The Next Heaven's Pardon Day Is December 16. The Rule Is One Line.

A Japanese temple's blog published a guide to Heaven's Pardon day (tenshabi) yesterday, October 26, 2026. It is a careful page. It lists six dates for this year and six for next, names the next one as Wednesday, December 16, and explains that the day is fixed by the stem and branch of the date, read against the season. Then it tells the reader that the dates cannot simply be calculated and should be checked against a settled calendar.

The first half of that is correct, and we checked every date on it. The second half is the part worth arguing with, because the rule is short enough to print in a sentence.

“

When Heaven's Pardon day falls is decided by the combination of the calendar's stems and branches (eto), and the qualifying stem-branch pair also changes with the season.

Which is to say: the date is decided by the calendar's combination of stem and branch, and the qualifying pair changes with the season. That is the rule, stated almost completely, and then stopped one step short of naming the four pairs.

The rule, in four pairs

Heaven's Pardon day is the day whose stem and branch match the pillar its season is assigned: Yang Earth Tiger in spring, Yang Wood Horse in summer, Yang Earth Monkey in autumn, Yang Wood Rat in winter. Season means the solar quarter opened by the Starts of Spring, Summer, Autumn and Winter, not the calendar month. That is the whole of it. One line, four fixed pairs, no free parameters and nothing to tune.

In Japan it sits in the lower band of calendar notes, and coverage of lucky days treats it as the strongest of the auspicious days. The same Heaven's Pardon appears among the auspicious spirits of the Chinese date-selection tradition. We are not going to claim an edition for it — these are transmitted tables, and we could not place the line ourselves — so everything below is a claim about dates, not about a text.

Run through this site's own two functions, the stem-branch counter and the solar-term solver that the almanac page already uses, the rule returns for 2026: March 5, May 4, May 20, July 19, October 1 and December 16. For 2027: February 28, April 29, May 15, July 14, September 26 and December 11. That is twelve for twelve against the two published lists we read, weekdays included. Two Heaven's Pardon days within two weeks in May 2026 looks like an error and is not: spring's Yang Earth Tiger falls on May 4, one day before the Start of Summer opens, and summer's Yang Wood Horse arrives on the 20th.

Two things printed beside the dates

The dates are right. Two statements that travel with them are not.

The first is the count. Both pages we read give Heaven's Pardon day as coming five or six times a year, and that is the figure repeated across this whole genre of coverage. Seven happens. By our computation 2019, 2020, 2024, 2031 and 2035 each hold seven, and 2029 and 2037 hold five. It is a small error with an awkward shape: a reader told six, who one day counts seven on a published calendar, has no way to work out which of the two sources is wrong.

The second is the framing, and it is the one that actually costs a reader something. These pages rate October 1, 2026, as the autumn's strongest day for opening a wallet or starting a plan — and the same pages print its place in the rokuyo, Japan's six-day cycle, as Butsumetsu, the worst of the six. December 16 is Shakko, the second worst. Nobody is misreporting: the rokuyo is (lunar month + lunar day) mod 6, and run against this site's own lunar code, October 1, 2026, was lunar 8/21 and December 16 is lunar 11/8, giving Butsumetsu and Shakko exactly as printed. The two systems simply disagree about these days. A date that is strongest in one count and worst in another is not strong on balance. It is named twice.

Where the rule is genuinely ambiguous, and it is not where anyone looked

If you want a real soft spot in Heaven's Pardon day, it is not the count. It is what happens when the qualifying day and the opening of its season land on the same date.

August 7, 2025, was a Yang Earth Monkey day. The Start of Autumn that year opened at 13:52 Beijing time on August 7, 2025. So the autumn pillar fell on the very day autumn opened — in the afternoon. Read by calendar day, that is a Heaven's Pardon day. Read by the moment, the morning of the 7th was still summer, and a Yang Earth Monkey day in summer is not a Heaven's Pardon day at all. Nothing we read says which reckoning it used.

That coincidence is rare and we can count it exactly. Across 2015 to 2040 the qualifying day lands on the opening day itself twice: November 7, 2022, a Yang Wood Rat day on which the Start of Winter opened at 18:46 Beijing, and August 7, 2025. Two more land on the day after an opening — February 5, 2020, and May 6, 2040 — which is close but not ambiguous, since by then the season has unambiguously begun. So there are exactly two dates in twenty-six years where two defensible readings of the same one-line rule can disagree, and neither is in the window anybody is currently writing about.

There is a second place you might expect trouble, and it turns out clean. The four Starts are moments, and a moment can fall on different calendar days in different time zones. The Start of Winter 2027 opens at 23:39 Beijing on November 7, which is 00:39 Tokyo on the 8th — different dates in the two places that most use this rule. Across 2015 to 2040 that happens to one of the four Starts six times. It never once moves a Heaven's Pardon day, because the qualifying stem-branch day is never near enough to a boundary for it to matter. That is a robustness result rather than a worry, and it is the sort of thing worth computing before repeating as one.

Our honest read

Check the date against a settled calendar. That is good advice and we would give it too. What is wrong is the reason attached to it. The page implies you should look the dates up because they cannot really be worked out. You should look them up because they can — because the rule is four stem-branch pairs and a solar-term table, which makes any published list falsifiable, and checking a list takes an afternoon rather than faith.

The difference matters to anybody selling a reading, this site included. A date you can only look up is a date you have to trust, and trust is exactly how a wrong list gets copied for a decade. A date you can derive is one where the second source either agrees or has some explaining to do, and the count above is what that looks like in practice: a figure repeated everywhere in this genre, wrong, and wrong in a way one script settles.

By the standard we apply before a layer goes on our own page — statable in one line, and a count that has to come out right or the arithmetic is broken — Heaven's Pardon day is among the most checkable things in the almanac. It is not one of the six portents we compute, and having now run it, we cannot claim that is because it failed the bar. Nobody had gotten around to it. That is a thinner reason than we would like, and the honest version of this piece has to include it.