How upswap measures cross-chain bridges v1.0 draft

Every 30 minutes, at :05 and :35 past the hour. A run requests every route × amount from every source once; missed slots are never back-filled, and gaps over 45 minutes are marked on every chart and in the log.

A full run makes about 1,218 HTTP requests and returns about 2,195 quotes in roughly two minutes. Each source has its own request interval and local rate limit so we stay inside public API limits. A run that finishes with only some sources is kept and labelled; it is never merged with another run.

Not integrated yet: LayerZero Value Transfer API（含 Stargate）, Squid Router, Rango Exchange, Stargate API (legacy), Bungee API v1.

Three fixed dollar amounts per route — $20, $200 and $2,000 — converted to token units with one CoinGecko price snapshot taken at the start of each run.

The same snapshot values both what goes in and what comes out, so quotes from different APIs in one run are directly comparable. Requests use placeholder addresses; nothing is signed or sent on-chain. Each raw API response is archived with its sha256 hash so any number can be traced back.

Cost is what you lose between input and net amount received, in basis points of the input, excluding gas.

net_usd = amount_out_usd − fee_extra_usd cost_excl_gas_bps = (amount_in_usd − net_usd) / amount_in_usd × 10,000

bps

1 basis point = 0.01 % of the input. On $200, 1 bps = $0.02.

fee_extra_usd

Fixed fees charged on top rather than deducted from the output (e.g. deBridge fixFee, LI.FI fees not included in the output).

Aggregator platform fee

LI.FI and Socket / Bungee public endpoints include 25 / 20 bps in their quotes; it is part of the cost, not added again.

Gas is always an estimate, shown in its own column and never part of the ranking. Since 2026-10-24 every bridge on a route shows the same figure: one source-chain gas estimate per run, source chain and input token (upswap gas basis v1). Across Protocol gas is never counted.

How the basis is computed: in each run we take every source-chain gas estimate (in USD, above zero) that any quote API returned for routes starting on the same chain with the same token — all routes and all three amounts, Across rows excluded. We take the median per API, then the median of those per-API medians, so each API counts once however many rows it returned. Today LI.FI and Relay return estimates (Socket / Bungee did until it was paused); deBridge, Mayan, NEAR Intents 1Click, Symbiosis and the Circle CCTP fee schedule do not, and now get the same basis instead of “n/a”. If no API returned an estimate for that chain and token in a run, the median of the previous 7 days is used and marked.

Why one basis: the APIs estimate gas differently (different contracts, approvals, simulation addresses), so their own figures are not comparable and would reward whichever API reports the lowest number. A single basis per route keeps gas comparable and makes clear it does not decide the ranking. Each API’s own estimate is still shown in small type next to the basis and kept in the raw data and CSV (gas_usd); the basis is in the CSV column gas_est_usd.

Across Protocol gas is never counted (decided 2026-10-23): its API simulates the transaction from the placeholder address, so the estimate depends on that address’s balance and approvals. When the simulation succeeds the originGas figure is archived with the raw response and flagged “gas est. archived”, but it is not shown as a cost, and Across rows (including Across reached through LI.FI) do not get the gas basis.

By net amount received, highest first, within one run, route and amount. The same bridge reached through different APIs appears as separate rows, which is how aggregator markup becomes visible.

Circle CCTP rows come from the published fee schedule, not a live quote; they are shown as reference rows and never ranked. Bridge review pages give separate metrics with their own sample sizes — there is no combined score and no star rating.

They are listed but never ranked. In run #538 there were 6.

A quote whose net amount exceeds the input (negative cost) usually means a pricing error, a temporary subsidy or an amount that will be adjusted at execution. Such rows sit in a separate group at the end of the table with rank “—” and the label “Quote above input · unverified”. They are excluded from best/second markers, key findings, meta descriptions, the home board, bridge win counts and medians, history lines and aggregator medians. They stay in the CSV with negative_cost = 1. The rule is lifted for a route type only after a real-money transfer confirms such an outcome, recorded in the changelog.

Only when the data behind it is deep enough. Pages below a gate are not indexed; the preview site shows them with a “Not enough data” label.

Charts cover the last 7 days, one point per run, with runs in order on the x axis. Amber hatched bars mark gaps longer than 45 minutes; lines never bridge a gap.

Cost axes use a log scale because bridges differ by two orders of magnitude. With fewer than 14 UTC days of history every chart carries “Not enough data · N/14 days” and no trend is claimed. All charts are rendered at build time as inline SVG: no scripts, no external requests.

Not yet — tests are planned after launch.

Each test will be done by hand through the bridge’s official front end and will list the route, wallet address, input, quoted output, approval / source / destination tx hashes, block times and what actually arrived. Real cost = input − received + source gas + approval gas. No example wallets or transactions are shown as if they were real.

Yes, under CC BY 4.0: reuse freely, credit “Arnold Burn / upswap.io” and the snapshot run.

Burn, A. / upswap.io (2026). {page title}, snapshot #{N}, {UTC time}. {URL}#{anchor}. CC BY 4.0. Accessed {date}.

Every route page links a 7-day history CSV; all quotes of the latest run are one CSV. A monthly Zenodo deposit with a DOI is planned; until it exists every DOI on the site is marked Example / pending.

No. There is no swap button, no referral or affiliate parameter and no paid placement.

Across Protocol’s API requires an integrator ID; ours is 0x03db and appears in the calldata Across returns. The app fee is 0, we never execute the returned transaction, and we earn nothing from it. All disclosures →