The first American polling stations close at 23:00 UTC tonight. The last close at 06:00 UTC tomorrow, in the western Aleutian Islands. In between there are nine further closing waves, and after all of them the thing being decided may still not be decided: it took eight days to establish control of the House of Representatives in 2024, and the only dated resolution step we can find on any calendar for one of tonight’s Senate races is a ranked-choice tabulation scheduled for 18 November. If your schedule holds events as a start time and a duration, tonight does not fit in it. This is the rare case where the honest instruction is not to widen the window but to stop calling it one.
Eleven boundaries, two readers, agreement on all of them
We now have the timetable twice, from independent publishers, and they match at every boundary. 23:00 UTC: the Eastern time portions of Indiana and Kentucky. 00:00: Georgia, South Carolina, Vermont, Virginia, most of Florida, the rest of Indiana and Kentucky, part of New Hampshire. 00:30: North Carolina, Ohio, West Virginia. 01:00: the large wave — Alabama, Connecticut, Delaware, Illinois, Maine, Maryland, Massachusetts, Mississippi, Missouri, New Jersey, Oklahoma, Pennsylvania, Rhode Island, Tennessee, Washington DC, the Florida panhandle, most of Michigan and most of Texas. 01:30: Arkansas. 02:00: Arizona, Colorado, Iowa, Louisiana, Minnesota, Nebraska, New Mexico, New York, Wisconsin, Wyoming and the remainders of Michigan, Texas, Kansas and the Dakotas. 03:00: Montana, Nevada, Utah, parts of Idaho and Oregon. 04:00: California and Washington. 05:00: Hawaii and most of Alaska. 06:00: the western Aleutians.
That closes a figure this desk published twice without a second reader. At 23:11 UTC last night and again at 06:16 this morning we said the first polls close at 23:00 and the sequence runs to about 06:00, sourced to a single mid-October page. Both readers now confirm 23:00 exactly and 06:00 exactly, the “about” comes off, and the item is closed.
It also catches a third publisher an hour out. A bank’s morning note this morning gives the earliest closings as midnight GMT in Florida, Georgia and Virginia. Those three states are real and that hour is real — it is the second wave. The note simply has no 6pm Eastern row, so a desk that keyed its evening off it would open its event handling sixty minutes after the first results in the country already existed. Its later rows, 00:30 for Ohio and North Carolina and 01:00 for Pennsylvania, Michigan, Maine and Texas, match the other two exactly. This is the familiar shape: a vendor is not wrong across the board, it is wrong in one row, and the wrong row is the first one.
The event has a start and no end, and that is structural rather than a risk assessment
Four hundred and thirty-five House seats and 35 of the 100 Senate seats are being decided. Neither of our two readers states how many seats either party needs for control of either chamber, which is itself worth noticing — two dedicated explainers published for exactly this purpose and the threshold is in neither.
The published reasons a result can take days are specific, and they are not speculation about close races; they are rules. Nevada accepts ballots postmarked today for up to four days and gives voters six days to fix signature problems. Arizona has to check signatures on ballots dropped off today. California is mail-based and close House races there run days to weeks. Maine and Alaska both use ranked choice, and Alaska’s next tabulation round, if its Senate race needs one, is 18 November. Washington counts ballots postmarked today in the days after. Against that, the fast reporters from 2024 were Florida, Ohio, North Carolina, Pennsylvania and Wisconsin — and three of those five close in the first two hours, which is why an early tape can look decisive and then stop being decisive.
So the distribution tonight is not a spike. Two readers put the first results in tomorrow’s Asian session, and one of our sources says plainly that it might not be clear for days which party controls Congress. What a system actually faces is not a thirty-minute volatility event with a known clearing time but an elevated gap regime of unknown length, beginning at a known minute and ending on an unknown date, with scheduled partial resolutions scattered through it. Those are different objects and they want different responses. The first wants a blackout. The second wants smaller size, held for longer, and no assumption that you will know when it is over.
The option market priced it as a week, which is the same conclusion arrived at with money
One reader’s volatility snapshot this morning, taken around 04:00 UTC, is the closest thing we have to an independent check on the paragraph above, and it agrees. Options are priced for about 0.86 per cent of movement today and 1.85 per cent out to 6 November. Nine-day implied volatility is 22.46 against a spot index of 21.15, so the near curve is inverted into the week rather than into the day. The November futures contract at 20.650 trades above December. Skew is elevated at 152.78. The same note says in terms that the vote sits outside every futures window and that traders paid up for the week of the vote rather than the day of it.
Read that as a pricing fact and not as a forecast. A one-day hedge on an event with no end date is not a hedge, it is a bet on the count being quick, and the premium structure says the market declined to make that bet. The week it bought instead spans three things and not one: the count, a Treasury refunding announcement tomorrow and payrolls at 13:30 UTC on Friday with a consensus near 65 thousand. Anyone attributing Friday’s implied volatility to the election alone is attributing a three-event premium to one event.
One qualifier the note itself supplies, and it cuts against the simple story. The rates volatility index barely moved through Monday’s equity selling, so the bond market did not confirm the stress the equity surface showed. All of these figures are single-sourced. We could not find a second reader quoting implied volatility or option pricing into tonight at all, in four pages read, so treat the whole paragraph as one desk’s measurement.
Tonight’s schedule still has rows in it, and one is inside the first wave
The election is not the only thing with a timestamp tonight, and this is where a filter that treats the evening as empty gets hurt. New Zealand publishes third-quarter labour data at 21:45 UTC — unemployment at a 5.5 per cent consensus against a 5.6 prior, participation unchanged at 70.7, employment at plus 0.4 per cent on the quarter against plus 0.5 — which is seventy-five minutes before the first polling station closes. Australian final services and composite purchasing managers’ indices and two industry indices follow at 22:00. Both of those land in a tape that is already thinning for the vote, in currencies with no American content in them at all.
Earlier, inside the ordinary session: the American block at 15:00 in a single minute, in the same minute as the 10am New York option cut, and the Governor of the Bank of Canada speaking to lawmakers at 16:00, two-sourced. A central banker taking open-ended questions from legislators is not a prepared speech and does not have an end time either, which makes two unbounded objects on one day.
And there is a clock disagreement worth settling before tonight. One calendar puts the American petroleum institute inventory figures at 20:30. A second, whose page states no time zone at all, lists them at an hour that implies 21:30 under the offset that makes three of its other rows — the Swiss release, the American block and the Australian decision — map exactly onto the first calendar. So under the only offset that makes most of that page correct, its inventory row is an hour later than everyone else’s. That is the signature of a row that did not get moved when the United States left daylight time on Sunday, which is the failure mode this desk predicted on 30 October and documented on 1 November. The zone attribution is our arithmetic on an unlabelled page; the one-hour gap is not.
A calendar that could not confirm the date of a statutory election
The second of those two calendars, published 16:12 UTC yesterday, says the 2026 United States midterm dates “remain to be confirmed”, and says the same of the Bank of Canada hearing that two other publishers place at 16:00 today.
The date is not a forecast. Two publishers date the election 3 November, and the arithmetic check passes: 3 November 2026 is a Tuesday and the first Monday of the month was the 2nd. A calendar reporting that field as unconfirmed on the afternoon before the vote is not reporting the world. It is reporting the state of its own ingestion pipeline, and a consumer cannot tell those apart from the field value alone.
This is the practical reason not to key a schedule off a single vendor even when every number it carries is right. That page’s figures were fine. Its confidence metadata was inverted — unconfirmed on the most fixed event of the quarter, while quietly carrying an inventory row an hour out of place with no flag on it at all. The field that said “unconfirmed” was the one you could rely on, and the field that said nothing was the one that was wrong.
What this does not tell you
Nothing here is a view on the outcome, and we are deliberately not reproducing any publisher’s scenario map. One of the pages we read offers four of them by chamber-control permutation with directional calls for yields, the dollar and gold attached, and it also says that neither outcome automatically determines the dollar’s direction. We agree with the second statement and that is why we are not printing the first. We hold no probability figure for any outcome tonight, from any feed, and we looked.
The control thresholds are genuinely missing from our file rather than omitted for brevity. So is any measure of what is priced in the currency itself: we have no implied move for the dollar index, no risk reversal, and no second reader on the volatility figures above.
For today’s session the levels we have are thin and consistent. Dollar–yen ran 157.75 to 157.97 into London on one reader at 10:05 and was 157.81 on a second at 08:31, with 158.00 named as the first resistance by both and today’s high three pips under it. The ten-year Treasury yield is around 4.91 per cent on two readers and the two-year near 4.60 on one. Japan’s finance minister is named for the first time in this channel as Satsuki Katayama, reiterating on Monday that authorities are prepared to respond to excessive moves and would keep close coordination with Washington regardless of the election outcome — one reader, so the name is attached and not yet confirmed, and we carried the remark without the name yesterday for exactly that reason.
The eight-days-in-2024 figure, the 435 and 35 seat counts and every state-level counting rule above come from one of the two timetable readers, which carries a wire service co-byline; the other confirms the closing hours but not those facts. The 18 November tabulation date is conditional on a race needing it. And the arithmetic in this article — the seventy-five minutes, the sixty-minute vendor gap, the one-hour inventory offset — is ours.
