[Flotapponnier]

Found while checking whether OKX could be measured on BNB. It cannot, for a
reason that turned out to be a defect in the column itself rather than in OKX.

What the BNB column publishes today

head_lag_seconds on bnb is receiveTime minus the provider's own
timestamp, because bnb is in neither referenceChains nor raceChains. The
spec justified that:

Reference on BNB Chain and Robinhood Chain: [...] their chain-supplied
timestamps sit within roughly 60 ms of the moment a trade is observable, so
they keep the on-chain timestamp as the zero point.

The gap between that series and the reference-matched one is that quantity.
It measures 569 ms, not 60.

What that does to the column

The dominant term is BNB's block time, which every provider shares, so the
column barely separates them. 24 h, 8,604 samples per provider, eu-west p50:

provider	published today	against the reference
mobula	0.621 s	0.052 s
serialized	0.657 s	0.102 s
codex	0.676 s	0.215 s
geckoterminal	33.881 s	31.196 s

The three push feeds sit within 9% of each other today and spread over a
factor of four against a common clock. p90 moves the same way, 1.0-1.2 s to
0.12-0.33 s.

The ranking is unchanged. mobula, then serialized, then codex, before and
after. This removes an offset they all carry; it does not reshuffle anyone.

Does it clear the gate

pending_match.go admits a chain only "once its reference is measured to arrive
before every provider". Share of trades where a provider beats our node:

	bnb	base (already shipping on the reference)
mobula	6.6%	4.2%
serialized	4.7%	3.1%
codex	3.7%	0.0%
geckoterminal	0.3%	—

Higher than Base, same order of magnitude, and Base ships at 4.2%. So this
clears the bar the project already applies to itself rather than a new one
invented for the occasion.

Scope

The reference subscription already exists and already populates
head_lag_ref_seconds on bnb. Nothing new is collected. The only change is
which series the headline reads.

What a reviewer should weigh

The published bnb numbers will drop by roughly 5-10x the day this deploys.
That is a methodology correction, not providers getting faster, and anyone
watching the page will see a step change. The spec now says so in the
methodology list where it used to carry the 60 ms claim. If that step should be
announced somewhere else too, say where.

TestReferenceChainsScope asserted bnb must NOT be here. I updated it
rather than deleted it, with the reason in the comment, and added
reference_chains_test.go pinning why bnb belongs now and why robinhood and
solana still do not. If you disagree with the switch, that test is the thing to
argue with.

Robinhood is untouched and still reads the provider timestamp. It has no
measurement yet, so I did not guess.

Follow-on

This unblocks OKX on BNB (#2791 excludes it precisely because the column read
provider timestamps and OKX sends whole seconds). That is a separate, small
change once both land, not part of either PR.

Checks

• go build ./... and go test ./cmd/script/ clean
• pnpm validate 235 specs valid
• pnpm exec tsc --noEmit clean
• bun test src/ 425 pass, 0 fail

🤖 Generated with Claude Code