One Saturday, an artist market in a converted school building in South Philly was not on our list. The market was fine. It had a crowd, a folding-table full of handmade things and a loyal group of neighbors. Its calendar just lived inside an embedded widget on a page, which a person sees and a lot of other readers do not.
We ran TinyJawns, a family guide for Philadelphia, and we found ourselves missing things like this more often than we liked. That guide is where Tailend comes from. Every problem below is one we hit ourselves before we wrote a line of the API.
1. Where does local information actually live?
Not in one tidy place. It lives on flyers photographed and posted to a page. In calendar widgets that look great in a browser and say very little to anything else. In booking portals with their own layout. In social posts that are gone in two days.
An organizer is doing nothing wrong by publishing this way. It is how a busy person tells their neighbors what is happening. But every one of those shapes has to be read into something consistent before anyone can compare a class with a class or a Saturday with a Saturday.
2. What does it cost to know a page was read correctly?
This one surprised us. Reading a messy page is one problem. Knowing you read it correctly is a bigger one. A date that is off by a week looks just like a date that is right. A price that belongs to a different class looks just like the price you wanted.
You can put your most expensive effort into every page, every time, and get good answers at a price that does not scale to a whole city. Or you can be cheap and be wrong quietly. We wanted a third path: read each page into a consistent shape that can be checked, and admit when we cannot tell.
That last part matters most. "We are not sure" is a useful thing for a listing to say. A confident wrong date is not.
3. Why is "checked once" never enough?
Because the listing you read last month is not the event that happens on Saturday. Dates move. A session fills up. A farm posts a rain date. A class skips a holiday. A festival is called off because of weather and the announcement goes out on a story at seven in the morning.
The same event needs to be looked at again, and looked at harder as the day gets closer. The week-of version of an event is the only version anyone actually acts on.
4. What gets lost?
The long tail. When you see the same handful of famous places recommended everywhere, it is easy to assume they are the only things going on. They are just the easiest to read. The class in the church hall, the twice-a-month storytime and the market in the school building are what make a neighborhood feel like yours, and they are the first to fall out of view.
A guide can fix that for a single city by hand. It does not scale by hand, and it does not help the assistants and apps that families are turning to.
Why this matters
If you are a local business owner, your neighbors cannot find you if what you publish is unreadable, and you cannot see your own market if everyone else's is unreadable too. If you are building an agent for a family, your users will judge it by whether it knows the small, current, nearby things or only the famous ones.
Both of those come back to the same question: can a machine read the real world as it is right now, and say honestly how sure it is?
What Tailend is for
Tailend was designed for exactly these four problems:
- Messy sources become consistent events, classes, camps and places.
- Every listing says when it was last confirmed, and says so plainly when it could not be confirmed.
- Listings are looked at again as their dates approach, so the version you see is closer to the version that happens.
- Small, local, non-famous things are first-class, and the API says what it cannot see instead of implying the neighborhood is empty.
We start in Philadelphia, where we learned all of this, and we are working outward from there.