How the rule was picked
Before deciding what counts as a trading market, the feeds were measured: every Chainlink round the eighteen wrote over thirty days, walked back round by round. The numbers on this page are that measurement, not an estimate, and they are what the window and the quorum were chosen from.
over 30 days, 18 feeds
every Saturday and Sunday, all 18 feeds
99.6% agreement with the trading week
What a month looks like
Rounds per UTC day. The gaps are not quiet markets, they are zeroes: not one feed wrote anything.
The feeds do not keep US session hours
This was the surprise. The feeds run around the clock on trading days, because Robinhood's stock tokens trade 24 hours a day, five days a week. The regular 13:30–20:00 session shows up only as a density peak, so a rule over feed activity can separate the trading week almost perfectly and the regular session not at all.
The trading week
Best rule scores 99.8% agreement over seven days and 99.6% over thirty, with a 60 second lag at the weekly open
The regular session
Best rule scores — and is already on before the bell on one day in five, so this contract does not claim it. Use a calendar for that
The weekly open, to the second
The first round of each trading week, over the five weeks measured. That is 20:00 New York time the evening before, when the 24-hour market opens.
| Week opens | First round, UTC | Feed that wrote it |
|---|
Every rule that was tried
Each pair of window and quorum was replayed minute by minute over thirty days and scored against the trading week. Accuracy alone hides the failure that matters for a gate, so the last two columns count how often a rule blinks shut in the middle of a working week.
| Rule | Agreement | False open | Lag at open | Overhang at close | Blinks shut | Minutes shut |
|---|
The deployed rule is the row in colour. A wider window rides through an overnight lull; a higher quorum refuses to call the market on when only the two busiest feeds are awake.
What would make it wrong
A half day was never seen
Thirty days contained no early close, so the behaviour at one is unverified. The rule would most likely hold the week open for the length of the window past the last tick, as it does at any close.
The sample of edges is small
Five weekly opens and four closes. The 60 second lag and the five minute median overhang rest on those, not on a year of them.
Two feeds do most of the writing
CRCL and MSTR alone account for about a third of all rounds. A calm week compresses every count, which is exactly why the quorum is five and not fifteen.
Clocks move
The weekly open is 20:00 New York time, so it sits at 00:00 UTC in summer and 01:00 UTC in winter. Nothing in the contract is tied to a UTC hour, which is the point of reading activity instead of a calendar.
A heartbeat is not a floor
Fifteen of the eighteen feeds went more than 48 hours without an update across a weekend, and SPY and QQQ have stalled for a full 24 hours mid-week. No single feed can carry the answer, which is why the rule counts them.