The week
Seven days on one axis, and the list it was summed from.
19.35% sounding Β· 80.65% tacet Β· 32.5 of 168 hours. Each barβs height is the fraction of that hour the auction was actually running, so a bell hour draws shorter than a full one.
| day | date | open | close | what it was |
|---|---|---|---|---|
| Friday | 2026-08-28 | 09:30 | 16:00 | regular session |
| Saturday | 2026-08-29 | β | β | tacet β Saturday |
| Sunday | 2026-08-30 | β | β | tacet β Sunday |
| Monday | 2026-08-31 | 09:30 | 16:00 | regular session |
| Tuesday | 2026-09-01 | 09:30 | 16:00 | regular session |
| Wednesday | 2026-09-02 | 09:30 | 16:00 | regular session |
| Thursday | 2026-09-03 | 09:30 | 16:00 | regular session |
5
days with a session
2
days fully shut
0
of those were holidays
Where the calendar comes from
published exchange calendar, applied to the exchange's own wall clock, in America/New_York, via the exchangeβs own wall clock rather than a fixed offset β the session is 13:30β20:00 UTC in summer and 14:30β21:00 in winter, so an offset implementation is wrong for an hour a day, half the year.
The holidays and the two 13:00 half days are written out literally rather than derived. Good Friday needs a computus and the exchange has closed for funerals; a published list can be checked in a minute, a clever rule cannot be checked at all. It runs through 2027-12-31, past which this page says so instead of extrapolating β an invented holiday is invisible.
The figures above are the last seven days, measured, not a typical week quoted from memory. A plain week of five full sessions is 32.5 hours of 168 β 80.65% with no live market β and any window containing a holiday reads darker than that. What that does to the prices.





