Every filing held, inspected, and stamped before it's a fact.
Tokenised US equities trade around the clock, but the SEC filings that move them land almost entirely — ~97% of last year's 8-Ks from the ten underlyings this tracks — while the US market is closed. AFTERHOURS holds every one as bonded cargo — three independent parses, a per-field agreement count, a challenge window — and releases it on-chain instead of asserting it.
~97% is stated at that precision, not a false-precision decimal: 171 8-Ks in the trailing year, and an independent recount puts 166–168 of them inside closed-market hours depending on how DST and holidays are handled — no published minute-by-minute classifier exists yet for this figure.
No wallet · no signature · no gas — this is a read, not a transaction.
Every filing is cargo until it clears inspection
This is how the dock runs, every time, from the moment a filing appears on EDGAR.
RECEIVE
The dock watches SEC EDGAR for every 8-K and Form 4 filed against a tracked ticker, around the clock — including every hour US equity trading is closed.
INSPECT ×3
Three independent LLM parses read the same document, blind to each other. Each field — event type, effective date, declared amount, affected token — gets its own agreement count, 0 to 3.
STAMP
The result posts to EventStateRegistry bonded in USDT0, carrying the source document's SHA-256 hash and every field's agreement level. Nothing is asserted without its receipt.
RELEASE
A challenge window opens. Anyone can dispute a posted field by proving it against the source document and take the bond if they're right. Silence, not trust, is what clears it.
What stale pools actually cost, measured, not modeled
Two measurement studies, method fixed and pre-registered before either ran, reported as they came out.
4.6 min
median gap, filing acceptance → first on-chain trade
30% of filings (14/46) saw no trade at all inside the ±60 minute window.
$0.0614
median realised LP loss per trade, 60 minutes out
Net of the 25% protocol fee; the reported window is the trade that already happened, not a simulation.
Both studies ran against X Layer's live reference USDG pools. Full method and raw per-event data are in the repository's reaction-latency-study.md and markout-study.md — nothing here is rounded up from a smaller number.
Two tickers cleared for inspection, eight queued
The poller only watches what it's configured to watch. Here's the honest state of coverage today — a queued ticker returns "not yet tracked," never a silent empty result.