The Silver Bulletin Year in Review

My favorite posts of 2025, some notes on the business, and a thank you to our readers.

On a train near Bergen, Norway, headed to the fjords, in a more reflective mood than usual. Nate Silver.

Hello readers,

Thank you so much for your continued support of this newsletter. I wish you happy holidays and an auspicious New Year. We made the big time: a couple of Fridays ago, Silver Bulletin was even featured in a New York Times crossword clue!

Two quick announcements:

• It’s a good time to submit questions for SBSQ #28 in the comments to the previous edition. I’ve got some halfway-finished thoughts about tanking in the NBA, taxes in California and a recent media controversy or two that might make their way into the SBSQ depending on what questions people ask. But otherwise it’s open season.

• We’re hiring a part-time editor! The original deadline passed and we have some promising candidates. But since I haven’t had a chance to review the applications in detail yet, I’ll go ahead and extend the deadline through next week (Jan. 7). I will give you a bonus point or two¹ if you applied by the original deadline, however.

We have some big plans in store for 2026. Our generic ballot polling average will begin publishing soon, then our new NCAA basketball model, COOPER. Later in the year, of course, we’ll have the midterm forecast, and I’m really, really hoping to find time for some sort of World Cup model, too. (I’m honestly less certain about the timing of NBA stuff.)

But since this newsletter is headlined “year in review”, I’m going to take that mandate fairly literally and mostly look back on 2025, highlighting some of my favorite posts along with a couple that didn’t quite work. I’ll also give you some insight into our business metrics, as I always find it interesting when other writers do this.

Riding the election wave up and gently down

When you run a newsletter that’s (partly) about politics, it can be hard to disaggregate how your business is going from how your life is going from how it’s going in the broader world. Presidential election years are insanely stressful but also very good for business (and the more drama there is in politics, the better it is from that standpoint).

It can also make for an abrupt transition when the election is over, however. You can feel like you’re hurtling along an exponential curve as Election Day approaches, but unless there’s some sort of extended recount, interest drops off quite quickly.

I don’t have the old FiveThirtyEight traffic data saved anywhere. But based on Google Trends data, search interest in FiveThirtyEight declined by amounts ranging from 72 percent to 88 percent in the immediate post-election year versus the election year (e.g., in 2013 versus 2012).

Silver Bulletin makes for a slightly apples-to-oranges comparison with FiveThirtyEight; we’re a paid newsletter rather than a free website and we have a staff of three people rather than ~30 (and I’m personally writing about a broader range of topics than I did during the FiveThirtyEight days). Still, I was expecting a decline, especially given that last year had way overperformed my expectations. (This also requires a mental adjustment; there’s not the same endorphin rush of Number Go Up when you’re fighting a receding tide from the election year crest, which tends to be stronger at particular inflection points.)² But it was gentle by the standards I was used to at FiveThirtyEight and quite close to the expectations I had written down³ ahead of time:

• Overall (paid + free) subscriptions are up 12 percent year-over-year.

• Paid subscriptions are down 27 percent. Realized subscription revenues⁴ are down less than that, 17 percent, because our election-year numbers had a high ratio of short-term monthly subscribers, whereas this year, we’re more weighted toward annual subs (as most Substacks are).

• Perhaps most gratifyingly, people are reading. The median post this year⁵ received 283,000 pageviews, actually slightly higher than last year’s 272,000 — though last year had a wider range and a higher average because the election forecast got so many views. I can’t speak for every newsletter, but at least for us, Substack has been a better way to get in front of more people than my partnership with Disney/ABC/ESPN ever had been.⁶

About 80 percent of my time is devoted to this newsletter (and I work a lot).⁷ Still, I’m going to be honest that it can sometimes be hard to find a routine. There’s not much rest for the weary; the minute I publish something, I’m thinking about (and sometimes feeling guilty about) the next post.

I also travel a lot, and even in non-election years, many of our beats are cyclical. March and April were extremely busy — almost an election-year pace — between the NCAA tournament and a couple of stories that I thought really played to our strengths, Trump’s tariffs and his relationship with Elon Musk. October/November was another really busy stretch between the off-year elections, the shutdown, a massive gambling scandal and the launch of ELWAY.

Other periods were slower. One of the reasons we’re hiring an editor is to raise the floor for content volume during those times. Nevertheless, I’m very much a believer in maintaining a high bar for publication and avoiding cheap hits.

Another goal I have for next year is to increase the volume of posts that aren’t about politics or sports. These typically take longer to prepare, but they’re often quite popular when we do get around to publishing them. (Our post about how early to leave for the airport was a surprising breakout hit, for instance.) Since SBSQ’s publish toward the start of the month, perhaps I’ll just mentally designate a slot for an “off-topic” post to publish mid-month; I have a considerable backlog of AI-related ideas, for example.

I don’t think last year’s version of this post quite clicked, so let’s try something that feels a bit more organic. Instead of a laundry list of best and worst posts — we don’t really make that many capital-P Predictions anyway in non-election years anyway — I’m going to walk you through some of what I felt like were the highlights and lowlights chronologically.

January

I’m not sure I had much of a plan for the immediate post-election period. There were quite a few election postmortems, which I felt like I was entitled to publish, especially after the Biden fiasco. But in retrospect, I wish we’d ripped the band-aid off and transitioned into post-election mode more quickly. I got a little phobic at some point of publishing non-politics posts — the two-part series on a very interesting Baseball Hall of Fame class fell kind of flat, for instance.

Nevertheless, I felt like the trio of posts we published around Trump’s inauguration was some of our better content of the year, along with another post on AI:

• Why Biden failed (Jan. 19)

• Are we entering a Conservative Golden Age? (Jan. 22)

• 113 predictions for Trump’s second term (Jan. 26)

• It’s time to come to grips with AI (Jan. 27)

The Trump post is full of testable predictions, something I wish (other) Trump critics did more of — though many of the predictions have a duration of his full 4-year term so I’m not sure whether I’m going to attempt some sort of interim update. The basics look about right, including the predictable backlash to Trump and resurgence for Democrats. You can argue that Trump has been even more norm-violating than I was expecting, but the initial predictions didn’t present a particularly optimistic portrayal to begin with and I’d like to evaluate them more systematically.

You could also argue that the AI post reads as slightly too confident given that my AI timelines have slightly lengthened over the course of the year. (More about that soon, maybe.) But the point of the post was that even at the 10th percentile of reasonable expectations for AI, it’s going to be far more impactful on society than merely being the “chatbot” that it’s sometimes dismissed as.

February

Another month where the transition into post-election mode sometimes felt clunky. But we got back into the sports swing after a story — the most inexplicable trade in league history — fell into every NBA fan’s lap. (I literally wondered whether I was hallucinating when the Luka trade scrolled across the ESPN bottom line during a random poker tournament I was playing.) I actually don’t think my story holds up particularly well, however, given how well Luka has played this year, averaging almost 34 points per game. Instead, I’ll direct you to this as our February highlight:

• So, how did the polls do in 2024? It’s complicated. (Feb. 17).

I’m going to be honest: pollster ratings aren’t a big seller, but they’re the backbone of our election models and part of what I see as the public-service function of this newsletter.

March

Outside of our election models, the NCAA tournament forecast is our biggest subscription driver. Publishing the model in 2024 was something of a last-minute decision. But this year we were substantially more organized about it, also developing a series of ratings for both men’s and women’s teams. This was also a moment when I was “feeling the AGI” as large language models were helpful in various ways in helping to expedite the modeling process.

The tourney — and a crosstown apartment move⁸ — did somewhat crowd out other posts in March, but I thought this was strong:

• America probably can’t have abundance. But we deserve a better government. (Mar. 26).

This is by no means a negative review of Abundance, the very popular book by Derek Thompson and Ezra Klein, both of whom I like a lot personally. In fact, I recommend the book. But the post explores some of the tension between whether Abundance is policy analysis or more of a political project; I’d argue it’s really more of the latter.

April

We were writing a lot about the tariffs in April. Frankly, I think there were a few misses, like this post discounting the idea that the stock market could serve as a check on Trump. Instead, Trump sometimes did TACO out, and stocks had a great year. Nor did the tariffs, even after being scaled down, have as large an impact on objective economic data as I was expecting or as most experts were, though they’ve contributed to a persistently gloomy consumer mood.

Meanwhile, my Democratic primary draft with Galen Druke — we’ll be doing another of these soon —- makes for an interesting time capsule. Galen and I both wanted Alexandria Ocasio-Cortez with the #1 overall pick, which looks prescient after Zohran Mamdani’s rise to fame later in the year. On the other hand, I drafted Andrew Cuomo (LOL) in the 5th round. On the other other hand, I got a steal by picking Gavin Newsom in Round 3. I might not be a fan of his approach to politics, but it’s evidently popular with a certain class of Democrats, as Governor Good Hair has rocketed to the top of prediction markets since then.

But here’s what I thought were our best posts of the month:

• What is Elon’s endgame? (Apr. 2).

• Defending democracy is easier when you listen to voters (Apr. 22)

The latter of these posts — a nuanced critique of Democrats’ emphasis on the Kilmar Abrego Garcia deportation case, but really a broader take on how democracy requires paying attention to public opinion — was popular with subscribers but generated a very mixed reaction on the broader internet. Trump’s approval rating on immigration did decline in April, but it remains his least-bad issue.

May

Some good variety this month, but here were my two favorites:

• Why didn’t anyone predict the American pope? (May 9)

• ChatGPT is shockingly bad at poker (May 21)

Although yes, I consult for Polymarket and am generally a “fan” of prediction markets, I also think we ought to understand their limitations. Oftentimes, like in sorting through incomplete returns on election nights, they’re uncannily accurate. But in other cases like that of the case of the papal conclave, which elected the American Robert Prevost even though he had only been trading at around a 1 percent chance, they can reflect feedback loops around what was obviously not particularly well-informed conventional wisdom.

Meanwhile, my frustrations with ChatGPT’s comically bad poker play led me to discount the notion of an extremely fast AI takeoff. While one should be careful not to extrapolate too much from personal experience, and while poker is uniquely challenging for language models in various ways, the idea expressed in the post — that AI-based intelligence is likely to be spiky in the immediate term rather than broad or necessarily general — increasingly became the conventional wisdom over the course of the year.

June

I spent most of the month at the World Series of Poker. While the WSOP itself went pretty well, including a deep Main Event run, my other poker tournament ventures⁹ this year mostly didn’t.¹⁰ I love poker, but tournaments are time-consuming, and the opportunity cost becomes more tangible when you’re working for yourself instead of being on someone else’s clock.

Still, there’s basically never a time when I’m totally “logged out”. In fact, the WSOP can lead to some extremely long days because I’ll basically try to get half a day of real work in on top of the poker. So we managed to get a couple of good posts out — although the Zohran post required very nearly an all-nighter.

• Why young men don’t like Democrats (June 2)

• Zohran delivered the Democratic establishment the thrashing it deserved. (June 25)

• Did the Thunder get too good too fast? (June 28)

The last of these, my annual post on the newly-crowned NBA champion (see also the Nuggets in 2023 or the Celtics in 2024) has become the closest thing we have to a tradition for a newsletter that’s only a couple of years old.

July

Undoubtedly, the most attention-grabbing post of the month was “Is Epstein the New Russiagate?” I think it mostly holds up well. Even after Congress voted to compel the release of the Epstein files — though the White House has not exactly taken the directive in good faith — it’s hard to draw too much of a connection between Epstein and Trump’s declining popularity. I suspect it’s a factor at the margin, but not in proportion to the predictions you were seeing this summer in more partisan outlets that the walls were finally caving in on Trump. And it’s not as important a story to rank-and-file voters as it is to political junkies.

But I thought these were the two best posts of the month:

• NBA Future of the Franchise Rankings 2.0 (July 16; see also Part II here.)

• Why Colbert got canceled (July 21)

I’d done a first version of the franchise ratings back in February, but I thought it worked much better as a team exercise with Joseph George and Jeremias Engelmann. Plus, our optimism about teams like the Spurs and pessimism about the Clippers is looking smart. The Colbert take was fairly equivocal but I think reads as slightly too credulous to CBS given subsequent events. But I think it’s still a good story having been informed by my own experiences working for a giant corporate news conglomerate.

August

By August, I was hard at work on ELWAY/QBERT. Ironically, sometimes when you’re busy with a model, your other newsletters can be surprisingly good because they’re a way of procrastinating from model work:

• Trump’s jobs data denialism won’t fool anyone (Aug. 3)

• I’ve tracked my last 800 flights. Here’s when you really need to get to the airport (Aug. 7)

• Real talk on models, moderation, and the misuse of academic authority (Aug. 20)

• Democrats can win the redistricting war (Aug. 25)

One complaint I’ll sometimes see on Bluesky/Twitter — not so much from actual subscribers — is that I’m too focused on Democrats’ shortcomings and not enough on Trump. Part of that is because a lot of Silver Bulletin’s content is election-focused. So we’ll often have mean things to say about Democrats when they lose, which they did a lot of in 2024, and nice things when they win.

And part of it is that there is an extreme abundance of newsletters in the marketplace whose orientation is basically “HERE’S 3 BAD THINGS TRUMP DID TODAY”. I usually agree that these bad things are, in fact, bad, but I’m not sure I have anything differentiated to say about them.¹¹ However, I’m more than happy to write about these topics when they overlap with my areas of interest, such as on tariffs, or after Trump fired the BLS commissioner. Those other newsletters can also swing between extreme optimism and extreme pessimism about Democrats’ political prospects. I’m citing the redistricting post because it proved fairly prescient, with redistricting having been fought to a draw when many Democrats were predicting doom over the summer.

The post on models and moderation could have been more concise and used a better headline. But the post it was responding to, from the academics Adam Bonica and Jake Grumbach, was so full of BS from the first sentence onward that it literally made me depressed about the state of academic political science.¹²

September

In late August and early September, I took a trip to Norway and Finland. The fjords were spectacular, but the surprising highlight of the trip was Helsinki: great food, great design with an eclectic mix of architectural influences, and I even surprised myself by sort of liking the sauna. Perhaps that put me in a good mood because I thought September was a strong month for posting:

• The best quarterbacks of all-time (Sept. 2)

• What is Blueskyism? (Sept. 5)

• What is Schumer’s shutdown endgame? (Sept. 22)

• Is America ready for a gay president? (Sept. 27)

I’m not sure how I feel, on the whole, about our coverage of the shutdown —by the end, it felt like we were repeating ourselves a lot — but that initial Schumer post proved to be prescient. The post on Mayor Pete was surprisingly popular, actually our best converter to paid subscriptions in the second half of the year outside of NFL stuff.

October

In October, I was mostly busy with ELWAY, so let me cite my favorite posts of the year from Joseph and Eli, respectively, along with a post on the sports betting and poker indictments that I absolutely had to make time for.

• Can Wemby make the MVP leap? (Oct. 2)

• I loved my time in the UK. But it needs an AC intervention. (Oct. 19)

• The NBA gambling scandal, explained by an actual gambler. (Oct. 25)

As for ELWAY itself, it got off to a strong start prediction-wise. But the demise of the Chiefs and the Lions was rough on the system.¹³ ELWAY was onto a few smart things early, however, such as the Seahawks and Drake Maye, so we’ll see if it can redeem itself in the playoffs. I think the bigger issue here from a planning standpoint is that ELWAY and QBERT are really ambitious systems, with a lot of moving parts, whereas maybe it would have been better to take a more iterative approach that had launched in time for Week 1 instead of midway through the season.

November

I thought our election coverage was generally quite strong, mostly striking appropriately optimistic notes about Democrats’ prospects — take Eli’s prescient post about how easy it would be for Jay Jones to win in Virginia, for instance. But it doesn’t necessarily have that much evergreen value for re-reading now. The rest of November was a little slow as I was recovering from the ELWAY/election crunch, but I thought these two posts were a lot of fun:

• Bring the big leagues to Mexico City (Nov. 3)

• The NFL has entered the Scorigami Era (Nov. 14)

December

It’s too recent to have much perspective on these. So I’m just going to cite the one newsletter that received the most likes of anything we published this year:

• What is Heather Cox Richardsonism? (Dec. 16)

Yes, I make my fair share of criticisms of certain types of Democrats. But I think the HCR post (and the Blueskyism one) are the best versions of these and show that there is some deeper philosophical thinking behind all of this. But honestly, there’s probably going to be less of that next year.¹⁴ Turning the page on 2025 means that we’re also mostly turning the page on 2024 postmortems and can be more forward-looking as we head into the midterms. Thanks for reading, and I’ll see you all in 2026!

1. I mean that quite literally; I’m quite rigorous about our hiring and I evaluate applications based on a point system.

2. Non-renewals roll out of the system either one month or one year after the initial subscription. So anniversaries of popular posts often result in lower numbers than you started the day with.

3. I do literally have to make some guesses from time to time for tax planning or other purposes.

4. Realized, meaning money that actually ends up in our Stripe account, as opposed to Substack’s “annualized revenue” estimate, which can be considerably inflated because it assumes that monthly subscribers stay signed up for a full year when, in our experience, many of them don’t.

5. Totals for both years exclude posts that were never sent to the entire email list; typically, these are things like job listings or methodology pages.

6. For newsletters published this year, the 10th percentile number of pageviews was 236K and the 90th percentile was 368K. Last year those numbers were 84K and 450K respectively. The average post, counting model landing pages, received 376K views this year versus 445K last year. Note that these figures are based on the date when the post was published, not when the pageview occurred. Our “back catalog” sometimes does fairly well, so some of the pageviews attributed to 2024 actually came this year.

7. For the record, the proportion of my income I derive from the newsletter is also about 80 percent.

8. The East Side is the best side (of Manhattan).

9. Increasingly, my volume is concentrated in cash games, but what happens in cash games stays in cash games.

10. Granted, not a huge volume outside the summer WSOP. Mainly, I ended the year feeling frustrated because I went to the WSOP “Paradise” event in the Bahamas in December after qualifying for various incentives rather than making my usual trip to the World Poker Tour event at the Wynn. I found the whole thing a bit cynical, with the WSOP pulling out every trick to ensure it hit its $50 million guaranteed prize pool for its $26K buy-in “Super Main Event”, including extremely liberal re-entry rules that resulted in some players entering the event literally a dozen times. Outside of the Super Main, however, the schedule was sparse — one rumor was that this was deliberate to funnel more people into the Super Main. Also, I can find the mega-resorts in the Bahamas to feel confining, whereas in Vegas, if you don’t feel like playing or bust out of a tournament, there are great restaurants, pro sports, shows, you name it. Location aside, it wouldn’t be that hard for the WSOP to replicate some of the feeling of the summer WSOP in the Bahamas with a better schedule that featured more variety of events at more reasonable buy-in levels.

11. And posts along the lines of “this is bad, but not quite as bad as you’re saying” seem to satisfy nobody.

12. As a believer in Brandolini’s law — “the amount of energy needed to refute bullshit is an order of magnitude bigger than that needed to produce it” — I’m going to give myself some credit for the occasional very deep dive like this, because they are a lot of work.

13. Even though both teams have excuses in the form of bad injury luck.

14. Though by no means zero. This tweet is a good candidate turn into a future post, for instance.