FSI's Category IV hours don't divide by FSI's own week

Here's a thing you can do in about 40 seconds, with a calculator, on a government web page anybody can open.

FSI, the Foreign Service Institute, is the US State Department's language school. It publishes four language categories, each with a week count and a class-hour count. Divide the second by the first and you get the class hours in an FSI week.

Category I: 552 hours over 24 weeks, and 690 over 30. Both give 23.

Category II: 828 over 36 weeks. 23 again. Category III: 1,012 over 44. Still 23.

Category IV: 2,200 hours over 88 weeks. That's 25.

Three categories agree with each other and with the sentence State prints a few lines above them. The fourth disagrees with all of it. I have a guess about why, which I'll get to and then argue against. The disagreement itself needs no explanation to be checked.

How many class hours are in an FSI week?

The answer is 23, printed in the paragraph that introduces the whole table on the page State publishes now:

So 23 is the published rate. It's stated once, at the top, and every category underneath is supposed to be an application of it.

Run all four.

FSI class hours divided by FSI weeks, all four categories

0

Category I, both ends of the range

0

Categories II and III

0

Category IV, the super-hard tier

No rounding anywhere. Five divisions, five whole numbers, and one of them is the wrong whole number.

The whole-number precision is what makes this readable at all. If the hours were rough estimates you'd get 22.7 and 23.4 and a shrug. Every figure on the page is a clean multiple, so every figure was computed, and a mismatch is a bug instead of noise.

The fourth category divides by 25 on both pages

Now put the archived version next to it. Same four categories, same week counts, different hours.

Category I was 600 to 750. Category II approximately 900. Category III approximately 1,100. Category IV, 2,200.

Divide those too. 600÷24 is 25. 750÷30 is 25. 900÷36 is 25. 1,100÷44 is 25. And 2,200÷88 is 25.

The old page is internally consistent at 25 across all four categories. The new page is consistent at 23 across three of them, and keeps the old number in the fourth.

So the rewrite was one edit: a rate change from 25 class hours a week to 23, with the week counts left alone. It was applied to Categories I, II and III, and not to Category IV.

I want to be careful about what the divide-by-25 pattern does and doesn't prove. State never says "we changed the rate."

I'm inferring the mechanism from the fact that every old figure divides by 25 and every new figure except one divides by 23, while not a single week count moved. That's a strong inference and it isn't a quotation, so hold it as the first thing and not the second.

What isn't an inference is the mismatch itself. 88 × 23 is 2,024. The page says 2,200.

Those differ by 176 class hours, which is 8.0% of the figure the page prints, and 7.7 extra weeks of class at the rate the same page publishes.

Category IV, computed both ways

published 2,200 88 weeks × 25 hours. The figure as printed, on both the current and archived pages.

consistent 2,024 88 weeks × 23 hours. What the page's own stated week gives.

I can't tell which of the two numbers is wrong

Here's where I have to stop, because the interesting question has no answer I can source.

Maybe Category IV genuinely runs a longer class week. Super-hard languages might carry more contact hours by design, and 2,200 could be a real observed figure the other three categories never had.

In that case the omission is deliberate, the page is fine, and the only fault is that it never says so. Or it's an oversight, and Arabic should read 2,024.

One test I can run. If 2,200 is a deliberate exception, the exception predates the rate change, because 2,200 sits unchanged on both pages. On the archived page it was the ordinary figure, 88 × 25, exactly like the other three categories.

Nothing about it was special in 2017. It only became the outlier when the other three moved and it stayed put. That's the signature of a cell that got skipped rather than one that got defended.

The unchanged 2,200 is evidence short of proof. A rate that was always right for Category IV and merely coincided with the old universal 25 would look identical from here.

From outside, I can't tell an oversight from a deliberate exception. State publishes one weekly figure and applies it to three categories, and I have no document explaining the fourth. I asked nobody at State, so treat this as an arithmetic finding and not a reported one.

The size of the Category IV discrepancy is worth holding next to the ordinary revisions, because it's larger than any of them. Category II lost 72 hours in the rewrite, Category III lost 88, and Category I lost 48 at the short end and 60 at the long end.

Category IV lost nothing. The correction it didn't get would have been 176 hours, twice the biggest one anybody did apply.

What I'll commit to is narrower and survives either explanation. The published Category IV figure isn't consistent with the published Category IV week.

Maybe you're quoting 2,200 hours for Mandarin next to "23 hours a week in class." Then you're quoting two numbers that were computed on different assumptions, and you should say which one you trust.

There's a related reason not to lean hard on any of these week counts. The other federal language school answers the same question differently.

The Defense Language Institute puts its Category IV languages at 64 weeks, where FSI says 88. DLI runs Categories I and II together at 36 weeks. That happens to match FSI exactly on Category II and sit above FSI's own Category I range of 24 to 30 weeks.

So the two schools agree in one band and diverge by 24 weeks in another. Which is a separate post, and it's the reason "the government says X hours" is a weaker citation than it sounds.

The unrevised cell implies a 25/15 week for the hardest languages

The mismatch lands in the shape of the week as well as the size of the total.

FSI's week is 40 hours: 23 in class, 17 alone, so class is 57.5% of it in Categories I, II and III at every week count. Category IV comes out at 62.5%, because 2,200 of 3,520 is 25÷40. The unrevised figure quietly implies a 25/15 week for Arabic and Mandarin students, where the sentence at the top of the same page says 23 and 17.

Published class hours as a share of the 40-hour FSI week

57.5%

Categories I, II and III, each exactly 57.5%

62.5%

Category IV, at 62.5%

37.5%

Category IV self-study share, implied, where the stated week gives 42.5%

The implied 25/15 week is the part worth carrying off this page, because the 15 is what you'd copy without noticing. Somebody planning a run at Mandarin reads 2,200, divides by an hour a night, and gets 6 years, with the self-study never entering the calculation at all.

88 weeks at 40 hours is 3,520 either way. Take the stated 17 and 1,496 of those hours are time alone. Take the cell and it's 1,320.

Between the two readings sits 176 hours of somebody's evenings, and no page on state.gov tells you which.

If you're quoting FSI for a Category IV language, quote 2,024 as the published week times the published weeks, or 2,200 as the cell as printed, and say which.

If you're planning your own study, use the ratio instead of either total. It's the only figure on the page that survives the inconsistency. And 23 to 17 comes closer to four hours with a teacher for every three on your own.

23÷17 is 1.353. Four to three is 1.333. The 3:2 it usually gets remembered as is 1.5, and that one is off by a tenth.

Scale the 17/23 ratio to the week you have. Multiply your class time by 17/23.

One 60-minute lesson gives 60 × 17 ÷ 23, which is 44.3 minutes of solo work. Two lessons give 88.7. Call it 45 minutes and an hour and a half.

45 minutes is a disappointing number to arrive at, and it's the useful one. The ratio divides down honestly where the totals refuse to.

A plan built on 2,200 hours tells a part-time learner they need 6 years and says nothing about Tuesday. A plan built on 17/23 says 45 minutes on Tuesday.

I'll lay out why that second half decides whether the first sticks in a post to come.

The self-study half is where Verbamor sits. It turns the vocabulary from your own lessons into gapped-sentence cards with native audio on an FSRS schedule (an algorithm that predicts when you're about to forget each card).

$19.99 a month, iOS only, and beyond about 650 starter words it teaches you nothing new. Anki is free and runs the same algorithm family, so if you'll genuinely maintain your own deck, use that instead.

---

Sources