraps that reference in its own index and mark-price formula, and some blend minor secondary sources for assets Pyth does not cover (OKX adds the Ondo API, Coinbase adds Blue Ocean ATS during US hours, Bitget adds dxFeed/Massive/Intrinio). Binance sources via its Atlas engine, which consumes Pyth and republishes it to BNB Chain. Attribution reflects the dominant underlying feed, not exclusivity.
- Venue-level attribution is refined per asset where a venue splits oracles across its book (Lighter, Extended, Hyperliquid builder markets); markets whose feed cannot be verified to a named provider are labeled Unverified rather than defaulted, and pure order-book or RFQ venues carry no oracle attribution at all.
- Spread, depth, and slippage in the venue popover always come from the latest daily liquidity snapshot, even when a historical as-of date is selected; historical market structure is not archived.
- Windowed averages use only days an asset actually traded; missing days are not counted as zeros.
- Spreads of 50%+ and books with nothing within 2% of mid are excluded as unmeasurable rather than averaged in.
- Trade counts follow each venue's authoritative source; on Hyperliquid that is candles, after wash-bot prints made the raw feed's counts unusable.
- Lighter trades are captured live from the native WebSocket market feed; api.0xarchive.io is a third-party historical trade archive used for backfill and as a heartbeat-gated safety net when the WebSocket goes stale, spending ~0 credits in normal operation.
- RFQ venues (Variational) can print 2-4 trades averaging $9M+; real prints, but they stretch the size axis. The 'Min 5 trades' toggle hides sub-5-trade points and is off by default so nothing disappears silently.
- Venue-asset pairs with zero trades or zero average size in the window are dropped, not plotted at the origin.
- Hover liquidity (spread/depth/slippage) is always the latest daily snapshot, not historical; Hyperliquid builder-dex clones are merged volume-weighted since only one clone gets sampled per cycle.
- Hourly flow playback exists from 2026-07-04 (collector go-live); daily history reaches back to 2026-04-14.
- Bars begin when a venue's collection began, not when the venue listed the asset; a venue joining the stack mid-window is usually a coverage event.
- Today's bar runs about 35 to 95 minutes behind: hourly candles are read after each hour closes. A venue that publishes only rolling 24-hour figures (Imperial Perps; Variational until 2026-11-20, before its hourly series covered a full day) is marked on today's bar, since its figure covers the trailing 24 hours rather than the UTC day.
- Hyperliquid builder-dex clones of the same asset are aggregated into one Hyperliquid contribution.
- Perpetuals only. A spot pair of the asset (a gold token, a tokenized stock on an onchain order book, a spot-only listing) is left out of these bars and shown under the Spot toggle and in Spot Markets; spot is never added to the perpetuals bars.
- OI is point-in-time per day and never restated; days before a venue's OI capture began are absent, not zero.
- RFQ venues (Variational) have no public order book; their cost comes from firm size-tiered quotes ($1K/$100K) and is starred as quote-based on the card.
- When no venue has real 2% depth, best execution falls back to lowest spread only and the card says so; with no sampled spreads at all, no crown is awarded.
- Spreads of 50%+ and sampled-but-empty books (nothing within 2% of mid) are excluded as unmeasurable (the EdgeX equity-band case).
- Venues whose feed returns a level-capped book (every reported level already inside 1% of mid) print identical 1% and 2% depth; their impact at size is computed against that capped depth and can be overstated, so the card daggers their impact columns. Rankings are unaffected.
- The identical 0.02bps spreads printed by the major gold books are real one-tick spreads (a $0.01 tick at roughly $4,060 gold, verified against raw venue books 2026-07-21), not a units artifact.
- Live data is the latest liquidity snapshot, refreshed on a ~10-minute poll; it is not an order-book replay.
- 7D and 30D book figures need 7 or 30 complete UTC days of hourly samples (at least 20 of 24 hours each); until a venue has them its row says how many days are collected. Hourly sampling began 2026-11-20. Carbon and EQX books have no window figure.
- A venue whose trades our stream does not capture still appears with its volume when its own candles show trading; its book columns then read not sampled.
- Perpetual books only. Spot books a venue also lists (gold tokens, tokenized stocks on Nado and Hyperliquid, EQX) are left out of its row; their cost at size is in Spot Markets.
- Both figures are two-sided depth; neither is directional. The +-2% depth is a superset of the +-1% depth by construction.
- Venues without a sampled book (RFQ, not-yet-sampled) are omitted from the chart rather than drawn at zero.
- Perpetual books only; spot books are not added to a venue's depth.
- These are wallet-attributed fills; on venues where capture is a sample (e.g. Hyperliquid's per-fill feed) they are not the venue's total - venue totals always come from the venue's own API. Phoenix is complete taker-fill history since the 2026-07 seat walk.
- The long/short split covers classified volume only; venues where the taker side is unobservable are listed with an explicit direction-n/a row instead of being folded into the split.
- Extended Exchange's row re-splits its onchain volume by the asset's observed taker ratio from its live trades feed (coverage begins 2026-07-09); the headline split takes only covered-days volume.
- Phoenix's row reads taker-side prints from the venue's own trade stream (live since 2026-07-15; 2026-05-01 to 2026-07-04 replayed from chain history verified at 100.0% agreement against the stream; 2026-07-05 to 2026-07-14 is a disclosed gap with volume but no direction). The prints are wallet-attributed and feed the tier breakdown; the earlier invalid byte-parity direction was removed from history.
- Pacifica's row comes from its public fills feed at venue level (no wallet attribution exists there, so it carries no wallet or tier contribution); coverage begins at the 2026-07-13 aggressor-mapping fix.
- Per-asset coverage notes render on the card itself because they change with the asset.
- Market-maker wallets are excluded (>=50 trades, >=85% long/short balance, active >=70% of days over a recent window) so the hold time reflects directional traders. On MM-heavy venues like Lighter, MMs generate the overwhelming majority of episodes at ~1min holds; the organic median is hours (Lighter ~7h). Same MM definition and default as /traders/position-durations.
- Completed, non-left-censored episodes only. The first episode of every wallet-symbol series is left-censored (it may be closing a position opened before capture began) and excluded; positions still open at build time are reported only as an open share, never assigned a duration.
- GM Trade timing is hour-resolution (its position log is hourly), so its shortest episodes floor near 1h; Lighter and Nado use fill timestamps.
- Flow reconstruction is exact at the position level, not sub-fill: a wallet whose net exposure oscillates within the flat band is one episode, and USD-vs-base-quantity drift is negligible at the scalp/intraday horizons that dominate, while long holds fall in coarse buckets where boundary error cannot change the answer.
- A per-venue cell with fewer than 20 completed episodes is shown as n/a rather than a noisy median.
- Market makers are excluded here as well; the Position Duration card on Trader Intelligence shows them as a separate series.
- Hyperliquid episodes are built from its complete trade stream (every print with both wallets) from 2026-08-30 onward, so its hold time covers that period only, not its earlier sampled history.
- High-frequency accounts (1,000+ trades per active UTC day on 5+ of the last 30 days, from each wallet's daily RWA perp trade count) are excluded from the hold time as well and counted in hf_wallets; the include option keeps them. Extended (no per-wallet daily trade counts since its move from Starknet to Arc on 2026-10-22) and Pacifica (none collected) are not screened for high-frequency accounts, so their medians exclude market makers only.
- APR annualizes each venue's own interval; a 1h venue's headline rate compounds 8x more often than an 8h venue's, so same-instant APRs are comparable but per-interval rates are not.
- Spread and TPM come from the latest daily asset_velocity aggregate, not a live book read.
- The anomaly flag marks a venue whose suspected market-maker share (50%+) and spread both look unusual against peers over the trailing 7 days; it is an investigative lead, not proof of manipulation.
- Venues listing several markets for one base asset show the average across those markets, labeled on the row.
- Extreme prints on thin books are shown as reported; funding on an untradeable book is a signal about the book, not a harvestable yield.
- Carbon prices its long and short side independently (asymmetric legs, one dealer per market). The venue row shows the declared long-side leg, marked on the card, with both legs on hover; the sign follows the same positive = longs pay convention as every other venue.
- Mark-price coverage is complete across all venues from 2026-07-13; earlier history has gaps for Lighter, OKX, Pacifica and part of Bybit (the rate itself was unaffected). Pre-launch synthetic markets without a real mark show no price rather than zero.
- A 0% lending APY on a yield-bearing token is often a collateral-only deployment, not a broken pool; where the underlying token distributes a native yield (e.g. tokenized-fund dividends) that rate is read issuer-direct (the issuer's API or the fund's own onchain oracle) and shown instead of a fake 0%, with DefiLlama only as a labeled fallback where no issuer source exists.
- n/a in the APY/utilization columns means the pool doesn't offer that function (not borrowable, nothing borrowed), never that data failed to load.
- The card only renders for assets with actual DeFi deployment; absence of the card means no tracked pools, while a load failure shows an explicit error state.
- Prices come from Hyperliquid's own oracle price on its Pyth-fed markets, labeled hl_oracle: on weekdays that is the Pyth publisher price, and while the cash session is closed the venue prices the market itself. Pyth's direct feed has required an API key since 2026-08-26 and is used only when one is configured. A ticker Hyperliquid lists as a different asset is excluded (CL is crude oil there, not Colgate). Where no oracle covers a ticker the fallback ladder is our cross-venue perp mark, then a liquidity-gated Jupiter DEX quote, then (for an explicit shortlist of exchange-listed tickers, e.g. SECZ, CURR and 28 Robinhood Chain names) the exchange's last close via a public quote API, labeled tradfi_close and removed per ticker the day an oracle covers it; the card footnote names which source priced the snapshot.
- Supply is circulating, not raw mint supply: xStocks supply comes from the issuer's proof-of-reserves API because raw onchain totalSupply counts pre-minted treasury inventory (median 50x overstatement); Ondo and Backpack legs are read onchain per chain.
- Spot / perp-OI compares outstanding tokenized shares against open interest on tracked venues only, so the ratio is a ceiling, not an exact multiple. OI in the ratio is counted one-sided (each contract once); a platform that double-counts both legs would halve this ratio.
- Per-issuer outstanding and share multiply each issuer's supply by the single oracle reference price; issuers are 1:1-backed twins of the same underlying, so one price is the honest common denominator.
- The In-DeFi, Holders and hold-time figures come from the tokenized-equity intelligence tables (see the DeFi Markets entries); holder counts are Solana-census accounts and do not yet include EVM holders. The Solana holder census refreshes weekly (onchain holder enumeration is expensive; composition moves slowly and daily supply-flow capture bridges the interval); EVM holder classification refreshes daily.
- Spot hold vs perp contrasts the value-weighted open-lot age of this ticker's tokenized spot with the median completed perp round-trip (market makers excluded); sub-hour routing hops are excluded from the spot side.
- 'In Kamino lending' is this ticker's equity deposited in Kamino, attributed to the real depositors from the onchain obligation snapshot (the reserve's total supply USD shared out by each depositor's raw deposit). The depositor still owns it (posted for yield), so it is a held position; Jupiter Lend holds a similar amount but has no obligation snapshot yet and is not counted here.
- The chain-split bar shows where the outstanding supply lives, in chain brand colors; xStocks reports all-chain circulating through its proof-of-reserves API, so its share renders as one multi-chain segment until per-chain supply reads exist.
- As perp margin is tokenized spot posted as collateral on a perp venue, not DeFi lending (that is the In lending line). Nado dollars are remaining posted units marked at each subaccount's last event oracle, not the signed event-USD net (event-USD overstates names that churn). Until nado_collateral_genesis is stamped the stock understates silent holders and the line says so. Lighter dollars are a live stock snapshot; Ethereum and Robinhood Chain are named separately and never summed without that split. Match is the census ticker (wSPYx and rhSPY both land on SPY; XAUT0 on XAUT; PAXG stays on PAXG). No trader count: addresses are not traders.
- A venue missing a market-day inside a window has no figure for that window and is named beside the total, never counted as zero.
- Exchanges not read (Gate and MEXC until their perpetuals are integrated, and any other exchange) are listed as not tracked.
- Spot has no open interest, funding or liquidations; those figures on the page are perpetuals.
- Dollar and fiat stablecoins and fiat currency pairs are not tokenized assets and are not counted.
- Only the hourly comparison basket is read on exchanges, so an asset outside it has no exchange cost figure.
- A concentrated-liquidity pool quote past the active price range is a floor (marked), and routing across pools is not modelled, so real execution can be better.
- Depth: two-sided resting depth within ยฑ1% and ยฑ2% of mid, the median of the last 24 hourly reads, recorded from 2026-11-21 on (the books are not stored, so earlier days have none); a band the fetched book does not reach is a floor, marked โฅ.
- A row marked * is not adjusted for the token's share multiplier, so part of its gap can be the multiplier, not a price difference.
- Pool-days trading more than 10,000 times their liquidity are left out of every window.
- Solana is shown apart as the sampled estimate with its 95% range: its DEXs publish volume but not the buy and sell sides. It is never added to the EVM figures.
- The long/short split covers only venues where direction is observed; venues without it (Flash Trade, Variational) contribute volume to totals but never to the skew. Extended Exchange joins the skew per-day from its live taker feed (coverage begins 2026-07-09), Pacifica from its corrected fills mapping (2026-07-13), and Phoenix from taker-side prints on its own trade stream (2026-05-01 to 07-04 replayed from verified chain history, live from 2026-07-15; the 2026-07-05 to 07-14 gap carries no direction); uncovered dates are excluded rather than extrapolated.
- Hedged volume is 1:1 greedy-matched (each leg pairs once, nearest opposite leg within 5% size and 15 minutes); the earlier flag-and-sum method inflated it roughly 8x and was replaced 2026-07-08. Confirmed = same wallet on both legs; coincidence = opposite flow across two wallets lining up on symbol, size, and time.
- Institutional share is computed over wallet-attributed volume on per-trade venues; venues without per-wallet data are outside its denominator.
- Per-fill rows are tx-hash deduplicated; the earlier additive-upsert pipeline double-counted re-polled fills up to 4.9x (fixed 2026-07-01, see incident log).
- A day's split can be dominated by whichever high-volume venue had direction coverage that day; the blend is volume-weighted, not venue-weighted.
- Extended Exchange enters the daily blend re-split by its own observed taker ratio for that day (live feed, coverage begins 2026-07-09); its onchain record stores both legs of every fill and is 50/50 by construction, so uncovered earlier days are excluded.
- Phoenix enters the daily blend through its own taker-side trade prints (chain-history replay covers 2026-05-01 to 07-04, live stream from 2026-07-15); days in the 2026-07-05 to 07-14 gap carry no direction and are excluded, never extrapolated.
- The Trades column is the venue's own total (candle-derived) where known; the long/short split behind it is a direction-attributed capture that can be a fraction of that total - the exact sample share per venue is disclosed on hover.
- CEX rows fall back to an OI-delta proxy when the window extends beyond live taker-side coverage; the card banners this, and the proxy's historical agreement with real data is audited whenever both exist.
- MM share is a heuristic (50+ trades, 85%+ long/short balance, active 70%+ of days), not a confirmed identity; the anomaly flag additionally requires a live spread over 1.5x the cross-venue median and is an investigative lead.
- No per-venue funding APR is shown here because this view blends every asset per venue; per-asset funding lives on the Asset Analyzer's Funding & Carry card.
- OI positioning rows (Variational, Flash Trade) are a point-in-time positioning snapshot, not taker flow, and are never blended into skew figures.
- Flash Trade's long/short flow split comes from fstats.io per-side daily totals (third-party, opaque methodology; Flash matches orders offchain, so no onchain fills exist). Flash's own API exposes only OI positioning, a different signal, so the flow split stays on fstats until a first-party per-side volume source exists.
- Extended Exchange and Pacifica re-split their real volume by an observed ratio (tagged as observed); Pacifica's aggressor coverage begins at its 2026-07-13 mapping fix. Phoenix (both programs) carries per-trade taker direction from the venue's own trade stream: live capture since 2026-07-15, and 2026-05-01 to 2026-07-04 replayed from raw chain history after the onchain decode matched the stream on 100.0% of 2,341 joined fills. 2026-07-05 to 2026-07-14 is a disclosed direction gap: those days contribute volume but never direction. The blended skews take only observed-window volume, never a re-split extrapolation over uncovered days. Positioning snapshot-diffs are still collected for Phoenix but are a different signal and are no longer blended into flow.
- Extended Exchange's taker feed changed transport on 2026-07-17, from a REST trade poll to the venue's public WebSocket trade stream, after four days of side-by-side parity (daily long-share within 1.2pp, volume coverage 94% or higher); the taker-side field, direction semantics and hourly buckets are unchanged.
- An OI delta is one number per venue per window; it cannot see intra-window reversals or distinguish longs closing from shorts opening.
- The agreement score exists only for windows fully covered by live taker data (collection began 2026-07-01).
- Coverage spans venues with onchain deposit/withdrawal tracking only; a venue absent here is untracked, not flat.
- QFEX is the Arbitrum USDC rail only. It also takes USD by ACH, wire and FedNow and USDT on TRON, none of which is onchain here, so a fiat deposit later withdrawn as USDC reads as an outflow with no matching inflow. Measured over 105 days to 2026-10-04: 76% of the rail's withdrawn dollars went to addresses that never deposited through it.
- Decibel (Aptos) counts only the deposit and withdraw entry functions and only the signer's own collateral row; transfers between Decibel's internal stores are not flows. Aptos addresses do not join EVM or Solana wallets in cross-venue views.
- Veranta flows are LP deposits and withdrawals on the avUSDC vault (ERC-4626 Deposit and Withdraw, wallet = the owner). They are not trader margin. Trader settlement and share transfers are excluded, and trader PnL moves the vault's TVL, so the net will not match the change in TVL.
- Perps is the default view because lending-vault churn (Sky, Kamino, Aave allocators) is a different kind of flow and would drown trader activity.
- Bridge routers, venue contracts, exchange custody and market-maker wallets are not counted as depositors: neither their dollars nor their wallets are in the totals, so dollars and wallet counts cover the same flows.
- Lighter deposits are credited to the depositing wallet from Lighter's own deposit record, so deposits routed through bridges count. Lighter moving funds to its own settlement chain and a solver's inventory moves are not withdrawals.
- Protocol treasury wallets that cycle large round amounts in and out of a savings vault (Sky sUSDS) are left out of the totals and shown on the venue's row; a wallet that entered once and left once is a holder and counts.
- Hyperliquid is its Arbitrum USDC bridge only; money reaching Hyperliquid by another route is not counted.
- EdgeX V2: deposits are stablecoin transfers into EdgeX's forwarding contract and vault (per-account deposit addresses, connect-wallet deposits, solver deliveries) on Ethereum, Arbitrum, Base, Optimism, Polygon and Arc; withdrawals are its CCTP payouts and direct payouts to users. BNB Chain and Solana deposits are not counted yet; the V1 Arbitrum pool is EdgeX-Arb.
- Correlation needs overlap: venues with deposit data but no matching volume series are excluded and listed by the API.
- A correlation under about 0.4 is treated as no reliable signal, and the insight sentence says so rather than crowning a weak leader.
- Protocol treasury/allocator wallets are labeled inline; their round-number capital cycling is operational, not trader sentiment.
- Only venues with per-wallet attribution contribute wallets; venue totals elsewhere include venues these tiers cannot see.
- DeFi lending flow (supply/borrow on Sky, Maple, Kamino, ...) is shown in its own column and never mixed into traded volume.
- RWA % is the average share of each wallet's activity that is RWA, from stored per-wallet category breakdowns; wallets without a breakdown are excluded from that average, not counted as 0%.
- High-frequency accounts (1,000+ trades per active UTC day on 5+ of the last 30 days) are left out of each tier's traded volume and DeFi-lending flow and counted in hf_wallets; they stay in the wallet counts and the RWA % average.
- Lending-venue wallets are a different species from perps traders; the Perps/Lending toggle exists so allocator wallets don't read as traders.
- Suspected market makers are flagged by heuristic (50+ trades, 85%+ balanced, active 70%+ of days), not confirmed identity.
- High-frequency accounts (1,000+ trades per active UTC day on 5+ of the last 30 days) are left out of the top-wallet lists and counted in top_wallets_hf_wallets.
- Completed, non-left-censored episodes only; the first episode per wallet-symbol is left-censored (may close a pre-window position) and excluded. Positions still open at build time are the open share, reported separately and never assigned a duration.
- Wallets are tiered by their most recent size tier (the same rule as the tier summary); a wallet never portfolio-scored is tiered from its own trailing-30d volume, so active wallets are not dropped into Unknown.
- Suspected market makers are a separate series by default (a toggle includes them): a two-sided MM's rapid churn would otherwise dominate the scalp bucket. The MM heuristic is applied at query time, so the stored episodes stay definition-agnostic.
- A tier cell with fewer than 20 completed episodes shows n/a instead of a noisy median.
- Venue set is GM Trade, Lighter and Nado; Hyperliquid joins once its direction sweep and position-action capture land. Phoenix, Extended, Pacifica and Flash are excluded (taker-only fills or no per-wallet attribution).
- High-frequency accounts (1,000+ trades per active UTC day on 5+ of the last 30 days) are excluded by default and counted in hf_wallets; the include option keeps them. Extended (no per-wallet daily trade counts since its move from Starknet to Arc on 2026-10-22) and Pacifica (none collected) are not screened for high-frequency accounts.
- Position data covers the 4 venues with per-wallet position tracking: Hyperliquid, Lighter, Extended Exchange, and Nado.
- A category's rise can mean new positions or new coverage of existing wallets; position tracking began later than trade tracking on some venues.
- Per-trade venues only (Hyperliquid, Extended Exchange, Lighter, Nado, Phoenix); Pacifica and Flash Trade have no per-wallet attribution and sit outside this chart.
- A wallet's tier is its classification at the window end; a wallet that grew mid-window is counted at its final tier.
- High-frequency accounts (1,000+ trades per active UTC day on 5+ of the last 30 days) are left out of the tier volumes and shares and counted in hf_wallets.
- The sample is the top wallets per tier from the segmentation endpoint, not all tracked wallets.
- Portfolio value comes from each wallet's most recent segmentation snapshot, so it can lag the trade data by up to a day.
- High-frequency accounts (1,000+ trades per active UTC day on 5+ of the last 30 days) are left out of the sampled wallets and counted in top_wallets_hf_wallets.
- The venue-exposure basis differs per venue (net deposits vs position notional vs perp account value; the response's exposure_basis field names it) - compare a venue with itself over time, not one venue's ratio against another's.
- Hyperliquid is a special case: self-custody there means HyperCore SPOT balances (parked, undeployed capital) plus the same wallets' Arbitrum wallet contents; its census cohort is wallets with at least $100K 30d volume, so smaller HL wallets are out of frame.
- Wallet contents are a point-in-time read of a curated list (stables + majors + registry RWA, no long-tail) and can be stale - each row carries its scored_at date; venue exposure is computed live at query time.
- Meaningful participants only: venue settlement/pool infra (huge flow footprint) and wallets with under $1K net deposits are excluded so unrelated chain balances can't inflate the ratio.
- The ratio is suppressed when the census covers less than 80% of a venue's exposure USD, and a single parked whale can dominate the in-wallet sum - the per-venue top-wallet share (in_wallet_top1_pct) is reported alongside so concentration is visible.
- Ondo GM is a spot venue: its tokenized stocks already sit in the trader's own wallet, so its 'venue exposure' is that holding and its ratio reads as dry powder kept next to the position; the holder universe comes from the equity-intelligence holder snapshot and only current holders (>= $1K of Ondo assets) count as participants.
- High-frequency accounts (1,000+ trades per active UTC day on 5+ of the last 30 days) are counted, never removed, and shown in hf_wallets.
- Stables + majors only (the dry-powder question), NOT the long-tail RWA registry - this is 'where do they keep cash', not a full portfolio.
- A wallet counts toward every venue it trades on, so summing venue totals double-counts shared traders; the overall figure is the deduplicated wallet union.
- EVM-native venues only. Extended is Starknet-native and its trader identity is an undeployed Starknet key with no reachable EVM address, so it is a documented gap, not a zero.
- Point-in-time balance read (scored_at); withdrawals after the scan are not reflected until the next census.
- High-frequency accounts (1,000+ trades per active UTC day on 5+ of the last 30 days) are counted, never removed, and shown in hf_wallets.
- Solana venues appear only where real flow events exist (Kamino, Flash Trade); absence of a Solana venue is a data gap, not zero flow.
- The cohort is wallets seen on 2+ venues in the same address space; single-venue capital is out of frame by construction.
- High-frequency accounts (1,000+ trades per active UTC day on 5+ of the last 30 days) are left out of the deposits and withdrawals by default and counted in venue_flow_hf; the include option keeps them.
- Matching is string-exact within an address space. Cross-space entity links are no longer invisible: declared bridge evidence (a wallet's Relay/LayerZero transfer naming its own address on the other space, recorded in bridge_transfers) proves EVM and Solana addresses belong to one entity; the live count of those declared links is shown in the card subtitle. Venue-overlap matrices above remain string-exact within one space.
- The per-venue activity column mixes evidence kinds (flows, balances, traded notional); compare a wallet's activity within one venue over time, not across venues.
- Solana lending-and-trading overlap (Kamino/Jupiter positions plus perps activity) is the onchain carry cohort, matched by identity rather than inference.
- High-frequency accounts (1,000+ trades per active UTC day on 5+ of the last 30 days) are counted, never removed, and shown in hf_wallets.
- Confirmed = the same wallet string on both legs; linked entity = different addresses proven to be one entity via the wallet-link graph, whose edges are either onchain (shared funder) or DECLARED bridge identity (the wallet sent a Relay or LayerZero transfer naming the other address as recipient, recorded in bridge_transfers); everything else is coincidence-window matching and may be two unrelated traders.
- This reconciles with the Flow Overview hedged KPI by construction (same pairing helper); the earlier flag-and-sum method overcounted ~8x.
- EVM and hybrid venues only today; venues without per-wallet fills cannot participate.
- High-frequency accounts (1,000+ trades per active UTC day on 5+ of the last 30 days) are left out of the pairing (their legs are dropped before matching) and counted in hf_wallets.
- A sequence is temporal adjacency, not proof the second leg was funded by the first.
- EVM address space only today.
- Basis default is traders-only (aggregating vaults excluded); toggling changes headline USD materially on DeFi lending venues.
- High-frequency accounts (1,000+ trades per active UTC day on 5+ of the last 30 days) are left out of every sequence and counted in hf_wallets.
- Cohorts require attribution history: dense coverage starts 2026-04-30/05-01 on every per-wallet venue (Hyperliquid, Extended, Lighter, Nado, Phoenix) after the July deep backfills; only Flash Trade/Pacifica start later (late June 2026), so early periods under-count cohorts on those two.
- A 'fill' is one open or one close; a round-trip counts as 2 fills against the min-trades threshold.
- A wallet lapsing can mean it left, or that its venue's collection lapsed; check the incident log before reading a cliff as churn.
- High-frequency accounts (1,000+ trades per active UTC day on 5+ of the last 30 days) are left out of the cohorts, the retention curve and the tier composition and counted in hf_wallets.
- Only wallets with a stored category breakdown contribute; cohort members without one are excluded from the percentages rather than guessed.
- High-frequency accounts (1,000+ trades per active UTC day on 5+ of the last 30 days) are left out of the cohorts and counted in hf_wallets.
- Suspected market maker is a heuristic flag (50+ trades, 85%+ balanced, active 70%+ of days), not a confirmed identity.
- High-frequency accounts (1,000+ trades per active UTC day on 5+ of the last 30 days) are left out of the list and counted in hf_wallets.
- CEX hourly detail exists only from 2026-07-01 (collector activation); hours before that are DEX-only, and each hour is marked with whether real CEX data is blended in.
- Hour-of-day says when volume prints, not where traders live; market-open effects (US equities RWAs) dominate timezone inference.
- Elysium (from 2026-11-09): wrapped xStocks and EQX tokens bridged from HyperEVM are counted where they sit on Elysium (holders rebuilt from every transfer, pool contracts as DeFi); the HyperEVM bridge escrow behind them is excluded there, so each unit is counted once.
- Scope is the equity-issuer registry (Ondo Global Markets, xStocks, Superstate, Securitize, Backpack, bStocks (Binance)): tokenized stocks plus those issuers' ETF wrappers; every non-stock ticker in it is an exchange-traded fund (commodity trusts like GLD, treasury-bill and bond ETFs like SGOV or TLT, crypto ETFs), together about a tenth of the universe by value, and the 'Stocks only' toggle removes them. Tokenized treasury and money-market FUNDS (BUIDL, PRIME, USYC, ...) and non-ETF commodity tokens (PAXG) are different products, covered by Tokenized Fund Issuance and the Yield section; their DeFi deployment is deliberately not in this card's figures.
- The denominator is measured circulating float: issuer treasuries, pre-mint inventory, bridge escrows and exchange custody are classified and excluded, so this DeFi share reads against real float, not raw mint supply.
- Protocol labels come from resolving each protocol account's owning program; a residual sits under 'unlabeled protocol accounts' and is shown, not hidden. The largest single position was traced to a Kamino lending reserve and cross-checked against its transaction history.
- Contract-held is NOT DeFi-held: on Ethereum the large contract holders turned out to be Gnosis Safe multisigs (institutional smart-wallet custody), verified individually against the explorer; they count as organic holders, and only labeled protocol contracts (Uniswap v4, ConditionalTokens, ...) count as DeFi. Without that check Ethereum would have read 45.8% DeFi-deployed; the honest figure is under 1%.
- On BSC, one Binance Wallet custody contract holds roughly half the raw supply (a 34-token basket, 8M+ transactions); it is classified exchange custody and excluded from the float denominator, the same treatment as Backpack's omnibus wallet on Solana. Contract holders that cannot be positively identified appear under 'unverified contracts' and are never counted as DeFi.
- The DeFi figure includes onchain trading and lending infrastructure holdings that are NAMED in the protocol split, including a market maker's RFQ inventory held in a CoW-solver settlement contract; that is deployed capital in DeFi rails, but it is dealer inventory, not user lending deposits, and the split says so.
- Registry tokens with no third-party price are valued at our own oracle reference (the same price the Tokenized Spot card uses), so DeFi-held xStocks value does not silently read as zero.
- Protocol names resolve from each account's owning program: Jupiter Lend positions sit in Fluid Liquidity accounts (Jupiter Lend is built on Instadapp Fluid) and Byreal CLMM is Bybit's Solana DEX. AMM entries (Raydium CLMM, Meteora, Orca, Byreal) are concentrated-liquidity pools: LP capital earning trading fees, which is also where DEX sells settle.
- Chain bars wear each chain's brand color; the week-over-week delta appears once seven days of daily utilization snapshots have accumulated.
- DeFi-held here is a point-in-time STATE read (what sits in protocol accounts), not a flow claim about why it got there; deposits to lending and liquidity in AMM pools both count.
- Same scope as that card: the equity-issuer registry (stocks plus their ETF wrappers); tokenized treasury/money-market funds (BUIDL, PRIME, ...) are a different universe and never appear in these bars.
- Position-size tiers bucket PROTOCOL-HELD positions (a Kamino reserve or an AMM pool position), not holder wallets; one institutional-size position can be a pool shared by many LPs.
- Ticker shares read against that ticker's own measured float, so a thin ticker can show a high DeFi share on small absolute dollars; the by-issuer bars carry the absolute context.
- Splitting this from the utilization card is presentational; both serve from the same snapshot and always agree.
- Aggregate leverage is total collateral divided by total equity, never an average of per-position ratios: a mean would let a $200 position weigh the same as a $2M one.
- The vault rows carry no collateral or debt column because a venue publishing terms is not publishing balances; those cells read 'terms only' rather than zero.
- THE HEADLINE FIGURES DESCRIBE THE ROWS ON SCREEN AND FOLLOW THE VENUE FILTER, computed from the same array the table renders so the two can never disagree. They were previously Kamino's book alone while the table listed six more venues, which put $419M above rows summing to $487M. Money sums across venues because these are different protocols holding different dollars. Two quantities deliberately do NOT sum: OWNERS, because one wallet can borrow on several venues and adding per-venue counts double-counts it, so an owner figure appears only when a single venue is selected; and the POSITION COUNTS of any venue that publishes terms rather than balances, because a supply-side vault position is not an RWA leverage position. Every venue on this card is now read per position, so that exclusion currently applies to nothing.
- A VAULT'S POSITION COUNT IS NOT ITS BORROWER COUNT, and on Jupiter Lend the two differ by more than seven to one: on 2026-11-15 its 16 RWA vaults reported 13,416 positions, of which 1,734 owe money. The rest are mostly SUPPLY-ONLY - lenders who have deposited and never borrowed. They carry no debt and cannot be liquidated, so they are not leverage. The borrower figures on this card and the whole of the near-liquidation card below count only the positions that actually owe money; the vault row keeps the venue's own total beside its terms.
- Loopscale is a first-class venue here, read per LOAN rather than per wallet: it is an order-book credit venue, so one borrower routinely holds several loans and each is a row. Its live book is CREDIT and fund collateral, not equities - ONyc, PST, PRIME and syrupUSDC carry effectively all of it, while the tokenized equities it lists as eligible collateral (TSLAx, SPYx, CRCLx, NVDAx) together hold about eight thousand dollars across thirteen loans. Those equity rows are published so the dust is visible, and they are not evidence of an equity-leverage market.
- Loopscale's lend side and its borrow side are the same dollars seen from two ends: a vault's deployed principal IS its borrowers' debt. The vault TVL on the Lending Markets card and the collateral figure here are therefore never added together.
- JUPITER OFFERBOOK IS ITS OWN VENUE, not Jupiter Lend and not Perena. It is a P2P fixed-rate book where USD* is posted as collateral against USDC. Each loan's account is the borrower wallet, so the owner count is COUNT DISTINCT of those wallets, not of loan pubkeys. The 5 percent on the Lending Markets row is the P2P loan rate; the looped net APY shown on this card is (USD* savings APY on collateral minus that borrow rate on debt) over equity, the same formula Kamino Multiply uses, derived at read time. A missing Perena rate withholds the number rather than writing zero. Looped USD* is a lens on who reused the savings mint; it is never added to Perena TVL.
- JUPITER LEND OWNERS ARE THE NFT HOLDERS. The position account itself carries no owner pubkey; the onchain key is vault_id:nft_id, which is a position identity. Counting those as owners would reprint the position count. The owner is the holder of the position mint, resolved on Triton via singles of getTokenLargestAccounts (a batch of this method 429s) then getMultipleAccounts (SPL owner at bytes 32:64). Rows key the wallet on account and vault:nft on market so two NFTs of one owner in one vault do not collide. Unknown owners still render as n/a, never as 0: a coalesced zero is how this card once read '0 owners' on a 2,231-borrower book. Fluid's resolver returns the NFT owner directly, and that wallet is what the owner count uses.
- NET APY IS RETURN ON EQUITY, not the Lending Markets supply rate. It is (collateral yield minus borrow cost) / equity, dollar-weighted so a $200 loop cannot weigh the same as a $2M one. Collateral yield is the vault's own supplyRate when that rate is positive (Jupiter Lend SPYx 3.00% is an incentive on the vault, not SPY yield; that is the number their screen shows). A published 0 on the vault (Jupiter syrupUSDC, Fluid PAXG) falls through to the token's native / savings rate (issuer feed, then the best non-Pendle/Exponent supply APY for that base ticker). A miss is n/a, never 0: coercing a miss to 0% is how syrupUSDC printed -26% while Maple was 4.67%, and how PST printed -30% with no issuer rate we hold. Equities and commodities with no venue supplyRate stay n/a, not a 0% that becomes -borrow x (leverage-1): Fluid's PAXG/USDC vault publishes 0% supply and 6.63% borrow and no Net APY, so printing -11% was a number they do not show. Kamino Multiply uses the venue's own per-position net_apy once that figure includes native yield; until then the card recomputes from the dominant collateral and the market's debt-stable borrow (never the collateral-only dust rate). Offerbook uses Perena's USD* 7-day APY minus the P2P borrow rate. Other venues use the vault's own borrow rate. A missing borrow rate withholds the number. Loopscale lend-side yield rows carry borrow_apy 0.0 by construction; that 0 is absence, not a free loan, so Loopscale net APY is withheld until the loan rate is persisted. Basket collateral (Compound) is unscored rather than blended. This number can be negative.
- Position counts here are in the thousands and a single large position moves every ratio, so this is an early signal rather than a market-structure claim.
- Jupiter Lend runs vaults its own API does not list. Those whose collateral is a Jupiter Lend AMM pool holding a tracked asset (reUSD, PST, sUSDai and others) are found onchain and shown with the pool as collateral. Their debt is read from the vault's own borrow position and valued only when it is a stablecoin; the collateral is pool shares, so its value, the position count and the health factor are left blank rather than estimated.
- The shock ladder is mechanical arithmetic, not a forecast and not a risk model. It moves ONLY the collateral asset and holds the debt leg still, which is correct for a stable-denominated debt and wrong for anything else. It ignores liquidation penalties, partial liquidations, oracle lag and any position the borrower tops up in the meantime.
- READ THE LADDER AGAINST THE COLLATERAL, because the same shock is not equally plausible for every asset on it. Most of the debt sitting closest to liquidation is a YIELD LOOP on stable-value collateral - tokenized treasuries and private credit such as PST and syrupUSDC - deliberately run at a health of about 1.02 because the collateral is not supposed to move; on 2026-08-11 that shape was $51.4M of the $57.2M within 5 percent of liquidation. A five percent fall in a treasury token is a credit event, not a market day, so those rows are near liquidation in arithmetic and not in the way an equity position at the same health would be. The tokenized-equity collateral on this card - SPYx, QQQx, NVDAx, TSLAx - is where a 20 or 30 percent column describes something that has actually happened to the underlying. The ladder does not weight by asset because any weighting would be our forecast rather than the venue's arithmetic; the collateral column is published beside it so the distinction stays the reader's to make.
- Positions whose venue publishes no health factor are counted in the header and EXCLUDED from the ladder; they are never assumed safe, and the count of unscored positions is shown.
- EVERY POSITION WE COLLECT IS IN THE HEADER AND THE LADDER; the list underneath is a SUBSET. Positions carrying under $10,000 of debt are kept off the list because a $27 position sitting above a $5M one is counting rather than informing, but they are real, we collect them, and they stay in the totals. The distinction matters most on Kamino, where 2,579 of 3,354 positions fall under that floor: they carry $3.5M of a $258M book, so they move the position COUNT a great deal and the DEBT barely at all. Read the ladder's debt column alongside its count for that reason. This was the other way round until 2026-08-10, when the floor was found sitting in the population query as well as the display, understating the book by 2,579 positions.
- Loopscale publishes a per-loan borrow LTV and liquidation threshold as integers on a scale where 1,000,000 is 100 percent. That scale was proved rather than read off the field names: the same ledger's annual rate reproduces its own interest-per-second on it, and 242 live loans sit above their borrow limit and below their liquidation threshold while none sits above the threshold, which is what a real book looks like and no wrong scale produces. Health is that threshold over the loan's current LTV, on the same liquidatable-at-1.00 scale as every other venue on this card.
- Loopscale's endpoint returns CLOSED loans alongside open ones, and their collateral valuations are meaningless: summed as served they report 148 trillion dollars of collateral, essentially all of it from one dead asset on loans carrying zero debt. Only open loans are read, and the flag is re-checked on our side as well as filtered at source.
- Jupiter Lend publishes no per-position debt anywhere, and its position accounts carry no owner pubkey. What each position does carry is a TICK, which in this design encodes a debt-to-collateral ratio, so the debt is reconstructed from the venue's own tick arithmetic against the position's collateral and the vault's borrow exchange price. That is a derivation, so it is gated on two figures the venue publishes and we do not: the decoded position count matches the venue's own count exactly on all 15 RWA vaults, and the decoded debt sums to 99.9 percent of the venue's reported total. Either gate failing holds the whole venue at its last good read rather than publishing a partial book. The liquidation threshold is the venue's own, on a per-mille scale (850 is 85 percent) proven against its wallet-level API rather than inferred from the numbers. The owner is the holder of the position NFT (mint at offset 14): Triton singles of getTokenLargestAccounts (a batch of this method 429s) then getMultipleAccounts on those token accounts. A crawl that leaves more than 10 percent unresolved is a fault, same as a failed decode. Rows key the wallet on account and vault:nft on market so two NFTs of one owner in one vault do not collide.
- Two Jupiter Lend exclusions, both stated because they move the count. Positions already through a liquidation branch carry residual debt that needs branch arithmetic we have not ported, so they are excluded rather than valued on the ordinary tick ratio, which would OVERSTATE their debt - 25 positions on 2026-08-11. And the venue runs vaults its own API does not list, two of them holding PST: 127 positions across 14 such vaults, with no published decimals, price or threshold to score them by, so they are counted here and left out of the ladder.
- Jupiter Offerbook is a different product from Jupiter Lend. Its loans have no published liquidation threshold, so they are counted as unscored rather than given a fabricated health. LTV and leverage are measured from the loan's own USD legs. Each loan is keyed on the borrower wallet, so the owner count on the leverage card is real.
- Fluid publishes no positions over REST at all, so its book is read from its own VaultPositionsResolver contract on Ethereum, which returns each position's collateral and debt already resolved. Gated the same way when it was built (the 16 RWA vaults Fluid listed in August 2026): the decoded count matched the venue's own count exactly, and the decoded debt summed to between 99.99 and 100.00 percent of the venue's reported total. Its risk integers are in basis points (7,500 is 75%), the opposite of Jupiter Lend's per-mille on an identically named field; only that reading leaves every liquidation threshold above its own borrow limit.
- SOME FLUID BORROWERS SHOW DEBT AND NO COLLATERAL, and that is deliberate rather than missing data. Its smart-collateral vaults take a DEX PAIR as collateral, so a position's collateral is denominated in POOL SHARES, not in tokens; pricing those as if they were tokens is what once produced a $2.79 quadrillion figure on a single vault. The card's unscored list gives the current count and debt of those positions. Their debt is real and is published; the collateral cell is left empty rather than filled with a wrong number, and with no collateral there is no health factor, so they are counted in the header and named as unscored instead of appearing on the ladder.
- This is a small set of large positions. One borrower repaying moves every number on the card, so the position count is stated beside the totals.
- A high share of pool means that pool's depth is one manager's position rather than many independent providers, which is a concentration fact worth knowing before reading its depth as a market-wide figure.
- These strategies DEPLOY INTO pools counted in the execution-depth card, so the two are the same dollars seen from different ends and must never be summed.
- A curator is the vault's declared manager, taken from the venue's own metadata. It is not a claim about who ultimately owns the deposits.
- Zodial and Nest, the two Solana spot-equity lending protocols, publish on different rate bases. Zodial's supply and borrow APYs are DECODED from its own onchain multipoint interest curve (five utilization/rate points per bank, verified against the venue's displayed deposit APY on seven assets to within 0.05 percentage points) and move with each bank's live utilization; an earlier read of the legacy rate fields found zeros and wrongly concluded the venue charges no interest, which is why the curve decode is gated against the venue's own UI. Nest remains APY-less BY DESIGN: its collateral earns no protocol rate, so a published APY there would be invented rather than measured, and a zero in its columns means not published, not a zero rate. Zodial's TVL is the sum of its bank liquidity vaults read onchain, and the whole-protocol sum is gated against DefiLlama's independent implementation of the same read before any row is written (measured 0.9994 to 1.0001); the card shows only the RWA banks, which are a subset of that protocol total. Nest's TVL is the collateral vault balances named in the issuer's own signed deployment manifest, each one checked to hold the mint the manifest declares. Nest has NO DefiLlama comparator: the `nest` slug there is usenest.xyz on Hyperliquid L1, a different protocol that merely shares the word. Non-RWA collateral (SOL, JitoSOL) is excluded from the published figure, and nUSD debt outstanding and snUSD staking are deliberately not added to collateral TVL, since they are a liability and a stake against the same dollars. Nest Exchange (that HyperEVM MetaDEX) appears here as DEX LP rows with estimated APR labeled as incentives; it is not Nest nUSD and is not a lending market.
- Loopscale is an ORDER-BOOK credit venue, not a utilization-curve money market: a vault lends one asset against a curated collateral set at fixed per-collateral terms, so its published rate is the vault's BLENDED lender APY over a rolling book of individual loans, and its utilization is the share of principal actually deployed rather than a point on a curve. Its rate is served as a decimal fraction by the venue and is converted once, which was proved against the same payload's own expected-24h-interest figure rather than assumed from the field name. The lend-side TVL shown here and the borrow collateral on the Leverage card are the same dollars seen from two ends and are never added together.
- Jupiter Offerbook is a P2P fixed-rate, fixed-term loop book, not Jupiter Lend and not Perena. USD* is the collateral; USDC is the principal. TVL is unique-creator fillable USDC plus active loan principal: Offerbook's availablePrincipal is min(remaining, a shared per-wallet escrow), so summing remaining across a creator's offers double-counts the same dollars. The rate is hundredths of a percent (500 = 5.00%), proved against the same payload's USD principal rather than assumed from the field name. Looped USDC is never added to Perena's mint-info TVL.
- Lending TVL is GROSS SUPPLIED - the total deposited into the market - not supplied minus borrowed. DefiLlama publishes the net-of-borrows figure for the same pools, so a side-by-side comparison shows us 3-10x higher on Aave, Compound, Spark and Kamino; the underlying markets are identical (their APY bases match ours to within 0.05 percentage points) and neither number is wrong, they answer different questions.
- TVL and APY come from the protocol's own contracts, not an aggregator; fund tokens that distribute yield natively rather than via the pool carry an issuer-direct rate: Hastra's Proof-of-Reserves API for PRIME, the Redstone daily-accrual oracle feed for BUIDL and VBILL (annualized as (1 + daily rate)^365 - 1), and each fund's own onchain NAV oracle for OUSG, USDY, USYC, TBILL, STAC and mTBILL (the ~30-day NAV change, annualized).
- Oracle-derived fund APYs are accepted only between 0 and 15 percent, so an oracle glitch can never paint a fund card; funds with no free issuer source (WisdomTree, Spiko, Galaxy, JPMorgan, Franklin Templeton, Ethena USDtb, Libeara) carry no yield figure rather than an aggregator's guess. DefiLlama remains only as a labeled fallback for ACRED/SACRED, which expose no free issuer rate.
- A pool appearing mid-history is usually collection coverage starting, not a launch; the summary's 7d TVL delta is computed same-pool-basis to exclude exactly this.
- Collateral-only listings legitimately show no supply APY; they are not broken rows. The flip side, found 2026-07-29: because a 0 there is legitimate, a MISSING native yield looks identical to a correct one, and PRIME's went unnoticed for 69 days. Its APY for 2026-05-01 to 2026-07-08 is a DefiLlama backfill of the same Hastra Proof-of-Reserves field we now read issuer-direct (see Pool History for the continuity check); TVL on those rows is our own.
- Kamino markets are ingested from an explicit allowlist of market addresses, so a market Kamino launches is not tracked until it is classified. The AUTO Market (Hastra AUTO, a US auto-loan book) was added on its 2026-07-29 opening day and the PAXG Market alongside it; a daily sweep now reports any unclassified Kamino market rather than leaving it silently absent.
- Monad lending (from 2026-11-12) is read onchain on Monad itself: Aave V3 and Morpho through their official APIs, Neverland (an Aave V3 fork) from each market's PoolDataProvider priced by its own AaveOracle, Curvance from its ProtocolReader (one call for every market) and Gearbox from its pools, credit managers and credit accounts priced by Gearbox's own price oracle. A Monad market or reserve is included as RWA only when one of its tokens is a Monad RWA by contract address (XAUt0, syrupUSDC, Valos vUSD, Midas mROX and mEDGE, Mu muBOND and loAZND, reUSD and its Pendle PT, rwaUSDi, sUSDat, apyUSD, Centrifuge JTRSY/JAAA/ACRDX), never by symbol; base dollars such as AUSD and mUSD never qualify a market on their own (Aave V3's stablecoin reserves are shown as before, as stablecoin rows). Curvance markets are isolated pairs and Gearbox strategies fund one collateral, so their stable lend leg is shown with the RWA collateral leg; in the Neverland cross pool only the RWA reserves are shown. Curvance and Gearbox rates are the venue's annualized rate as each venue prints it; Aave and Neverland rates are compounded APY. Collateral-only reserves show the token's own native yield where we measure one, otherwise the venue's 0. Monad markets under $1,000 (most syrupUSDC and reUSD Morpho markets, the loAZND market's AUSD leg) are not drawn.
- Euler v2 (from 2026-11-14) is read onchain on Monad, Ethereum, Base, Arbitrum, BNB Chain and Avalanche: every vault its factory lists on that chain, with each vault's asset, deposits, borrows, interest rate and accepted collateral. A vault is shown when its asset is an RWA of that chain by contract address; a vault that lends only against such RWA vaults (on Monad, AUSD lent against Valos vUSD) is shown as that collateral's stable leg, and a vault that also lends against other collateral is not. Dollar values use Euler's own price oracles; a vault no Euler oracle can price at the time of the read (some tokenized-stock vaults when their feeds pause outside market hours) is left out rather than priced at a guessed value. Rates are compounded APY from the vault's per-second borrow rate; the supply rate is that rate times utilization, less the vault's interest fee. Vaults under $1,000 are not drawn.
- Maple syrup on Ethereum lending venues (from 2026-11-14): syrupUSDC, syrupUSDT and syrupUSDG are matched by their Ethereum contract addresses on Aave V3, Euler v2, Pendle and Spark (Morpho already matched them by symbol). They are the deployed side: the Maple pools themselves stay fund issuance. A syrup reserve that pays no lender rate (Aave V3 syrupUSDT is collateral-only) shows Maple's own pool yield, labelled with its source.
- The fixed rate is implied by the PT's current price and only realized if held to maturity; selling early re-exposes the position to rates.
- Locking the PT rate trades the variable underlying yield for certainty until maturity; whichever leg outperforms is only knowable after the fact.
- Governance-set rates (Sky) step rather than drift - a flat line is normal; vault-ratio yields (sUSDai) are trailing realized rates, not promises.
- Sky sUSDS TVL and savings APY are read onchain from the vault (the per-second ssr rate and totalAssets); info-sky.blockanalitica is a disclosed fallback used only if the Ethereum RPC is unavailable, roughly zero calls per day in normal operation.
- The set grows as protocols are added; absence means untracked, not unavailable.
- Perena USD*-J and USD*-P are other share classes of the Perena vault, shown as their own savings rows. USD*-P's rate is the onchain seniorFixedApyBps. USD*-J's rate is the vault-sdk settleTrancheNav residual applied to the issuer's USD* 7-day APY.
- Collateral-only fund tokens earn inside the fund itself, not via Aave; a fund's own yield is shown where tracked (Superstate, Centrifuge) and blank where not, never 0%.
- Horizon TVL is gross supplied, same basis as the other lending markets, so it reads above DefiLlama's net-of-borrows Horizon figure; the collateral-only markets match theirs to the dollar and the difference is entirely the borrowed side.
- Hollow points are backfilled: official protocol history (Morpho, Kamino, Superstate, Centrifuge) where available, Sky savings-rate history via the blockanalitica archive (third-party, backfill only, never the live number), otherwise DefiLlama APY-only estimates with TVL omitted where their figure isn't comparable to ours; solid points are our own collection. Every point discloses its source on hover.
- One APY-only backfill is not marked hollow, and is disclosed here instead: PRIME on Kamino's Figure Market carries a DefiLlama-sourced supply APY for 2026-05-01 to 2026-07-08 (69 days). Kamino's own history supplied the TVL for that window but reports PRIME's LENDING rate, which is genuinely 0 for a collateral-only reserve, so the fund's native yield was simply absent until our issuer-direct feed started 2026-07-09. The backfill reads DefiLlama's hastra pool, whose apyBase comes from the same Hastra Proof-of-Reserves field we now read directly; its last backfilled value (6.469% on 07-08) sits 0.02 percentage points from our first live reading (6.449% on 07-09). TVL on those rows is untouched and remains our own.
- Short histories are labeled with their snapshot count instead of pretending to be a trend.
- ISSUER MINT-INFO TVL AND DEFILLAMA'S VAULT TVL ARE TWO DIFFERENT SCOPES and are never added. This row is the USD* token; Llama's Perena figure counts vault deposits, so summing them double-counts the same dollars.
- Utilization is 0.00% because USD* has no borrow side, not because a number is missing. It therefore cannot plot on the utilization-vs-APY scatter, exactly like Nest snUSD.
- The APY is the issuer's declared trailing 7-day figure. A missing tvl or apy is a FAULT (last-good kept), never a zero, because a zero would read as 'the product pays nothing'.
- SOL* and Smart Borrow are out of scope for this row, and the Neutral Trade Drift vault is not published at all.
- USD* looping on Jupiter Offerbook is a DIFFERENT product and a different protocol string (Jupiter Offerbook). Those looped USDC dollars are not added here.
- USD*-J and USD*-P are OTHER SHARE CLASSES of the same vault, collected from onchain VaultTrancheState under their own protocol strings. They are not extra copies of this USD* row, and DefiLlama vault TVL is still never added on top of any of them.
- THIS IS NOT THE USD* ROW. USD* is the untranched class from the issuer REST mint-info; USD*-J is a different mint. They share a vault and must never share a protocol string, because defi_yields wipes per (date, protocol).
- Jupiter AMM prices are not used. On 2026-08-19 the Jupiter mark on USD*-J was ~$1.75 against an onchain share price of ~$1.07, on $105 of AMM liquidity.
- GitBook's ~2-3x USD* multiplier is not used. REST ?class=junior is ignored by the issuer and still returns the USD* rate. The waterfall is the program's own split, labeled derived:perena_junior_waterfall.
- Onchain unstake lockup and fees are read from VaultTrancheConfig (measured 1 day / 10 bps standard / 100 bps early on 2026-08-19) and can differ from the GitBook copy.
- A failed RPC keeps last-good. A class that fails the mint or supply gate is omitted, never written as zero. A missing USD* 7-day APY withholds the junior row this pass so last-good J is kept while P still publishes.
- THIS IS NOT THE USD* ROW. Senior is USD*-P, not USD*. The SDK comment that called USD* 'senior' is wrong; the config mints are the proof.
- Jupiter has no USD*-P price. TVL is the tranche value, never an AMM mark.
- A seniorFixedApyBps of 0 withholds the rate rather than writing 0%.
- A failed RPC keeps last-good. A class that fails the mint or supply gate is omitted, never written as zero.
- THIS IS NOT JUPITER LEND AND NOT PERENA. Jupiter Lend is a pool money market; Perena USD* is hold-to-earn savings. The three share a mint and must never share a protocol string, because defi_yields wipes per (date, protocol).
- Offerbook's availablePrincipal is min(remaining, a shared per-wallet escrow). Summing remaining across a creator's offers double-counts the same USDC, so fillable depth is unique-by-creator.
- The APY int is hundredths of a percent (500 = 5.00%), proved against the same payload's USD principal. Token-metrics collateralOfferMetrics is interval flow, not stock, and is not used as TVL.
- The 5 percent on the Lending Markets row is the P2P loan rate. Looped net APY (the figure a loop UI quotes, often ~25-35 percent) is derived at read time from Perena's USD* 7-day APY and this book's borrow rate, and is shown on the leverage card, never written into defi_yields.supply_apy.
- A missing or partial book writes NO row rather than a zero.
- Only pools with nonzero utilization AND nonzero supply APY plot: a log axis cannot place zero, and mint-only legs (e.g. GHO), fixed-yield, savings, and collateral-only listings have no meaningful position on this map; they live in their own cards.
- Exclusion from this chart is a chart constraint, not a data gap; the Lending Markets table shows every tracked pool.
- DEX LP rows (Raydium, Orca, Aero, Uniswap, Nest Exchange, and any other venue we collect with a usable LP APR) are merged at read time from dex_native_pools, not written into defi_yields. They do not plot on the scatter (no borrow utilization). Estimated APR is trading-fee APR plus incentive APR where paid; Nest Exchange is emissions only, because swap fees accrue to veNEST. Distinct from Nest nUSD on Solana. This APR is never Best Yield Today. A four-digit fee APR is treated as a units fault and withheld, not published as a rate.
- This is liquidity standing ready to TRADE these assets, not capital earning a yield. It is not summed with the lending or fund cards: the same dollar can be an LP position here and collateral elsewhere.
- Matching is by mint because symbols are adversarial: Orca lists 49 distinct gold-symbol tokens including six different mints all claiming to be 'PAX Gold', plus names like 'goldman sachs'. A symbol match would ingest spoofs.
- Pool discovery is a periodic bounded sweep, not a daily scan: Orca's pool list runs past 120,000 pools without exhausting and is not reliably TVL-ordered at depth, so a scan that stopped early would silently miss material pools. The sweep states whether it reached the end or stopped at its cap; a newly-listed RWA pool is therefore picked up on the next sweep, not the next day.
- Reserve history is each pool's two reserves as TOKEN amounts from our daily snapshot of Orca's API (the last hourly read of each UTC day, since 2026-07-30), for the 10 largest pools by TVL over the last 90 days. It is never a reconstructed USD value, because no per-day historical price exists for these tokens; a pool whose token decimals are not recorded shows no amounts rather than raw units.
- A pool's TVL is Orca's own USDC valuation of its two reserves, so it can differ from our figure for the same asset elsewhere on the site, which prices from issuer NAV or oracles.
- DefiLlama is not a source for this card. It is used only as a counter-check against our own figures (check_dex_liquidity_llama_band.py) and is never published. It was dropped as the source after it reported a BSC Uniswap v4 QQQB-USDT pool at $379,980,394 of 24h volume on $76,167 of TVL while that pool's own Swap events summed $4,444,452.
- This is liquidity standing ready to TRADE these assets, not capital earning a yield. It is never summed with the lending or fund cards; the same dollar can be an LP position here and collateral elsewhere.
- An absent volume is not a zero. A pool whose 24h tape is not fully covered, whose swaps cannot be priced from a stablecoin leg or one of our own marks, or whose venue has no volume source yet (the summary lists those venues) shows no volume rather than $0.
- Depth is the trustworthy headline; raw volume is not. Onchain equity and gold tapes are dominated by MEV and arbitrage loops, so the card reports volume excluding pools that trade more than 100% of their own depth in a day alongside the raw figure, and holds any pool above 10,000% for review, out of every volume sum.
- Each pool carries what its other leg is: a stablecoin, a major (native asset, wrapper, BTC), another tracked RWA, or other. Other is overwhelmingly memecoins; a consumer that counts a meme-paired tape separately must exclude those pools.
- Uniswap v4 keeps every pool's tokens in one contract, so per-pool TVL is rebuilt from the pool's own liquidity events and published only where that ledger reproduces the PoolManager's active liquidity exactly. It excludes unclaimed fees and anything a hook holds outside the PoolManager. Arbitrum v4 pools are found only through standard fee tiers without hooks; hooked Arbitrum pools are not tracked.
- Matching is by chain AND address because both symbols and bare addresses are unsafe: Orca lists six different mints claiming to be 'PAX Gold', and 0x01bffโฆ1071 is XAUt0 (gold) on Monad but USDโฎ0 (a stablecoin) on Optimism.
- Kuru (Monad order book) is read from the venue's own API: its volume is the venue's rolling 24h and its depth is the two-sided book within 1% and 2% of mid; an order book holds no pool TVL, so Kuru carries depth and volume but no TVL. Not collected: Supernova. Byreal volume is the day-over-day change in each pool's own onchain swap counters (published once a reading 22-26 hours old exists); Fluid volume is summed from each Fluid pool's own Swap events. Meme-paired pools count in these totals on every venue.
- Jupiter Lend AMM pools (Solana) hold no tokens of their own: their reserves are supply positions on the Jupiter Lend liquidity layer, read onchain with interest applied. Volume is each pool's own swap events, valued on the stablecoin leg; a swap whose event was cut from a long transaction log is read from its instruction instead, and its non-stablecoin side is priced at the pool's own 24-hour average. Each of these pools is also the collateral of a Jupiter Lend vault, so the same dollar sits on the leverage card as that vault's collateral.
- Concentrated-liquidity quotes are a FLOOR, not an estimate, once a trade would leave the current tick range: liquidity outside that range is not modelled, so real slippage is worse. Every quote is flagged, and the response reports how many are bounded rather than estimated; on 2026-07-30 that was 154 of them.
- Constant-product pools (Uniswap v2 and forks) are exact, because reserves fully determine price impact there.
- Solana concentrated-liquidity pools are quoted only where the pool row carries its in-range liquidity and current tick, which today means Orca; Raydium and Byreal concentrated pools carry reserves only and are not quoted here. Their depth still appears in the DEX Liquidity card, so this endpoint covers less of the market than that one and says so.
- Routing and aggregator splitting are not modelled. Real execution routed across several pools can beat these figures, so treat them as single-pool worst-case rather than the best available price.
- This is the honest counterweight to volume: a pool can show large 24h volume and still be unable to absorb a modest sale, because MEV loops trade in both directions without ever moving size.
- ROUTER AND POOL FIGURES ARE NOT RANKED TOGETHER, even though one card now carries both. A router splits a trade across several pools, so it can legitimately beat every individual pool here. Merging them into one 'best venue' would claim a comparison the data does not support, so they stay separate blocks with their own headers.
- Router and venue quotes are sampled hourly and carry their timestamp. They are not 24-hour depth and not volume.
- An asset with no router row is not illiquid, it is unsampled by that method. Aggregators are sampled on Solana and on the EVM chains they serve; venue quotes come from Raydium on Solana, from Metric on Robinhood Chain, BSC and Ethereum, and from Uniswap on Robinhood Chain, where no aggregator quotes; Hyperliquid and Nado are quoted from their order books.
- Uniswap on Robinhood Chain is quoted with Uniswap's own onchain quoters, so each figure is the exact output of the swap at that size. Uniswap v3 quotes the best single USDG pool. Uniswap v4 covers tokens with no USDG pool of their own, such as the Ondo tokens, whose only market is an Ondo pool against the Robinhood token on the same stock: the quote goes USDG to the Robinhood token on its best v3 pool, then into the Ondo token, and the reverse to sell. Both tokens are one share each, so cost is measured against the share price, and where the Ondo pool prices its token above or below the Robinhood token that premium is part of the cost (a sell can then return more than the share price). A size no route can fill shows as no route, not as a cost.
- Metric pools price off an oracle bid and ask, not reserves, so pool math does not run on them. Their venue quote is walked from the pool's own bid/ask bins at each size, measured against the pool's mid so the half spread is part of the cost, the same basis as every other venue quote here. The walk was checked against Metric's own router on the same pools and matched it to the raw unit. Where several Metric pools hold one token on one chain, the deepest is quoted.
- A pool showing 'not valued' has exact known reserves but no USD conversion, because neither side is a stablecoin and no venue publishes a price for it.
- Fee APR is trading fees only: the pool's own 24h Swap-event volume times its fee tier, over its TVL, annualized. (It was derived from feeGrowthGlobal until 2026-09-13; that estimate read 2.4-3.3x high on Aerodrome Slipstream and 0.55-0.83x on Uniswap v3 against the Swap events.) It is NOT Merkl, AERO, or any other incentive. Those launch emissions are a different number and must not be written into defi_yields.supply_apy or this column. A null is a pool whose 24h Swap tape is not fully covered or cannot be priced, or a pool with no TVL, never a 0% rate. LP vault wrappers (Beefy CLM and the like) are a future kind, not this table.
- TAGGED FLOW (added 2026-10-18) labels each pool's last 24 hours of swaps by behaviour, read from the pools' own Swap events grouped by transaction. Atomic arbitrage is one transaction that buys and sells the same asset on two tracked pools, split by whether one of them is a Metric pool. Routed is a trade sent through a known router or aggregator (Binance Wallet, OKX, Relay, Uniswap and PancakeSwap routers, smart wallets). Automated is one contract that calls a pool directly, many times a day, from at most two wallets; it can be hedging elsewhere, so it is never called arbitrage. Everything else is untagged. 'Excl. arbitrage' removes only the atomic arbitrage: it scales the pool's published 24h volume by the share of the tagged window that was not arbitrage. Arbitrage whose second leg is in a pool we do not track, or on a centralized exchange, is not visible, so the arbitrage share is a floor. Labels describe behaviour; motive is not inferred. The card shows one row per venue and pair, the deepest pool, so a smaller pool with the same pair label (for example PancakeSwap AAPL-USDT at 0.01%) is not shown with its own tags.
- Correlation needs overlapping coverage of both series; tickers are skipped rather than padded when either side is short.
- TVL moves for custody and allocator reasons unrelated to trader positioning; a correlation here is a lead, not a mechanism.
- The short venue is picked by realized 7d average funding, not the latest print, so a one-hour spike on a thin book cannot win the ranking; Consistency is the share of the week's hourly readings where shorts collected.
- Carbon's funding legs are asymmetric: the venue dealer prices each side independently, so the ranking uses its declared long-side leg (marked on the card, both legs on hover) and the short side's actual rate can differ from the number shown.
- Isolated-market protocols (Kamino, Morpho) only allow borrowing inside the collateral's own market; rows marked 'own capital' have no stable borrow market at all, and their net carry assumes unlevered capital.
- Execution costs (spread, slippage, gas) are not modeled; this is the rate stack, not a P&L forecast.
- Issuance is the fund's outstanding token supply x NAV, not its DeFi-deployed TVL; the two are compared on the Today tab's issued-vs-deployed card.
- This measures ISSUED INSTRUMENTS, not unique underlying Treasuries: wrapper funds hold other tokenized funds (Ondo OUSG is backed largely by BUIDL and USYC), so the same Treasuries can be counted inside more than one instrument.
- NAV comes from the issuer's own onchain oracle where one exists (Ondo OUSG/USDY, verified live) or an attested issuer/rwa.xyz NAV otherwise (near-par money-market funds are exact; high-NAV funds refreshed on a cadence). Supply is always live onchain; rwa.xyz is never called at runtime - attested NAVs are manual constants carrying a per-fund as-of date.
- A few funds live partly on non-EVM/Solana rails (XRPL, Stellar, Aptos, Cosmos) not yet read, so those totals are slightly conservative.
- When a fund's supply or NAV read fails on a day, each of its chain rows carries the last value we stored (at most 7 days old) instead of dropping out of the total; such a row is a repeat of that earlier day, not a new read. A row with nothing stored inside 7 days is left out for the day (from 2026-11-14, when a failed Saturn sUSDat NAV read had removed $36.8M from the day's total).
- Bridged legs are counted once. Where a bridge locks the home-chain token and mints a copy elsewhere (sUSDat, apyUSD, XAUt0), the copy is a slice of the home supply, so a bridged chain row is carved out of the home row, never added beside it; XAUt's Ethereum row is its total supply minus the legs read on Monad, Arbitrum, Avalanche, Polygon and Solana; XAUt0 on chains not yet read stays inside that Ethereum row. Where both sides burn and mint (rwaUSDi, Re, Superstate, Etherfuse), per-chain supplies add.
- Maple's syrup pools (syrupUSDC, syrupUSDT, syrupUSDG and Maple's other pools) are issuance: each pool's own onchain total assets (cash plus loans plus accrued interest), not the pool's idle cash. syrupUSDC bridged to Monad is shown as its own chain row, carved out of the Ethereum pool. Syrup posted as collateral on Aave, Morpho or Kamino is the deployed side and is shown on the lending cards.
- Tokenized commodities (Tether Gold, XAUt) are shown beside the funds and are not part of the fund total: one token is one troy ounce, valued at the Chainlink XAU/USD price.
- Etherfuse stablebonds are local-currency bond tokens: 1 KTB is priced at about 1.02 Korean won (each token is a claim on about one won of Korean government bond face value, so well under a US cent) and 1 CETES at about 1.19 Mexican pesos, not a dollar. Each is valued at Etherfuse's own price in that currency times the day's FX rate (Yahoo), and a value more than 2% from Etherfuse's own dollar price is withheld.
- BUIDL aggregates two share classes (BUIDL + BUIDL-I), both issued via Securitize.
- USYC's BNB Chain leg (~$2.86B, ~96% of the fund) is real, verified issuance, not a collector artifact: the onchain totalSupply matches Circle's official BNB contract and rwa.xyz to the token (checked 2026-07-21). USYC is natively issued on BNB Chain and that leg is predominantly Binance institutional off-exchange collateral held via Ceffu, so it represents supply parked with one venue's custody rather than broad onchain distribution.
- Only wallets visible on both a tracked perps venue and a tracked lending protocol in the same address space qualify; the cohort is a floor, not the population.
- Basis default is traders-only; toggling changes headline USD materially on DeFi lending venues (a Morpho curator vault is capital, not a trader).
- High-frequency accounts (1,000+ trades per active UTC day on 5+ of the last 30 days) are left out of both wallet lists and the summary counts and counted in hf_wallets.
- This is a behavioral screen (collateral + borrow + perp deposit), not confirmation that the perp leg is short the same asset; direction of the perp position is not always observable.
- High-frequency accounts (1,000+ trades per active UTC day on 5+ of the last 30 days) are left out of the candidates and the summary counts and counted in hf_wallets.
- HOLDER COUNTS ARE WALLETS HOLDING AT LEAST $1 from 2026-11-16, the same floor as every other wallet count we publish; the count of wallets holding any balance is shown beside it. Before that date the count was any positive balance, and 1.57M of the 2.36M organic holders then held under $1, so the 7d and 30d holder-count change is not shown across the switch. The dollars include every organic balance (wallets under $1 hold under 0.3% of them).
- Outstanding (the market-share bar and the Spot Markets headline) is supply times the latest hourly reference price. The holder composition (organic, custody, DeFi, market maker, hub-suspect) is measured once a day at that day's prices, so when prices move during the day the issuer-reserve/residual slice absorbs the difference until the next daily measurement. The 7d and 30d views compare against the daily measurement of those days.
- EQX (EQX SPC Ltd., Hyperliquid HyperCore + HyperEVM, live 2026-10-26) is counted at its PUBLIC float: every holder except the issuer's treasury and deployer (genesis supply, not issued) and the issuer-funded distribution and market-making wallets (a wallet funded by a direct send from the treasury, the deployer or a wallet already in that set), reconciled hourly between the Hypurrscan census and Hyperliquid's own balances and withheld when they disagree. EQX's own backing API (created_supply, daily cut) is the cross-check. Its holder lots come from each wallet's own fills and spot-transfer ledger; an order-book sale is counted as Transferred. 6 of the 94 EQX wallets with holder lots trade so often on Hyperliquid that the 10,000 most recent fills it serves reach back only to dates between 2026-11-04 and 2026-11-20, after their first EQX activity, so their EQX order-book trades before those dates are not in the lots; on 2026-11-21 they held under 1% of the open EQX value. Their transfers in and out are complete, and any balance those records do not explain is counted as one open lot priced at that day's mid, so holder balances and tiers are complete while those six wallets' hold times and closed lots before those dates are a lower bound. From 2026-11-21 every fill is kept as it is read, so no further history is lost this way.
- ISSUERS ON THIS CHART ARE NOT ALL THE SAME INSTRUMENT, and an equal bar height can mean three different things. A share-backed wrapper (xStocks/Backed, Ondo GM, bStocks) is a token against a real share held in custody, with published reserve attestation. PreStocks holds shares through SPVs and the token conveys indirect economic exposure only: no ownership, voting, dividend or information rights, and the attestation reports promised at launch have not been published. Tessera is not a share claim at all: a T-Token is a LOAN PARTICIPATION RIGHT against a dedicated issuer entity, on a non-security opinion under Singapore law, explicitly NOT collateralised by the underlying shares, repayable only on a liquidity event, with total loss of principal possible. The difference between them is CREDIT RISK, not size. Related and live: in May 2026 Anthropic and OpenAI both stated that SPV transfers of their stock are void under their transfer restrictions; ANTHROPIC is 50.0% of PreStocks' outstanding and OPENAI a further 10.5%. Pre-IPO legs are priced from a single source (jup_stock) with no public listing to cross-check against, and only SPACEX has a sibling leg (xStocks SPCX) for the cross-leg price control.
- Holding stats are VALUE-weighted and cover organic holders only: protocol accounts (AMM pools, lending reserves), issuer treasuries and exchange custody are excluded. Count-weighted figures sit in hover detail because dust dominates them: the count-median open age reads 130 days while the value-weighted median is 46, and neither is wrong, they answer different questions.
- HIGH-FREQUENCY ACCOUNTS AND MARKET MAKERS ARE LEFT OUT OF HOLDING BEHAVIOUR, NEVER OUT OF HOLDINGS. A high-frequency account averaged 1,000 or more trades per active UTC day on at least 5 of the last 30 complete UTC days, and stays flagged for 30 days after it last qualified; a market maker is a perp wallet whose long and short volume over the last 45 days are within 15% of each other, active on at least 70% of those days with 50 or more trades. Their lots are excluded from hold times, open-lot age, closed-lot counts, the under-an-hour share and the disposal mix, and are still counted in holder counts, holder dollars and tiers; the number of such wallets and the high-frequency accounts' open dollars are shown beside each figure, and the API's include_hf option returns the figures with them kept in. The perp trade count is RWA markets only, while HyperCore spot holders are counted on their fills across all Hyperliquid markets. Wallets are matched by address, so an onchain holder on a chain with no trade-frequency source is not identified as high-frequency.
- Hyperliquid serves only an account's newest 10,000 fills, so a busy HyperCore spot holder's older days cannot be read. A holder whose every readable complete UTC day in the window has 1,000 or more fills is counted as a provisional high-frequency account until 5 such days are known; the 5-of-30-days rule then decides, and a provisional flag lapses at once when the wallet fails it. On 2026-11-22, 3 HyperCore spot holders were provisional high-frequency accounts.
- Sub-hour custody lots (routing and aggregator hops) are excluded from headline hold times and reported as their own percentage; a token passing through a router for minutes is not a holder.
- The 'into DeFi' exit leg is split by DESTINATION PROTOCOL: AMM/DEX venue (Raydium, Orca, Meteora, Phoenix, ...), lending protocol (Kamino, Jupiter Lend), or unclassified (an ambiguous or unknown owning program, left uncategorized rather than guessed). This is the destination VENUE, not the intent: an AMM destination can be a swap-sale OR a liquidity-provision position. The flow-based leg also UNDERCOUNTS lending, because the mint-flow ledger is missing the owner-to-lending-PDA deposit legs (the same hot-address truncation the census-clamp compensates for) - the honest held-in-lending figure comes from the holdings snapshot instead (next caveat).
- Held-in-lending: tokenized equity deposited as collateral in onchain lending, which the flow ledger barely sees. Kamino (~$31M) is attributed to the real depositors from its onchain obligation snapshot (each reserve's total supply USD shared out by each depositor's raw deposit, folded to wallet-link clusters). Jupiter Lend (~$20M, TSLA/SPY/QQQ/NVDA collateral in its Fluid borrow vaults) is included as a per-vault TOTAL from the Jupiter Lend API, but NOT yet split by depositor: its positions API is per-wallet with no bulk enumeration, so depositor attribution waits on an onchain position enumeration (tracked as a follow-up). The depositor still owns it (posted for yield), so it is a held position, not a sale.
- xStocks redemptions happen offchain (transfer to issuer, then burn off-ledger), so the redeemed leg undercounts for xStocks; Ondo GM burns are onchain and counted.
- Holder tiers are balance-based: Institutional $1M+, Whale $100K+, Medium $10K+, Small $1K+, Micro below; these are portfolio-value thresholds, distinct from the trade-size tiers used in Trader Intelligence.
- Perp comparisons use completed, non-market-maker round-trips only; positions open at a venue's coverage start are excluded rather than guessed. Perp behavior is bimodal: the median round-trip closes in minutes while averages run 18-34 hours, so both are shown.
- Dust is excluded before anything is computed: transfers under $1 never open or close a lot, and wallets whose lifetime peak never reached $10 are dropped entirely, so the Micro tier is real small holders, not airdrop residue.
- Net issuance week-over-week compares the two most recent complete onchain weeks (mints minus burns priced at the oracle); the current partial week is never shown as if it were finished.
- Backpack's omnibus custody wallet is labeled exchange custody and excluded from holder stats; its user-level composition is not observable onchain.
- Hub-suspect wallets are organic-by-elimination wallets whose shape matches an omnibus (at least 10 distinct tickers AND at least $1M); every confirmed CEX wallet showed 11+ tickers while genuine whales held 6 or fewer. They are excluded from organic float and holder stats and shown as their own slice, never folded into CEX custody, because suspicion is not identification; each is queued for a manual identification pass and graduates to the named custody list (or back to organic) once resolved.
- Anchored aStocks (Anchored Capital Ltd., from 2026-11-12) are ERC-20 tokens at one address on Ethereum, Base, Arbitrum and Monad, 1:1 backed by shares at a U.S. broker-dealer. Every token comes from one factory, read onchain on all four chains; supply is minted on a fill, so there is no issuer inventory to subtract. Dividends are paid in USDC, so balances never rebase. Holders come from the full transfer history on Monad, Base and Ethereum, proven per token (replayed balances sum to totalSupply at the census block); Anchored's own trading-account Stock contracts hold most of the Monad supply on behalf of its users and count as exchange custody, out of organic and float, like the Backpack omnibus. SCOPE: Arbitrum (about 0.1% of Anchored's value) has no holder census, so its holders are not counted and its supply is outstanding only.
- bStocks are issued by BTech Holdings Limited, a Binance Group affiliate, as BEP-20 tokens on BNB Chain, 1:1 backed by shares at a regulated custodian (labeled 'bStocks (Binance)' here). Like xStocks, all supply is minted into one custodian wallet and distributed from there, so its net issuance is reserve-net DISTRIBUTION (custodian-to-holder minus holder-to-custodian), and the custodian is excluded from float. bStock DeFi deployment is Venus core-pool collateral on BNB Chain.
- 'In exchange custody' is token an exchange holds for its users: identified centralized-exchange omnibus/hot wallets (Binance, Kraken, Gate, Bybit, VALR, Bitget) and, on Monad, the user-balance contracts of two onchain exchanges (Monday Trade's Stock contracts for Anchored, Kuru's AccountCore for xStocks). It is NOT self-custody, and it is excluded from organic and from float. Much of the tokenized-equity supply sits here (xStocks ~$187M, Ondo ~$144M, bStocks ~$159M). CAVEAT: 'organic' is defined by ELIMINATION (a wallet not matched to any known contract / CEX / reserve list), so an unlabeled exchange or treasury wallet can still be counted as organic until its address is added; known ones are excluded by address (a labels feed would harden this).
- This chart shows issuance ACTIVITY (net flow), deliberately not cumulative outstanding: cumulating it would overstate supply (xStocks distribution is bounded by its reserve-inclusive mints).
- Backpack issuance is materially undercounted pending a decoder fix: its launch-window supply mints are barely captured (the same launch-mint gap slightly undercounts other issuers' first weeks).
- Pricing uses the current reference price, not the price at mint time; the series reads as today's value of historical flow, not historical dollar volume.
- Restated 2026-08-03. Redemptions for reserve-distribution issuers were previously counted only as transfers back into the reserve, but bStocks and xStocks retire supply by BURNING it, from an address that touches no reserve. Under the old method both issuers therefore showed no redemptions at all, although bStocks had been burning supply since 2026-W24 and xStocks burned in one week (2026-W25). bStocks weekly redemptions are now non-zero from W24 onward and its net issuance falls accordingly.
- bStocks redemption is CEX-mediated, not wallet-level: supply returns via a Binance exchange hot wallet before it is burned, so a redemption here is an exchange retiring supply, not an identified end-holder exiting.
- The current partial week is excluded from week-over-week comparisons.
- HOLDER COUNTS ARE WALLETS HOLDING AT LEAST $1 from 2026-11-16, the same floor as every other wallet count we publish; the count of wallets holding any balance is shown beside it. Before that date the count was any positive balance, and 1.57M of the 2.36M organic holders then held under $1, so the 7d and 30d holder-count change is not shown across the switch. The dollars include every organic balance (wallets under $1 hold under 0.3% of them).
- Tiers are portfolio-value thresholds over ORGANIC holders: protocol accounts, smart-wallet custody, exchange custody and treasuries are classified and excluded first.
- Hold time now covers Solana lots PLUS EVM lots (13.05M lots as of 2026-07-31). Until 2026-07-31 it was Solana-only, because the EVM FIFO lot builder could not complete; figures published before that date reflect the Solana slice alone and are not directly comparable to later ones.
- bStocks QQQ is EXCLUDED from hold-time and tier statistics. Measured over its full lot history it shows 30.4M lots (70.5% of all EVM lots) across 54,247 holders, averaging 561 round-trips per holder at an average hold of 86 SECONDS, in $790 clips totalling $24.06B of cumulative position opens against a float of $25.0M, i.e. roughly 964x turnover of the entire token. Every comparable ticker sits at 1-6 days, including SPCX from the same issuer on the same chain at 1.18 days, so this is specific to that one pair and not an issuer-wide or collector artefact. These are custody statistics, and that behaviour is not custody: included, it would pull the published median toward 86 seconds on a population that is 70% one pair. This follows the same principle that already excludes AMM pools and lending reserves as the spot analog of market makers. We do NOT assert why the turnover occurs: incentivised volume programs, wash trading and a single high-frequency market maker all produce this signature. Counterparty analysis run 2026-07-31 RULES OUT direct wallet-to-wallet washing: every one of the five dominant counterparties is a CONTRACT (verified by eth_getCode: 180B, 2,820B, 20,743B, 4,084B and 4,265B of bytecode), they carry 75% of all transfers, and each passes through almost exactly as much as it receives. That is routers and pools, i.e. ordinary DEX plumbing, and it also explains the 84% reciprocity we measured (every swap is one transfer in and one out by construction). So the turnover is real and router-mediated rather than two accounts trading with each other. What drives it economically is still undetermined: identifying the beneficiaries needs transaction-level analysis of who INITIATES the swaps, which transfer logs cannot show. Other issuers' QQQ (e.g. xStocks) is unaffected and remains included.
- Short holds are common in this data generally, not only in the excluded pair. The pct_closed_under_1h column carries this explicitly rather than letting a median imply longer conviction than the distribution supports; it counts behaviour lots only, with high-frequency accounts and market makers excluded.
- Issuers with under $500k of organic value are omitted so a dust book does not take a row. Tessera is shown regardless: a tracked issuer, currently below that floor, kept on the card because the holder book is real (thousands of organic wallets) and growing.
- Issuers with under $10K of open lots show n/a rather than a number built on dust.
- Hyperliquid HyperCore xStocks holder counts are withheld: two wallets hold โ99.7% of the $1.09M float (2026-08-11), so a raw count would describe market-maker inventory, not a holder base. The MM-shaped share is stated here as scope rather than published as a headline holder number.
- Wallets whose lifetime peak never reached $10 are dropped as dust before tiering.
- Elysium pool TVL (from 2026-11-09) is measured for v2 and v3 pools from their token balances, with HYPE priced from our own stock/WHYPE pool; Uniswap v4 pools keep their tokens in the shared PoolManager, so their per-pool depth is not measured and is left out, never shown as zero.
- The issuer column is a stacked bar of each issuer's share of that ticker's tokenized spot supply in USD, on the supply-times-reference-price basis, so it is the same basis as the spot USD column and NOT the DEX liquidity column; the label under the bar is the dominant issuer by supply. A single-issuer ticker shows a plain swatch and name instead of a bar.
- AMM POOL TVL is one of three bases. Ethereum: native-mint pool census is sparse; Backed's wrapped wTICKERx Uniswap v3 pools carry the real AMM depth (measured ~$285K USDC-side across SPYx and TSLAx on 2026-08-11) and are NOT yet indexed here. Solana is as listed. Ink: only dust Velodrome pools exist (measured ~1 wei USDC per pair), so Ink AMM is untracked, not zero. Uniswap v4, Curve, Balancer and other venues are not yet read (measured at well under $100K across tickers when checked).
- CLOB BOOK DEPTH is the second basis. Hyperliquid HyperCore xStocks and EQX spot books are collected hourly in xstocks_hl_snapshots (resting bid/ask notional) and walked at $1k-$1M in the ticker picker, where EQX is its own venue row ('Hyperliquid ยท EQX'); EQX has no AMM pools, so its DEX-liquidity column reads no pool. Nado on Ink lists live wrapped xStock spot books (wSPYx, wTSLAx, wQQQx, wAAPLx, wGOOGLx, wMETAx, wNVDAx); measured 2026-08-11 at ~$1.76M bid / $1.86M ask top-10 across the 7 live books. These are order books, not AMM pools, and are published separately from pool TVL.
- UNTRACKED is the third basis: Ink AMM depth (dust only), TON, and any venue not listed above. Untracked means we have not measured it; it is never reported as zero.
- The liquidity-by-chain bar wears chain brand colors; a ticker with no discovered pools shows n/a, which means no pool found by these readers, not proof that none exists.
- Launch unit is DISTINCT token on PK (chain, protocol, token). Four.meme TokenCreate has no quote; n_24h is every launch in the window, labelled launches (all quotes), never stock n.
- 7d/30d USD needs a daily mark for every day with wei. Missing mark => that window is None, not a collect-time spot sum. volume_mark_basis is binance_daily_klines or collect_time_spot. 30d is not public on collect-time spot.
- Headline rewards + buybacks = tokenized-equity holder payouts (OTC distributed in the reward stock, priced NFT deliveries, Ember 24h holders). Protocol-token burns (STONK dailyHoldersRevenue, OTC toBuyback, Four.meme 4Stock) stay on the rewards card, not the Overall KPI. StonkFun launched-coin holders lifetime is GET /rewards distributedTokens ร Jupiter, not a tweet. Stock-quoted share denom includes StonkFun/Ember/Signal all-quote excess and Four.meme native BNB. Of spot equity uses /defi/dex-liquidity by_asset folded onto the equity book (AAPLx/AAPLon), never the all-RWA DEX summary.
- UNI/RAY mcap is market-wide, not attributable to this narrative. Fee-bearing volume by venue is. STONK/PONS/GIGA/HOOKR/UP fdv is n/a until pool mark ร supply is onchain in this fixture.
- Dormant is sidecar 24h vol < $1k (StonkFun) or stored TokenLaunched minus 24h WETH-active (Pons). DexScreener is not a source. Every dormant count has a denominator.
- Relay overlap: RH set = Pons v1/v2 buyers + Uni v4 swappers; Solana id = Relay recipient vs StonkFun creator. complete=false until the VPS K.16 join; zero from a missed join is never published.
- Chain scan (2026-09-08): Ethereum L1, Arbitrum, HyperEVM, Plasma, Ink, Avalanche, Monad scanned none. Covered: Solana, Robinhood, Base, BNB. Untracked is listed, never counted as zero.
- OpenSea floor is allowed for NFT inventory pricing only. Anvil fill volume is vault NFTBought/NFTSold, never OpenSea. Clock In CreditAccrued is not volume.
- LIT is Lighter, not LITE (Lumentum). Do not alias LIT to LITE. DJT has no verified mark (Parqet returned DWAC).
- Uniswap v4 tvl_usd is always null. GIGA fee-center USDG is inventory, not 24h revenue. Take-rate is measured rev/vol else documented bps (Pons 30, Flap/Four.meme 100). Named no-tape brands (up., Delta, Hookr, pools.trade, Fables) collapse to one row on the tab.
- Incomplete windows render as partial (dashed), never as a finished 30d. n/a has a hover reason. Day-1 keep-set (KUMA, DOGCOW, MOMO, MARSCOIN) is a sample, not desk-wide retention.
- DexScreener, GMGN, and stonk.fyi unofficial fdv are not published numbers. Llama pons-v2 is a named bound when the onchain curve walk is incomplete, never a silent substitute.
- Elysium (meta elysium_windows, built hourly by collect_elysium.py from its own stored tape, 0 extra RPC calls): pools where one leg is an Elysium tokenized stock and the other a launched coin (not USDC, USDT0, WHYPE or HYPE), split by venue (kStock, Lumora, BOXA, Uniswap v4 hookless and the rest). USD is the stock leg x the reference price at collection. A window the tape does not cover is n/a with its partial sum named as a lower bound, never 0. Buyers are not measured on Elysium (a v4 Swap names the router). Lumora's test coins are excluded only where the coin name is stored; until then they are included and counted in the blob.
- On Elysium (from 2026-11-09) the stock tokens are canonical-bridge copies of the HyperEVM wrapped xStocks and EQX tokens, matched by contract address; venues are resolved per pool from its PoolManager, hook or factory (BOXA, Uniswap v4, kStock, Lumora, Homer, Eleos, Stockium, Kleos, Ramses, Helios, Agora, Zyora, Merklewood, Nest, Aero, Metric). Volume is Swap events only, so a liquidity manager's mint and burn churn never counts; a hooked pool on Uniswap's shared PoolManager belongs to the hook's owner, never to Uniswap. Elysium 7d and 30d are summed from our own Elysium swap tape and read n/a until a window is fully covered.
- A pool-day whose swap volume is more than 10,000 times the pool's TVL is left out of every figure here: volume farmed through near-empty liquidity (2026-10-18: a near-zero-fee BSC AAPL pool with about $11 of TVL printed $17.0M of balanced swaps). The swaps stay stored and labelled. Metric's oracle-priced pools are exempt, and a pool with no TVL reading cannot be tested.
- Excl. arbitrage is shown for a venue only when pools carrying at least 95% of its volume are tagged; otherwise it reads n/a with the reason.
- The arbitrage share is a floor: a second leg in a pool we do not track, or on a centralized exchange, is invisible, so that trade stays in.
- Arbitrage tags on EVM chains come from every swap (from 2026-10-18 for the per-swap rows). On Solana they are a sampled estimate (from 2026-10-25): each sampled or large swap is labelled, and a swap where the stock moved between two pools in one transaction with no end wallet counts as arbitrage, in any pool, so it is not limited to tracked pools as the EVM test is; a route that uses the stock as a middle hop reads the same way. Session views split arbitrage on EVM only.
- Aggregators (Jupiter, KyberSwap, Binance Wallet) route through these pools and are not counted as venues here, so nothing is double-counted.
- Meme-paired stock pools count toward their venue's volume.
- On HyperEVM (from 2026-10-27) the stock tokens are the wrapped xStocks and EQX's linked tokens, matched by contract address. Venues: Nest, KittenSwap, Project X, Ramses (CL and v2), UltraSolid and HyperSwap, found from the stock tokens' own transfers; the 7d and 30d history before that date was read back from each pool's swap events.
- All (the default) covers stocks, ETFs and commodity tokens; Stocks only keeps single stocks, leaving ETFs and commodity tokens out. Stocks only covers all hours: a trading-session view is for All.
- A trading-session view (US regular hours, outside US hours, pre-market, after-hours, overnight, US closed, or the Tokyo, Hong Kong and Seoul opens) sums only the swaps inside that session, by each swap's own time and each exchange's calendar (holidays and early closes included). It covers 24h and 7d on chains with per-swap records; Solana is listed without a figure until its swaps are timed.
- Split (the default) draws one slice per venue version as each venue reports it. Combined draws one slice per protocol on each chain (Uniswap v2, v3 and v4; PancakeSwap v3 and Infinity; Aerodrome Slipstream and vAMM; Ramses CL and v2), its versions' volumes added; the chain's total does not change. Elysium's own venues are never merged with another chain's.
- A pool-day whose swap volume is more than 10,000 times the pool's TVL is left out of every figure here: volume farmed through near-empty liquidity (2026-10-18: a near-zero-fee BSC AAPL pool with about $11 of TVL printed $17.0M of balanced swaps). The swaps stay stored and labelled. Metric's oracle-priced pools are exempt, and a pool with no TVL reading cannot be tested.
- BSC, Base, Ethereum, Robinhood Chain, Arbitrum and Hyperliquid's HyperEVM are read from every swap. Solana's DEXs publish volume but not buy and sell sides, so Solana is estimated: every trade above a per-pool size is read from the chain and smaller ones from an hourly sample, scaled to each pool's volume, with a 95% range. A stock's Solana estimate joins its row (24h and 7d) when its bought and sold ranges are within 25% of the estimate; where the net could be either sign inside Solana's 95% buy-share range (the range applied to Solana's volume, on each basis), the net shows the best estimate, with its range in the hover. Swap counts are counted swaps on EVM chains; on Solana they are estimated (every large trade counted, smaller swaps scaled up from the sample).
- A chain filter shows each stock's buys, sells, net and swaps on that chain alone; the list holds the top 100 stocks plus the top 30 of every chain.
- All (the default) lists stocks, ETFs and the commodity tokens PAXG and XAUt (two issuers, kept apart; XAUt0 counts as XAUt, under Commodities). Stocks only lists single stocks.
- A swap in a pool where both tokens are stocks is not counted for either (which stock was demanded is ambiguous).
- Excl. arbitrage removes atomic arbitrage only; routed, automated and untagged flow stay in.
- 7d sums complete UTC days only and says how many days it covers; 30d unlocks once 30 complete days are collected (Solana per stock over 30 days once 30 days of its pool volume are stored).
- Hovering a stock shows each chain's share of its buys and, under each chain, each DEX's share of that stock's pool volume there (all flow, from the venue table).
- A trading-session view counts only the swaps inside that session, by each swap's own time.
- This is the pool's other token, not the wallet's starting asset: a route that starts in ETH and passes through a stablecoin pool reads as a stablecoin buy.
- Metric pools are classified by their quote token's symbol.
- Venues whose stock pools carry no fee figure are listed as 'not collected', never as zero.
- A trading-session view estimates each venue's session fees as its in-session volume at the venue's 24h average fee rate, for 24h and 7d; the top-pools list covers all hours.
- The top-pools list holds the top 25 pools by the window's fees plus the top 20 of every chain, so a chain filter shows that chain's leaders.
- Incentive emissions (AERO, Merkl and the like) are not included; on Aero, staked positions earn emissions while their fees go to voters.
- A pool with no covered 24h swap tape or no TVL shows n/a.
- Pools under $1,000 of TVL are left out and a fee APR above 1,000% shows as 1,000% (the house LP-rate floor and cap). A pool the venue does not name is labelled from its tokens: the tracked stock and the other leg's symbol, or 'other token'.
- Where the venue's own reading gives no TVL (most often Robinhood Uniswap v4 pools), TVL is valued from our liquidity-position ledger at the last trade of the latest replayed day: each position's token amounts at that price, stablecoins at $1 and stocks at their stablecoin-pool price on the same chain. Those values are a day old and marked as such.
- Combined, By platform (the default) merges a protocol's versions (Uniswap v2, v3 and v4; PancakeSwap v3 and Infinity; Aerodrome Slipstream and vAMM; Ramses CL and v2) on every chain shown, after the chain filter; By chain merges them on the same chain; Split lists each venue version as the venue reports it. Fees and TVL are added, and fee APR is recomputed from the added fees and TVL, never averaged. A merged cell is the sum of what is stored, and its hover names any version that is $0 because it had no trades in the window, any version whose chain has only part of the window collected so far (with the days collected; its fees are annualised over those days), and any version left out because its fees are not collected (the fee APR is then withheld). Elysium's own venues are never merged with another chain's.
- A position held through a position manager we could not trace to a wallet is shown as a custody contract, not spread across wallets.
- Fees that no known position can claim (liquidity added before our history, or a swap whose price we could not place) are shown as not attributable rather than given to the positions we do know.
- One address on two chains counts as one earner in the all-chains bar. New means first seen earning since the first day we replayed, not first ever.
- 7d and 30d sum the fees over that many consecutive replayed days (each unlocks once those days exist); in them, new means first earned inside the window, and wallet size is the liquidity at the last day's last trade.
- Solana (Raydium, Orca, Meteora, Byreal): every liquidity position in the stock pools carrying 99% of Solana stock-pool fees is read once a day at the start of the UTC day. A concentrated position earned its liquidity times the change in the pool's fee growth inside its price range between the day's start and end, the pool's own accounting; a Meteora DLMM position the same per price bin; a constant-product pool's fees go pro rata to LP-token balances at the day's end. The pool's growth is the LPs' share, so wallet amounts are scaled up by the venue's protocol share read from the pool, and the day's total is the venue's own 24h fees. The wallet is the holder of the position NFT, the position's owner on Meteora, or the LP-token holder; a program-controlled holder (a vault or a locker) is shown as a custody contract.
- Solana limits: a position opened and closed inside the same day, fees collected between the two readings, and fees in pools outside the 99% set are not attributable and show as such; a position touched more than once in a day is an estimate. Solana days appear once the other chains' day is in. Solana wallets have no wallet type yet and read as unclassified.
- Before a day's Solana split exists, Solana is listed with the fees its stock pools generated that day (each pool's 24h fees at its venue's last reading of the day, pools under $1,000 of TVL left out) and no split; those fees are not in the all-traced-chains bar.
- Commodity-token pools (PAXG, XAUt) are not included yet, and this card has no Stocks only view.
- Pairs held for review, fiat currency pairs (such as EUR/USD) and stablecoins (such as RLUSD and EURC) are left out of every figure, and Bitget's own stock tokens count from their first daily turnover read and are absent from the hourly view.
- This is spot only and is never added to the perpetuals volume shown on other pages.
- A window needs every pair's every day: an exchange with a pair-day not collected inside the window shows n/a with the count, never a partial week.
- Exchanges not listed here are not read by this tab; they are not counted as zero.
- Silver, oil and other commodities trade on these exchanges almost entirely as perpetuals, which are on the perpetuals pages; their spot tokens are small here.
- A pool-day trading more than 10,000 times its TVL is left out of the onchain figures, as on the DEX Deep Dive (Metric's oracle-priced pools excepted).
- An asset not listed on these exchanges, or with no tracked onchain pool, says so instead of showing $0.
- A pool's last reading of a day can fall up to about 40 minutes before midnight UTC, so its day is shifted by that much.
- A pool-day trading more than 10,000 times its TVL is left out (Metric excepted), as on the DEX Deep Dive.
- A token bought on an exchange and later withdrawn onchain appears in both.
- Bitget's own stock tokens are left out of the session split: Bitget publishes only a rolling 24-hour turnover for them, and their candles carry the listed company's volume, not Bitget's (measured 2026-11-11). Bitget's other pairs are split.
- The exchange filter shows one exchange's sessions and weeks on the same basis.
- Weekend volume on 24/7 exchanges is real trading, not a gap.
- OKX and Kraken list xStocks our registry does not carry yet; they are credited to xStocks by the issuer's own published list. A token in neither list reads as Issuer not identified, never folded into another issuer.
- Issuer is decided by the token's own naming and contract where the exchange publishes it, never by the ticker alone.
- Not shown on the tab: most exchanges publish no trade count, and the site's live trade tables key spot and perpetual trades of one asset together, so they cannot supply one. Kept in the payload as a parked surface.
- Exchanges count trades differently, so compare an exchange with itself over time rather than exchanges by trades.
- Where the order book we fetched holds less than the size within 10% of the mid price, the cell says so instead of a number; orders resting further away are not counted as depth, and a size is compared only across exchanges whose fetched book reaches it.
- Each exchange serves a limited number of book levels (Bybit 200, Binance as requested), so a deep book can read as not reaching a size it would fill.
- Realized cost (what traders actually paid) needs an exchange fill tape with the mid at each fill; no exchange publishes one, so it is not shown.
- Counted at identified exchange addresses only: an exchange wallet we have not identified reads as an ordinary holder, and large unidentified hub wallets are kept apart, never assigned to an exchange.
- Moves between one exchange's own addresses, and mints, burns, redemptions to the issuer's burn wallet and issuer-reserve transfers, are left out: those are issuance, not wallets moving tokens on or off an exchange.
- A token no tracked exchange lists has no exchange close; it is valued at the platform's reference price for its underlying that day, taken oracle first (Pyth, directly or through Hyperliquid), then a perpetual mark across venues, then the token's onchain DEX price or the stock's exchange last close, and the cell counts how many tokens were valued that way. A token with no price at all that day is counted in transfers but not valued, and the cell says so.
- Solana deposits and withdrawals are read from each custody token account's own transactions, every account once a day, and checked against the account's balance (each transaction must start where the last one ended, and the last must match the live balance). An exchange's Solana figure appears for a window only when every one of its accounts was walked for every day of it; until then its net change from balance readings is shown instead.
- Moves to or from large unidentified wallets (holding 10 or more tickers and $1M or more, the census's hub rule) are shown apart and never counted as user flows: an exchange's own unlabelled wallets look like this, and so does a market maker.
- Coinbase lists no tokenized stock on its exchange, so it has no stock custody flow. Backpack issues its own stock tokens and keeps them off chain: it mints a token when a user withdraws and burns it when it comes back, so for Backpack a burn is a deposit and a mint is a withdrawal. Its deposits and withdrawals are read from every mint and burn of its tokens, each signed by that token's own mint authority, and checked every day against the tokens' total supply (the supply at the last reading plus every mint and burn since must equal the supply now). A holder burning its own tokens without Backpack lowers the supply too: burns not signed by Backpack, up to 0.1% of that token's mints and burns in the period (at least 0.05 tokens), are kept apart and not counted as deposits; anything else that does not reconcile keeps the days out. Before every day of a window is read, Backpack's change is minus the change in total supply from daily readings.
- Venue-issued tokens have no onchain token and so no custody flow.
- Spot is divided by the share multiplier only where the exchange publishes one for its own pair (Bybit, for its xStocks pairs). Everywhere else the figure is the raw price gap, never adjusted with an assumed multiplier: a steady gap of a few tenths of a percent there can be the issuer's dividend multiplier rather than a mispricing. Kraken's xStocks perpetuals are on the token itself, so no multiplier applies there.
- A perpetual is paired with a spot market by name only when their prices agree: a pair whose median gap is beyond 10% is two different instruments and is dropped.
- A pair with no paired closes in the window is not shown, and the count is stated.
Provenance classes
Every directional or derived figure carries one of these classes; blended skews only mix observed venues.
| observed | Read directly from the venue's own data (fills, taker flags, position changes). |
| inferred | Derived from a related observable (e.g. OI deltas standing in for flow). |
| proxy | A substitute signal used where the target is unobservable. |
| backfilled | Reconstructed after the fact from a historical source; flagged per point. |
| derived | Computed from other stored metrics rather than collected. |
| unavailable | Structurally unobservable for this venue; excluded from blended figures. |
Wallet size tiers
Wallets are tiered by average daily traded volume (30-day window รท 30).
| Tier | Min avg daily volume |
|---|---|
| Institutional | $100,000 |
| Whale | $10,000 |
| Medium | $1,000 |
| Small | $100 |
| Micro | $0 |
Asset taxonomies
Two classification systems run in parallel because they describe different objects. rwa_utils.get_rwa_type() buckets a TRADABLE instrument (perp/spot) by its underlying asset class; defi_categories.classify() buckets a DeFi pool/position by its underlying EXPOSURE, not by protocol. Equities and Commodities are shared buckets: a token in either means the same asset class on both sides, but the DeFi side holds only tokenized metals, since no tokenized energy or agricultural product is tracked in DeFi. The rest are domain-specific: FX and Indices exist only for tradable instruments, while Stablecoins, Treasuries, Credit, and Other exist only in DeFi (there is no tokenized-treasury or stablecoin on the trading side, and no FX/index perp analogue among tokenized yield products).
Trading: rwa_utils.get_rwa_type(symbol) matches the cleaned base ticker against curated per-class sets, defaulting to Equities. DeFi: defi_categories.classify(protocol, symbol) applies protocol-level overrides first (Maple, USD.ai -> Credit), then matches every token in the pool symbol against curated exposure sets (first match wins in Treasuries, Credit, Commodities, Other, Equities, Stablecoins order), then a 'โฆon'/'โฆX' equity-suffix heuristic, defaulting to Other.
Permanent gaps
Structural limits of the underlying venues: these cannot be backfilled or engineered around, and the dashboard labels them rather than estimating through them.
- Pre-capture history cannot be reconstructed for venues without a historical API: Pacifica (40-row cap), Flash Trade, Variational (RFQ), and CEX taker streams begin at their listed live dates. (Phoenix left this list 2026-07-10: its full seat history was walked onchain back to 2026-05-01.)
- Hyperliquid builder-dexes km/cash/vntl/flx publish no order-book or quote data on any public endpoint; spread/depth/slippage are structurally unavailable for those symbols (volume and trades are unaffected).
- Extended Exchange records both legs of every fill without an aggressor flag; its venue-level long/short is reconstructed from the REST taker ratio, not per-fill observation.
- Variational direction is unobservable (RFQ, no public fills); its per-side OI is a positioning signal and is excluded from blended taker-flow skews.
- Intra-hour round-trips are invisible to GM Trade's position-delta method (open+close within one snapshot interval nets to zero); venue totals always come from the venue's own API.
- Wallet-attributed trade counts are SAMPLES on sampled venues and are always labeled as such next to the candle-derived venue total.
- Bitget, OKX and Bybit rows before 2026-08-05 are the venue's rolling trailing-24h ticker figures stamped into a UTC-day row, not day buckets; from 2026-08-05 they are true day buckets. Binance's pre-cutover history was restated to the day basis on 2026-08-08, but these three cannot be: their historical candle endpoints publish no per-candle trade count (Binance's do), and volume and trades on those rows both come from the same rolling snapshot, so restating volume alone would break the internal consistency the rows currently have. They are therefore left on the rolling basis deliberately, and a single pre-2026-08-05 day on those three venues is not directly comparable to a single day after it.
- Securitize's SECZ carries $191.3M of onchain supply, 99.4% of it on Avalanche, and no DEX liquidity figure. Avalanche is readable and was read directly on 2026-08-08: the token is a paused, permissioned Securitize DS Protocol DSToken whose investor registry holds 19 wallets, none of them contracts, and whose entire transfer history is 19 mints from the zero address with zero secondary transfers. The supply is held, not traded, so there is no pool for a liquidity reader to find and none could legally hold it while transfers are blocked. We therefore do not run DEX pool readers on Avalanche for SECZ; its liquidity cell is blank rather than zero. A daily onchain check watches the two events that would reverse this - the issuer unpausing the token, or a contract joining the registry - and we begin probing when either fires.
- GM Trade's market API reports zero open interest on its RWA markets (equities, FX, metals) outside trading hours, typically Friday through Saturday, while real positions remain open onchain. Measured 2026-08-08: the venue reported $0.42M against 784 positions worth $31.1M in our own onchain census, while its crypto book agreed to within 1% over the same hour. On those rows only, open interest is our onchain position census rather than the venue's figure, and the row is tagged as such; a non-zero venue figure is never overridden, and the substitution is refused outright if our own snapshot is stale.
- Holder counts and holder value are two different measurements and we publish them on two different bases, deliberately. The per-asset holder count is the weekly Solana onchain census: distinct token accounts with a non-zero balance, no floor. Holder VALUE on the EVM legs is reconstructed from the full transfer history and carries a 1e-6 share dust floor, so a wallet holding less than a millionth of a share is excluded from the value ledger. That floor is kept because it removes airdrop residue from every dollar figure without moving the dollars, and it is the reason our EVM holder counts run below a block explorer's: measured 2026-08-19 on Ondo GM's Ethereum book, an explorer reports 2,207 holders for the Apple token and 4,281 for Nvidia, of which 1,112 and 2,726 hold more than the floor, against our 1,115 and 2,700. The two bases are never added together into a single holders number.
Incident log
Data-quality issues we found, root-caused, and repaired, published because trusting a dashboard requires knowing how it fails and what was done about it.
| Date | Incident | What happened / fix |
|---|---|---|
| 2026-07-13 | Phoenix direction restored via onchain position-diff | The byte-parity fill direction nulled on 2026-07-11 is now replaced by a real signal. Phoenix Rise is closed-source, so the per-trader account struct was locked against the authoritative public SDK (github.com/Ellipsis-Labs/rise-public, const_assert-checked): a 224B header (authority wallet @72), then a len-bounded packed array of 40B position entries: asset_id u64@+0, base_lot_position i64@+8 whose SIGN is the direction. An earlier research pass had brute-forced the wrong offset (reading virtual_quote/funding fields) and mis-declared it 'solved'; re-deriving from source fixed it. The hourly gPA snapshot-diff engine (same as GM Trade) now decodes every Trader account on both programs, and each wallet's signed-position change feeds the long/short split as observed:position_delta. Validated on a fresh gPA and cross-checked against our own fill history (96% wallet/symbol match). Because Phoenix is a spline-AMM, trader positions don't sum to venue OI (the reconciliation ratio is ~129x, expected, not a defect); the delta captures active trader flow while idle/stale accounts contribute nothing. Volume/wallet identity still come from the fill walk; only direction changed source. |
| 2026-07-11 | Phoenix sentinel sizes + invalid byte-parity direction | The Phoenix decoder's trade-size heuristic (max u64 in the program-return blob) swallowed u64::MAX sentinel slots on the equity program: 508 fills recorded at $18.4T each (2^64/1e6), briefly showing ~$1,399T of Phoenix volume in Flow Overview. The same audit found the raw[8] byte-parity direction read produced zero shorts across 221K directed rows on both programs - direction was never valid. Fix: sentinel slots and >$50M single-fill values excluded from size extraction; direction gated to NULL until ground-truth validation lands (volume and wallet identity remain attributed); poisoned sizes zeroed (return blobs carry no recoverable size), all Phoenix directions nulled, wallet_daily_stats rebuilt for affected dates. |
| 2026-07-10 | Phoenix wallet identities were pseudo-IDs | The byte offset read as the trader's wallet on Phoenix seat accounts was not a public key at all (~6% ed25519 validity vs 99.6% for real wallets); every Phoenix wallet since capture began was a stable pseudo-ID that could not be cross-referenced onchain, and fee-payer attribution mislabeled relayer-submitted fills. Root-caused via curve validation; the real identity is the seat's registered signing key, which equals the fill instruction's signer account. Decoder and backfill fixed the same day; all historical rows re-attributed from the chain. |
| 2026-07-09 | Hyperliquid users[] mis-attribution | HL trade payloads carry users = [buyer, seller], not [maker, taker] as assumed since 06-28. Every buy-aggressor fill had attributed taker flow to the seller. Caught by cross-validating 0xArchive HIP-3 crossed flags (0.0% match on longs / 99.7% on shorts); writer fixed side-dependently and 966K rows repaired in place. Venue-level long/short was never affected (side-based). |
| 2026-07-08 | Lighter ~240x trade undercount | The per-wallet log scanner sampled ~0.4% of fills; replaced with the native market feed (aggressor flag included) and a 2.26M-leg backfill. |
| 2026-07-06 | Nado ~46x undercount | Archive API ignored min_submission_idx and re-sampled the same ~100 events; fixed with backward-index pagination and a 1.46M-row backfill. |
| 2026-07-03 | HL builder-DEX velocity overwrite | Same-symbol clones across HL builder dexes overwrote each other's daily velocity rows (last-clone-wins); fixed by aggregating before upsert, 4,649 rows repaired from candles. |
| 2026-07-02 | Extended Exchange dropped legs | ~94% of trade legs were dropped by registry holes and a cursor-skip bug; fixed and re-backfilled from 2026-04-30 (~54K wallets). |
| 2026-07-01 | directional_bias inflation | Additive upserts double-counted re-polled fills for per-trade venues (up to 4.9x); directional stats moved to tx-hash-deduped per-fill rows. |