One of the stranger problems in making Yearbook is that something can be technically correct and still obviously be wrong.
That sounds contradictory.
It makes more sense when you meet Barbenheimer.
For a 2023 puzzle, Yearbook had a source event about Barbie and Oppenheimer opening in theaters on the same day.
Great Yearbook material.
It's recognizable. It's recent. It's weird. It was one of those cultural moments that will probably make people say “oh yeah” for a long time.
The problem was what happened next.
The automated system could look at the source event, follow the articles associated with it, and land on something like:
Oppenheimer.
Which is related.
It is factual.
It is also not the answer.
The answer is obviously Barbenheimer.
And that tiny distinction ended up being a good example of why I don't think you can completely automate editorial judgment.
The source event was fine
The underlying historical event wasn't the problem.
Barbie and Oppenheimer really did open on the same day in July 2023, and the simultaneous release turned into a whole internet phenomenon.
People made memes about seeing both movies.
There were jokes about the tonal whiplash of going from a bright pink comedy about Barbie to a three-hour Christopher Nolan movie about the atomic bomb.
People dressed up.
The term Barbenheimer became the shorthand for the whole thing.
So when that event appeared as a candidate for Yearbook, I liked it.
This is exactly the kind of thing I want in the game.
It isn't “the most historically consequential thing that occurred in 2023.”
That's not really the standard.
It is something people remember.
If you lived through that summer, there is a decent chance that seeing a clue about it brings back the moment immediately.
That matters more to me.
Then the computer had to decide what the event was “about”
This is where things get messy.
Yearbook doesn't just need an event description.
Eventually the card has to flip over and show an answer.
It also needs somewhere useful to send the player if they click Learn More.
So the system looks at the pages connected to the source event and tries to determine which one best represents what happened.
Most of the time, that's reasonably straightforward.
If the event is about a particular spacecraft landing somewhere, the specific spacecraft is probably a better answer than the general article about space probes.
If a particular person wins something, the person or event may be the obvious reveal.
But Barbenheimer is a different kind of thing.
The source material references two very famous movies. Those movies have substantial Wikipedia articles. They are obviously relevant to the event.
So it isn't ridiculous for an automated resolver to think one of those movies is the central subject.
That's the problem.
It's answering a slightly different question from the one a person would ask.
Technically related is not the same as what people remember
If I said:
Remember that thing in 2023 when Barbie and Oppenheimer came out together?
Nobody would answer:
Ah yes. Oppenheimer.
I mean, I guess they could.
But that's not the cultural memory.
The memory is that somehow these two completely different movies got combined into one thing.
Barbenheimer.
That's the fun answer.
It's the thing that makes the clue worth using in the first place.
If Yearbook reveals Oppenheimer, the facts haven't necessarily collapsed.
The clue could still be associated with a real movie released that day.
The link could take you to a perfectly legitimate Wikipedia article.
All the plumbing might work.
And yet the player would probably look at the answer and think:
That's not what that clue was about.
That reaction matters.
This is the kind of mistake that is hard to express as a rule
A lot of Yearbook problems can be turned into relatively clean checks.
Don't put the year in the clue.
Don't include the answer.
Don't introduce unsupported names or numbers.
Don't make playful jokes about serious events.
Make sure the source event actually occurred in the target year.
Those are things you can write rules around.
“Did this answer capture the cultural meaning of the event?” is a little harder.
You can create signals.
You can compare article titles.
You can rank how well a page aligns to an event description.
You can make the automated system better and better at choosing specific pages instead of broad ones.
But at some point you're asking a question about what people actually mean when they talk about something.
That is much fuzzier.
The Barbenheimer event wasn't memorable because Oppenheimer existed.
It was memorable because Oppenheimer and Barbie somehow became Barbenheimer.
A person reads the event and gets that immediately.
The system needed some help.
So I gave it an override
The solution was intentionally narrow.
I did not want to build a giant rule saying:
Whenever two movies are released together and the internet gives them a combined nickname, always use the nickname.
That would be a very specific piece of software for a problem that has happened exactly once so far.
Instead, Yearbook now supports an editor-approved correction for this particular locked event.
For that source event, the intended reveal is Barbenheimer.
The important part is that the override doesn't give the editor unlimited permission to replace the source with whatever sounds more fun.
It still has guardrails.
The event itself remains the same locked Wikimedia event. The corrected destination still has to be one of the Wikipedia pages actually linked from that source material. If the override tried to point somewhere completely unrelated, the system would reject it.
So the human correction can say:
No, this is what the event is actually about.
It cannot say:
While we're here, let's make up something better.
That distinction is important to me.
I don't want the override to become the system
There is a temptation when you're building something like this to keep adding exceptions until the entire thing becomes exceptions.
That is not the goal.
If the automated resolver repeatedly makes the same kind of mistake, I'd rather improve the resolver.
If Yearbook consistently chooses broad articles instead of event-specific ones, fix that behavior.
If it routinely interprets cultural phenomena too literally, look for a general way to improve that.
But sometimes there really is a weird edge case.
Barbenheimer is a weird edge case.
The system can handle the normal cases.
I can handle Barbenheimer.
That seems like a reasonable division of labor.
It also changed how I think about “correct”
Trivia games make factual correctness feel binary.
Something happened in 1994 or it didn't.
These two people were teammates or they weren't.
A movie opened on a date or it didn't.
And there absolutely are places where the answer needs to be that simple.
If I put an event in the wrong year, that's wrong.
No philosophical discussion required.
But there is another layer of correctness in something like Yearbook.
Did I choose the right thing to reveal?
Did the clue point toward the memory I intended?
Will the player understand why this answer belongs on the other side of the card?
Those questions are editorial rather than purely factual.
I care about them because the reveal is part of the payoff.
You spend a few guesses trying to work out what the clue is talking about.
Then the card flips.
There should ideally be a little:
Ohhh, right.
Not:
...I guess?
That's also why the Wikipedia link matters
One of my favorite parts of Yearbook happens after the game is over.
The cards flip and you can click through to read more about each subject.
I do this almost every day.
Sometimes I already knew the event and just want to revisit it.
Other times the clue has introduced me to something I'd never heard of.
That means the article link isn't just technical metadata.
It's part of the game.
If the clue is really about Barbenheimer and the player gets dumped into an article that makes the answer feel like it was just Oppenheimer, something has been lost even if the link technically contains relevant information.
The reveal and the link should help explain the clue you just played.
That's another reason I cared enough to fix what probably sounds like an absurdly small issue.
Which, to be clear, it is.
This is an edge case in a free daily trivia game.
Nobody's power grid was depending on the Barbenheimer answer being correct.
But if I see something and think, that's clearly not right, I'm going to want to fix it.
I have apparently chosen a relaxing hobby.
The nice part about finding problems like this
Every weird case makes the system a little better.
Sometimes the fix is a new general safeguard.
Sometimes it's a test.
Sometimes it's a change in how sources are ranked.
And occasionally it's a very specific note in the system that essentially says:
Yes, I know what the automatic process thinks.
Trust me on this one.
Barbenheimer got one of those.
The repository now has the correction explicitly encoded and tested so that the event resolves to the intended Barbenheimer reveal while remaining tied to the locked source event.
Which is quite a lot of infrastructure devoted to making sure a future trivia player sees the correct name for the summer when everybody watched a Barbie movie and an atomic-bomb movie back-to-back.
I stand by it.