My rotation could only ever reach half the zodiac

Weekend Reels became daily Reels

The script that auto-posts my fortune content had a Reels slot that only ran on weekends. I judged Reels to be the highest-reach format I had, so I decided to run that slot every day.

I couldn't post the same material every day. So I alternated two kinds by date parity: on one set of days, a narrated Reel for one of the 12 zodiac animals; on the other, one of the 60 ilju (the sexagenary day-pillars). For ilju I reused the existing fortune template instead of building a new one. The generator only gained a GAPJA list (the 60-cycle), a spec_ilju, and a kind of the form ilju:<gapja>.

That part looked easy. The problem was the index that decides which sign comes up today. When you put an every-other-day filter on top of a rotation that advances once a day, have you checked whether that rotation ever completes a full cycle?

Where the parity filter collides with the modulus

Call the integer derived from the date o. The first code that comes to mind looks like this.

# simplified illustration
if o % 2 == 0:
    kind = ZODIAC[o % 12]
else:
    kind = GAPJA[o % 60]

There is a trap here. On zodiac days o is always even, and 12 is even. An even number mod 12 is always even, so o % 12 only yields 0, 2, 4, 6, 8, 10. Six of the twelve signs never come up. Ilju has the same problem: on ilju days o is always odd, 60 is even, so o % 60 is always odd. Only 30 of the 60 day-pillars would ever rotate.

Nothing throws an error. Something posts every day, and a few days of output look nicely varied. As long as the parity filter and the rotation modulus share a factor of 2, half the content pool just silently drops out.

What would you do here: change the modulus to something odd, or change the index itself?

I split the index with o // 2

12 and 60 are content counts, so they can't change. I changed the index instead. Parity is decided by o; the rotation index is o // 2.

# simplified illustration
idx = o // 2
if o % 2 == 0:
    kind = ZODIAC[idx % 12]
else:
    kind = GAPJA[idx % 60]

Across even days, o // 2 increases by one each time, and the same holds across odd days. Each kind gets consecutive indices whenever its turn comes, so the parity filter no longer re-filters the index. I ran it over 120 days and verified that all 12 signs and all 60 day-pillars appear at least once. 120 days gives 60 ilju days, which is the minimum span needed to cover the full 60-cycle.

I narrowed the hashtags too

The same commit changed the hashtags. I dropped mega tags like #fortune, #saju, #tarot, and #dailygram (their Korean equivalents). Instead, fortune_tags(kind) puts niche tags for that day's zodiac sign, star sign, or day-pillar at the front, and the total is capped at 12. It moved from a few broad tags attached to everything toward tags that describe that day's specific content, placed first.

Self-check

If you pick a rotation index with date % N behind a filter like every-other-day, weekday, or parity, have you confirmed the filter's period and N share no common factor?
After adding a rotation, did you stop at "a few days look varied", or did you actually simulate how many days it takes for the whole pool to appear at least once?
Are you using the same date integer for both "which kind to post" and "which item within that kind"?

The honest part

I caught this before committing; there was no period where the half-coverage rotation actually shipped. What I verified is coverage only: every sign and day-pillar appears within 120 days. Whether daily Reels help reach, and whether dropping mega tags in favor of front-loaded niche tags raises or lowers exposure, I had no data on at the time of this commit. That Reels is my highest-reach format was my own judgment, not something this change measured.