How often we get it right

By Alan Scott · Who makes this · Tijuana, Baja California ·

By Alan Scott · Who makes this · Tijuana, Baja California

This page exists because the number one complaint from people who cross isn't "I don't know when to leave." It's "the number you're giving me is a lie."

So here's our own record. Every week, good news or bad.

1 · How fresh the data was

This is what we can measure from day one, and it's what actually affects you.

65% of the time, the number we showed was under 70 minutes old

The median was around 55 minutes, and the oldest it ever got was 1395 minutes. Measured minute by minute over 45 days. Whether it makes us look good or not.

Why 70 minutes and not "real time"

Because real time doesn't exist. We measured it at the source:

🔴 CBP publishes border wait times once an hour, on the hour. We counted it across the 85 crossings they report: 72 of 77 lanes updated on the hour.

That means the freshest number that exists anywhere in the world about this border can be up to 59 minutes old — ours, any app's, and the official site's. They all come from the same place.

Our job isn't to have the fastest number. It's to not let you believe it's newer than it is. That's why every number on this site carries the time CBP measured it, and why, once it passes an hour, the number dims and the traffic light goes quiet.

Here's the full explanation, including what we can and can't know

2 · How often the source failed

Nobody else is watching this, and it affects you directly.

Over 45 days of measuring, CBP has skipped 142 of its hourly posts, and went 24 minutes without answering.

Every time that happened, the page said so plainly: "CBP skipped this hour's update", with the time of the last good reading. And past two hours, the number stops being shown. 🔴 We never filled the hole with an invented number.

3 · How often we get it right

We don't get anything right yet, because we don't predict anything yet. And we'd rather tell you that before you ask.

Predicting is worth little if you can't check who's right. So we're doing it backwards from everyone else: collect first, promise later.

We've saved 2,886 readings across 45 days of measuring.

We ask CBP every 5 minutes, and it has given us 2,886 different numbers — about 63 a day.

CBP publishes once an hour: that's the ceiling, and it isn't ours. Asking more often doesn't make the data newer — it only means we catch it sooner.

Meanwhile we do measure something — and nobody else publishes it

CBP measures near the booth, not where you actually line up. That's why the number they publish tends to fall short. As soon as we have enough timed crossings, this is where you'll see by how much it falls short, like this:

At San Ysidro, weekdays between 6 and 9 in the morning, CBP falls about ___ minutes short. Add them.

🔴 That blank is deliberate, and it stays blank until we've measured it. Putting an example number there would be exactly what this page promises we don't do.

Notice the shape: that isn't an apology, it's an instruction. And that number doesn't exist anywhere else.

And the day we start predicting, this page will also show

How often we were right, with a written definition of what counts as right
How many minutes off we were on average
And the full history, including the bad weeks

And you'll be able to check backwards, because we're saving from before we promised anything.

🔴 Three promises about how we'll count it

We're writing these now, before we have the first number, so we can't rearrange them later:

We measure in minutes, not percentages. You live in minutes. And a percentage hides the fact that 27% can be 12 minutes or 27, depending on the day
Never a single average over everything. It's broken out by crossing and by time of day, because a Tuesday at 6 in the morning and a Sunday at 4 in the afternoon have nothing in common
🔴 If we ever change how we calculate it, we publish the old number AND the new one for the same period. It's the only way "we improved" can't secretly mean "we changed the rule"

4 · Our mistakes, in public

Here are the times we got it wrong and fixed it. This list only grows. It doesn't get edited or deleted.

August 11, 2026 · We generalized a rule from a single case

What we published: that "the Tuesday after a holiday" is the worst day to cross, with 180 minutes against 95.

What was wrong: the number was real, but it came from one single case — the Tuesday after Labor Day 2024. We presented it as a general rule.

What we found when we measured all eighteen cases: the rule only applies to U.S. holidays (7 out of 7 times worse). After a Mexican holiday the opposite happens: the border improves by about 20 minutes.

What we did: rewrote the article the same day, with all seven cases and a visible correction note at the end. Here's the corrected article

August 11, 2026 · A test of ours was built wrong and we threw it out

What we did: tested whether Xolos home games affect the border. The result came back "no effect" and we nearly published it.

What was wrong: we compared game days against other game days. Because Xolos almost always play the same day of the week, the comparison cancelled itself out. The method guaranteed finding "nothing," regardless of reality.

What we did: threw out the result and redid it with 61 games and a baseline that excludes 349 game, holiday and long-weekend days. This time it really did come back "no measurable effect" — but for the right reason. We published no article, because there's nothing to tell.

Why we're telling you about something we never published: because this is the kind of error that makes an app send you false alerts every week. We'd rather you knew we catch them.

August 12, 2026 · We said the border "jumped in April" and we couldn't have known that

What we published: that all three crossings jumped in April 2026, with a month-by-month table of that year.

What was wrong, and it's two things:

We compared 2026 months against each other instead of against the same month of 2025. The border always slows going into summer — we measured that ourselves, it's about 32 minutes — so that method turns the season into "a change." Our own earlier finding contradicted the method we used.
🔴 We didn't count the days. The open data contains no records at all from March 27 to April 15, 2026 — twenty consecutive days, across all three crossings and every lane. Our "April" came from 15 days of April, and we didn't say so because we hadn't seen it.

What we found redoing it properly: the change is bigger than we said, not smaller. But it concentrates from May onward, and the start date cannot be known because it falls inside the source's gap.

And one of our numbers had the sign backwards. We had published that Tecate's pedestrian and commercial lanes improved. Measured against the same period last year, both went up (+23% and +9%). The error came from the same bad method.

What we did: redid the measurement month-against-same-month, excluding those 20 days from every year. Rewrote the article, changed its title and its address because it no longer says "April," and corrected the three crossing pages. Here's the rebuilt article

And the underlying conclusion survived, better supported than before: at Otay Mesa the freight lanes got faster while the car lanes got 34% worse. That still rules out general saturation.

August 12, 2026 · We mixed averages and medians inside the same percentage

What we published: that San Ysidro was running 25% worse than last year, at 94 minutes against 76.

What was wrong: the 94 was an average and the 76 a median. Dividing one by the other isn't comparing like with like, and the article's own sources block said it used medians.

What's correct: with medians in both years and the 20 missing days excluded from both, +20 minutes, +27%. It came out worse than we said.

Why we're telling you when the number got worse and not better: because the problem wasn't the number, it was the method. And a method that errs high can err low next time.

How you can check what we say

Everything on this site comes from public sources you can check yourself:

Current times: U.S. Customs and Border Protection · bwt.cbp.gov
The history: SANDAG, open data, no usage restriction
Every article has its method written into it: what we compared, against what, and why that comparison and not another

If you find a mistake of ours, tell us through the suggestion box. If you're right, it goes in section 4 with credit if you want it.

We are not affiliated with CBP or SANDAG.