[Flotapponnier]

terminal-fill-quality and terminal-execution-quality both rank a provider named Binance. The row measures Binance Wallet's swap router — one address (0xb300…028d) on BNB, Ethereum and Base, read like every other on-chain trading app in the cohort, next to Phantom's in-wallet swap.

On the page this is clear: the provider carries a tag, Binance Wallet swap, BNB + Ethereum + Base, beside its name. Every other surface drops the tag and keeps the name. And the row currently leads both boards, so what ships is:

Binance posts the lowest cost per swap at 172 bps (p50, 24h) across 12 ranked providers.

in the citable API, the share card, and anything consuming the headline. A reader concludes the exchange is the cheapest venue measured, on a bench titled "what a swap costs on Axiom, GMGN, FOMO, Photon, Trojan". The measurement supports no such claim.

Renames the display name on both specs and in the harness product map. The slug stays binance: the product page deliberately covers the exchange's RPC and its wallet together, which is an argument about the slug and never was one about the name. The tag drops the words the name now carries.

What this is not

Reported as "we're still wrong on the execution cost benchmark". After auditing it, the cost is not wrong:

• 165 bps is unremarkable for a wallet router charging ~50 bps on the pairs it trades. Phantom charges 85 bps on its own.
• The 1 bp terminal fee on the Base row is already identified in native.go as fee-free Binance Alpha round trips, filtered with DropLoops, and documented in the row's note.
• The sample floors hold. tfq_ranked uses effective ≥ 100 on pooled rows (phantom 82 unranked, binance 137 ranked, fomo 160 ranked) and the size buckets use their own documented 20/40 floors, stated in the methodology.
• The hold logic is more careful than it looks: a pooled row waits when no chain row publishes on its own, waits again when the excluded chains carry most of the flow (the comment names Banana Gun's Ethereum buys), and a hold on a row propagates to its buckets.

The 20 tfq_health=0 series are all rows under the publish floor — the gate working, not an outage.

Left open, deliberately not touched here

• No score_scope: contested_chains, so the All-chains view ranks products that never meet: on Solana the leader is pump.fun at 213 bps and this row is absent. That is a disclosed design choice and a discussion, not a patch.
• A chain="funding" label exists in the metrics without being a declared dimension value. Unpublished, cosmetic.

Verified

pnpm validate 235 specs, pnpm test 430 pass / 0 fail, go build and the harness suite clean, gofmt clean.