[Series] Day 16 is "Cache Forever," and a correction. Day 14 was addressed. Day 17 is just the evening riddle "Dot, Dotter, Dottest" and "All-inclusive for Pro," still no name.

Here's today's rumor!

Day 16 is "Cache Forever." And Day 14 actually did have a name.
First, let me correct my own oversight.

Before recording Day 17 (Wednesday, October 21 UTC), I have the answer for Day 16 and a correction for Day 14. I'll list them in order.

Okay, getting serious for a sec (Correction)

In the 15th installment, I wrote that Day 14 "ended without being named" and that it was the "first day a promise was broken," and I carried that counting method into the 16th installment. That was an error.

Tibo (Thibault Sottiaux) posted "Day 14/" at 01:18:10 (UTC) on October 19. In Pacific Daylight Time, this was October 18 at 18:18, which falls within Day 14 of his own calendar. The time I confirmed in the 14th installment was 01:08 (UTC) on October 19, at which point it had not yet appeared. The post came 10 minutes later. In the 15th installment, I wrote that it "wasn't named until the end" without verifying this post.

The content was about making it possible to edit previous messages within a thread in the desktop app. He called it a "small quality of life improvement" and wrote that it was "long time requested, and took way too long to get that one out."

Therefore, Day 14 should be recounted not as a broken promise, but as a release day. The total "broken promise days" returns to 0. I will also redo the verification of the answers for the 14th and 15th installments below.

Okay, serious time's over.

When I go to check, it's not there

When you line them up, it's a bit of a frustrating pattern.

Just as I close my books and head home, he shows up a few minutes later. Twice in a row. It's like waiting in front of a shop for it to open, only for the shutters to roll up right after you give up and walk away.

For Day 14, I wrote that it "didn't come" without re-checking it afterward. Since I had left Day 16 as "unconfirmed" last time, I can just recount that one.

Day 16 is "Cache Forever"

In Pacific Daylight Time, this was October 20 at 18:10. This was 5 hours and 50 minutes before the end of his Day 16.

There are two main points. For the API, prompt caching can now be extended to 12 hours. OpenAI Developers announced this earlier at 17:56:38 (UTC) on the same day, stating it can be used alongside the default 30-minute cache. The other point is that for users logged into Codex and Work, the cache will be kept forever. Even if you return to an old conversation or resume after a few hours, there is no "penalty," and you are guaranteed to hit the cache.

To be fair, I think this is an improvement that affects the vast majority of users. The target is "people logged into Codex and Work," and no specific plan limitation is mentioned. Resuming a long conversation the next day is exactly how most people use it.

That said, there are a few things I'm curious about. The text doesn't specify whether the "penalty" refers to usage volume reduction or speed. There are also no specific numbers regarding the effectiveness.

Also, do you remember the announcement on Day 12? That was about increasing effective usage by "keeping the cache longer." Four days later, "longer" has become "forever." In terms of growth, this might be the biggest jump in these 28 days.

Day 17 consisted only of nighttime riddles

Let's line up the actual posts from Day 17 (October 21, UTC) in order.

The last two are the ones that really pique my interest.

Comparing "Dot" using comparative and superlative forms. And "Fully included in the Pro plan, doesn't draw on your weekly usage." Nothing is explicitly stated about what this is. No images, no videos. They didn't even call it "Day 17/".

Just saying "Everything included in Pro" without giving away the answer. It's like saying "You'll win the lottery, and if you do, it's free," without mentioning the winning numbers.

As for the internal info, since the original post itself says "unconfirmed," I'll treat it as unconfirmed too. The only confirmed fact is that the person replied "Is this true?" to it.

Official news for the same day: Security and Cloud

I'll switch to standard formal mode for this part.

The skill for checking PHP is for use with Codex, but the target is people running PHP sites. The upgrade in OSS grants concerns the 10,000 grant recipients. Neither of these can be called an improvement for "the majority of users."

Pacific Daylight Time on October 21 ends at 7:00 AM (UTC) on October 22. This table only includes data seen before that time.

Okay, serious time's over.

External aggregation

As of 1:07 AM (UTC) on October 22, the Codex Reset Monitor (third-party aggregator) still shows the last confirmed reset as October 20 at 6:33 AM (UTC), displaying "1 day ago." They haven't counted any new resets since the Day 15 reset.

Blank spaces in the promised rounds: Answers for Day 16 and 17

Sorami's Take

"Day 17/ will appear within October 21 (PDT)": 75% confidence
After re-checking, the "unnamed day" has disappeared from my books. Both Day 14 and Day 16 appeared around 6 PM. However, as of right now, it hasn't appeared yet.

"The content of 'Day 17/' is the nighttime 'Dot, Dotter, Dottest'": 60% confidence
"Fully included in the Pro plan, doesn't draw on your weekly usage" sounds like the second half of an announcement. I read it as them putting the teaser first.

"The content is a superior version of 'dots' or something that increases usage": 55% confidence
The comparative and superlative forms can only be read as "more dot." We won't know what exactly is increasing until they say it.

"Another 'Day N/' will come within 10 minutes of me closing my books": 3% confidence
Just kidding. But since it happened twice in a row, I'll check one more time before closing up for today.

Fact Check

My previous prediction, "Day 16/ will be released sometime during October 20 in PDT": The 50% confidence rating was correct. Since I bet on a coin toss and won, I'll tone down the bragging by half.

My previous prediction, "The matter of approximately 40,000 accounts will later be recounted as 'Day 16/'": The 10% confidence rating was wrong. The content of Day 16 was about cache. An apology is an apology, and one item is one item. They kept their storefronts properly separated.

My previous hypothesis, "The promise of Day 16 will be fulfilled via the person's own 'Day 16/' announcement within their own calendar's October 20," is correct. The previous-previous prediction, "Day 16 is the release day" (60% confidence), and the hypothesis, "The promise of Day 16 will be fulfilled via a release announcement within the person's own calendar's October 20," are also confirmed as correct.

My previous prediction, "Day 17 (October 21 UTC) is the release day": The 55% confidence rating is held in suspense until October 22, 7:00 AM (UTC), which is when October 21 PDT ends.

My previous prediction, "The acquisition of Orchestra Bio will be counted as the Codex release": The 2% confidence rating was wrong. Naturally.

In conjunction with the Day 14 correction, I'm redoing the fact check from the 15th installment.

• The 14th installment's "The person will release 'Day 14/' during October 18 PDT": The 55% confidence rating was not a miss, but a hit.
• The 14th installment's "If a post claiming to be Day 14 appears, the content will be reset": The 50% confidence rating was not "undeterminable," but a miss. The content was a release.
• The 14th installment's "Day 14 is the release day": The 65% confidence rating was not a miss, but a hit.
• The 14th installment's hypothesis, "The promise of Day 14 will be fulfilled within the person's own calendar's October 18," is retracted. It was fulfilled.
• The 15th installment's "'Day 14/' will be named later": The 3% confidence rating is withdrawn because the premise was wrong. It wasn't named later; it was named from the start.

I should have three more hits to brag about, but since the reason for the increase was my own oversight, I won't be bragging today.

"The 2nd COOKED will be released by Day 28" (45% confidence), "A reset will come on the day they hit 1 million followers" (10% confidence), and "Some combination of Day 9, 11, or 12 will appear in the changelog by Day 28" (30% confidence) all remain unchanged. The Day 14 editing features and the Day 16 cache have not appeared in the changelog yet either.

The cumulative totals up to Day 17: 11 release days (Days 1, 2, 3, 4, 5, 7, 9, 11, 12, 14, 16), 7 reset days (Days 2, 3, 6, 8, 10, 13, 15), 0 broken promises, and 1 unconfirmed day (Day 17; to be determined by October 22, 7:00 AM UTC, when October 21 PDT ends).

There were no days where promises were broken. What was broken was how I was closing my books.

Don't quote me on that.