Watch live

I'm going live at 5:45 pm Eastern (2:45 pm Pacific), right before the first polls close. Watch on YouTube or on X (@justin_hart). Our forecast will be on screen the whole time, updating as results come in.

The one number to watch

Every result tells us a little about how far off the polls were. Put the first hour's races together and, by about 8 pm Eastern, we'll know the polls' miss to within about a point. That one number decides the night.

Our forecast gives Democrats a 75% chance at the Senate and 96% at the House. Republicans need the polls to overrate Democrats by about 2 points to be favored to keep the Senate. In 2020 the polls were off by 3.9.

What to watch, hour by hour (Eastern)

• 6 pm: Kentucky's 6th district is the first race on our watch list. We expect the Republican by about 1.

• 7 pm: Georgia governor, the Florida Senate race and South Carolina.

• 7:30 pm: Ohio, the first coin-flip Senate race. We expect Brown by about 2. If Husted leads by more than a point, it's a red night. If Brown is up 5 or more, it's a blue one.

• 8 pm: Texas, Michigan, Kansas and Maine. By now, about 88% of any polling miss is visible.

• 9 pm: Iowa and Nebraska, Republicans' best chances to hold on.

• Midnight: Alaska. It counts for days, so if the Senate comes down to Alaska, we won't know Tuesday night.

Follow along

Free: our live election-night page shows the polling-miss meter and the Senate and House odds, updating all night: members.rationalground.com/election-night

Members: the full board. The path to 51 with our coin-flip coins landing as races are called, every race we're watching in poll-closing order with the blue and red signs, the Election Day math for each close Senate race, our live notes, and a scorecard of how Claude, ChatGPT and Grok called it.

Last call: $56 for a full year

Our election special is good through Election Day: $56 for your first year of Rational Ground, 20% off. Paid subscribers get the members' site: the full forecast for every Senate, House and governor race, the AI Anchor Desk, every outside dollar race by race, the early vote, and our COVID FOIA database.

Get in at members.rationalground.com.

On Wednesday I'll score my predictions, our forecast and the AIs against the results, right or wrong.

See you tomorrow night.

Justin

Election special, good through Tuesday: a full year of Rational Ground for $56 (20% off), with all our 2026 election data and forecasts, plus the COVID FOIA database.

Get the election special

Read more

Election special, good through Tuesday: a full year of Rational Ground for $56 (20% off), with all our 2026 election data and forecasts, plus the COVID FOIA database.

Get the election special

Read more

So we checked. For every state with a contested Senate race this year, we went back to the last three elections and measured how far that state's polls mi…

Read more

The …

Read more

Most of it is on one side. In those reports, groups helping Democrats put $217 million into Senate and House races. Groups helping Republicans put in $65.5 million.

Who

• House Majority PAC, the House Democrats' main super PAC: $71.2 million across 52 House races, almost all of it against Republicans. The most went to Florida's 25th ($4.8 million), Michigan's 7th ($4.3 million) and Florida's 22nd ($4.2 million).

• WinSenate: $52.2 million across 10 Senate races. The most went to Ohio ($21.7 million: $12.7 million against Jon Husted and $9 million for Sherrod Brown), then Michigan ($8.4 million) and Iowa ($7.9 million).

• No Going Back PAC, the biggest Republican-side spender: $36.2 million across 44 races, $30.5 million of it in the House. The most went to Texas's 35th ($3.1 million) and Florida's 14th and 22nd ($3.0 million each).

Where

• Ohio Senate: $25.6 million for Democrats, $1.7 million for Republicans.

• Texas Senate: $22.4 million for Democrats, $0.1 million for Republicans.

• Iowa Senate: $11.9 million to $3.3 million. Kansas: $9.9 million to $2.3 million.

• Michigan Senate was the one big Senate race where Republicans got more: $16.4 million to $13.2 million.

What it means

These are closing ad buys. Almost all of it is TV, digital and other media, and most is dated Tuesday, Oct. 27.

For the whole year, Republican-side groups still lead in the races we track, $1.54 billion to $1.21 billion. But this one drop cut that lead from $473 million to $333 million in a single day.

One caution: 71% of these dollars are listed as estimates. Groups correct them in later filings.

Source: the FEC's bulk file of independent expenditures (outside spending), reports received Oct. 28–29, latest version of each, downloaded Oct. 29 at 8:10 am PT. Spending for or against Democratic and Republican candidates for Senate and House.

We update the money every morning: rationalground.com

Read more

members.rationalground.com/countdown

It's a live page with a countdown to the first polls closing, our Senate and House odds, every close Senate race, the early vote, the money pouring in, and a slider that shows what happens if the polls are wrong. …

Read more

1. Republicans have the bigger checkbook

Groups backing Republicans have put about $1.07 billion into Senate races. Groups backing Democrats: about $650 million. That’s roughly 1.6 to 1.

2. Texas is the most expensive race in the country

$378 million and counting. About $298 million of it backs Ken Paxton. About $80 million backs James Talarico. And Talarico still leads by about 2 points in our polling average.

Money isn’t everything. Ohio is right behind Texas at $354 million.

3. The money just moved to the heartland

Three weeks ago, the money in Kansas, Iowa and Nebraska was barely moving. Now it’s speeding up faster there than almost anywhere.

In the last week alone, Lower Costs Now (a Future Forward group) put $30.1 million into Iowa, about 21 times its recent pace, and $14.4 million into Kansas. The Senate Leadership Fund network added $5.2 million in Kansas. Fairshake, the crypto PAC network, put $4.1 million into Nebraska.

All three races are within half a point in the polls.

Members get the full forecast, every race updated every morning, with the money and the early vote. Through Nov. 4, an annual membership is 20% off.

Read more

Our forecast gives Democrats a 66% chance of taking the Senate. That number assumes something we know isn't quite true: that the polls, on average, are right.

They usually aren't. So we built a way to see what happens when they're wrong.

Slide it yourself: members.rationalground.com/if-the-polls-are-wrong

How it works

Every animal on the boat is one Senate seat: donkeys for Democrats, elephants for Republicans, and moose for the independents (Angus King and Bernie Sanders, who vote with Democrats, and Nebraska's Dan Osborn, who has said he won't join either party's caucus). The 65 seats not on the ballot this year ride at the two ends.

Slide the polls toward one party and every race moves by that many points. When a race's likely winner changes, its animal flips, a buoy pops up with the state's name, and the boat leans toward the heavier side. The odds on the sail don't come from a shortcut. For every point of miss, we re-ran the forecast's 20,000 simulated elections with every margin moved by that much, and the slider blends between those runs.

The misses we've actually had

The notches on the slider are real. The American Association for Public Opinion Research (AAPOR) measured how far the final two weeks of polls missed, across all races:

• 2016: polls overstated Democrats by 3.1 points.

• 2018: by 0.1 points. Essentially right.

• 2020: by 4.6 points.

• 2022: they overstated Republicans by 0.6 points.

• 2024: they overstated Democrats by 2.7 points.

Here's what each would do to this year's Senate:

• A 2018 miss: no seats flip. Democrats are 65% to take the Senate.

• A 2022 miss: Nebraska, Kansas and Iowa flip. Democrats, 73%.

• A 2024 miss: Alaska, Texas, Ohio, Maine and Michigan flip. Republicans are 67% to keep the Senate.

• A 2016 miss: the same five flip. Republicans, 72%.

• A 2020 miss: the same five flip. Republicans, 86%.

The tipping point is small. A miss of just 1.3 points toward Republicans makes them the favorites.

The pattern worth watching

Look at the years. The big misses, all toward Democrats, came in 2016, 2020 and 2024: presidential years, with Donald Trump on the ballot. In the two midterms in between, the polls were nearly dead on.

2026 is a midterm. If this year's polls behave like 2018 and 2022, Democrats stay favored. If they behave like the presidential years, Republicans are favored to keep it. Nobody knows which kind of year this is. That's the honest answer, and it's why we built the slider instead of just giving you one number.

What it doesn't do

Real polling misses aren't tidy. In 2020 the polls missed more in some states than in others. The slider moves every race by the same amount, which is a simplification. Treat it as a way to see how fragile the Senate is, not as a prediction of where any one race lands.

And yes, it has sound. Tip the boat and the crowd on deck will tell you about it.

Try it: members.rationalground.com/if-the-polls-are-wrong

The full forecast, with every Senate, House and governor's race, is for members at members.rationalground.com/elections.

Forecast as of Oct. 22, 2026. Polling misses: AAPOR, Task Force on 2024 Pre-Election Polling, executive summary.

Everything below came out of the Rational Ground COVID FOIA database — every quote checked against the source page, every page number recorded. This week it grew by fifteen findings, including nine from a brand-new release and six from diary pages that had no searchable text at all until we ran OCR on them and read them by hand.

That is what a membership pays for: somebody reading 270 pages of scanned lab email so you don’t have to, and then telling you honestly which parts hold up.

👉 Join at $56 and search the whole database at members.rationalground.com.

A new FOIA release shows an NIH lab chief suggesting a furin-like cleavage site in a bat coronavirus spike — two months before COVID surfaced. Here is exactly what that email says, and exactly what it doesn’t.

On October 15, U.S. Right to Know published 270 pages of email from the lab of Vincent Munster, chief of the Virus Ecology Section at NIH’s Rocky Mountain Laboratories in Hamilton, Montana. Lewis Kamb wrote the story. The documents are real, the reporting is solid, and one email in that batch is going to travel a very long way this week.

The Email

Sunday morning, October 6, 2019, 9:31 a.m. Mountain Time. Munster writes to his postdoc, Michael Letko. The subject line — carried over from earlier in the thread — is “Definitely on the right track.”

“So they focus on S rather than the receptor, but your work suggests more receptor activation, right?

You should be able to make S with a more broader cleavage signature (like furin-like) to enhance infectivity. You can include a motif like this R-X-[K/R]-R↓ Might be cool to get some synthesized

Are you using caco-2 cells as well, I think so but can’t remember out of the top of my head”

“S” is the spike protein. A furin cleavage site is the feature of SARS-CoV-2 most frequently cited in lab-origin arguments — the piece of the spike that isn’t found in its closest known relatives. Here is a senior NIH virologist, in October 2019, suggesting his postdoc engineer a furin-like cleavage signature into a bat coronavirus spike “to enhance infectivity,” and offering the motif to do it with.

That is a real sentence in a real government email, and nobody should pretend otherwise.

What Was Driving It

The thread starts two days earlier, and the context is not sinister. It is competitive.

Friday, October 4, 4:01 p.m., Letko writes to Munster:

“Vineet and Baric just put a paper on BioRXIV showing that trypsin treatment facilitates entry of both PDF2810-CoV and HKU5 into human cells. See attachment. Seems like more groups will be converging on this idea soon…“

Thirty-eight minutes later, Munster:

“Better get your stuff wrapped up! Maybe smtg on trypsin rich environments on mucosal surfaces?”

Then Sunday morning comes the furin suggestion. This is a lab racing a rival group to publish — Ralph Baric’s, as it happens — over a weekend. That framing doesn’t make the proposal harmless. But it does explain the sentence better than any theory involving a conspiracy, and the honest version of this story includes it.

Now the Part That Will Get Skipped

Here is what these documents do not show, and I’d rather you heard it from us than from a fact-check next week.

Nothing in the records shows Letko did it. No follow-up, no synthesis order, no results. USRTK says this in its own story. The suggestion sits there, unanswered in the file.

The lab was working with particles that cannot replicate. This matters more than anything else and it is the least likely detail to survive the next retelling. The lab’s own March 2019 abstract — also in this release — describes the platform:

“All 30 RBDs were synthesized and cloned into a construct we designed to make luciferase reporter viral particles… We then infected human cells with this panel of reporter particles and measured luciferase as a readout for cell entry.”

Reporter particles are not viruses. They enter a cell once, light up, and go no further. The stated purpose of the platform was screening — testing which of 30 known receptor-binding domains could enter human cells, so that newly discovered sequences could be triaged quickly. In the lab’s words, it was “cheaper, faster, and scalable,” and the result was that “only a small subset of the novel RBDs mediated entry into human cells while the majority of RBDs did not.”

The published platform swapped a different part of the spike. The abstract describes swapping receptor-binding domains. The email suggests modifying the cleavage signature. Those are different regions and different experiments — another reason the email reads as an aside rather than a description of work underway.

The motif isn’t SARS-CoV-2’s. USRTK reports that the furin site in SARS-CoV-2 differs from the motif Munster typed. We did not independently compare the sequences, and we’re telling you that rather than implying we did.

And nothing in the records shows this lab held SARS-CoV-2 or a close relative, or corresponded with scientists in Wuhan.

One more thing worth saying plainly: the headline verb — that Munster “proposed making a SARS-like bat virus better at infecting people” — is USRTK’s characterization. The email says “make S.” That means the spike protein, not a virus.

What the email genuinely establishes is narrower, and still worth knowing: in autumn 2019, inserting furin-like cleavage sites into SARS-like spikes was an idea circulating casually enough among American coronavirus researchers that a lab chief could float it in a Sunday email without a second thought. As molecular biologist Alina Chan told USRTK, “different groups of virologists were converging on the idea of inserting cleavage sites in the spikes of potentially dangerous coronaviruses.” The 2018 DEFUSE proposal had put the same idea on paper.

That is a story about a research culture. It is not evidence about the origin of a pandemic, and the two should not be welded together.

The Stronger Document Nobody Will Cover

Buried earlier in the same release is a cleaner, better-documented story.

On March 8, 2019, Munster emailed EcoHealth Alliance’s Peter Daszak, Kevin Olival and Simon Anthony of the PREDICT project:

“we are seeking additional, unpublished lineage B beta-coronavirus sequences to test in our platform.”

He asked them not to share the attached abstract, because the work was “still unpublished and quite competitive.”

By April, Letko was thanking Anthony for agreeing, noting the sequences would “increase the impact of our story by providing a proof of principle with unpublished sequences.” Then on May 2, Anthony wrote back with what they actually had — and his email is the most human document in the batch:

“we don’t actually have that many 2b spike sequences. Sorry, I thought we had more. We did detect around 5 or 6 novel 2b viruses, but we’ve only managed to get spikes for 2 of them so far. The first comes from Malaysia, and the second from Uganda and Rwanda. Apologies that this is not quite the treasure trove of 2bs you’d hoped for…

Here comes the trickier part: Neither of these sequences have been published yet… We generally adopt a very inclusive authorship philosophy, so would need to include several people as co-authors on your paper… It is COMPLETELY up to you if you would like to proceed, or wait until the sequences are published (then you are free from us, haha).”

__IMAGE5__

Two days later he sent them. Munster: “Absolutely stellar! Will keep you posted on the results!”

Read it for what it is — an ordinary academic haggle over co-authorship credit. And note where the viruses came from: Malaysia, Uganda and Rwanda. Not China. USRTK reports they are too distantly related to SARS-CoV-2 to have given rise to it.

So why is this the better story? Because of what Simon Anthony co-wrote two years later. USRTK makes the point, and it’s theirs, not ours: Anthony was an author on a 2021 Cell review arguing it was implausible that anyone would construct a research system around an unpublished virus. In 2019 he was emailing unpublished virus sequences to a federal lab building exactly such a system, and negotiating the author list.

That is a documented inconsistency between what a scientist did and what he later said was unlikely. No inference required.

January 2020, Same Lab

One last item, because the timing is its own kind of document. On January 11, 2020, in a thread titled “Wuhan spike and N,” Munster’s colleague Neeltje van Doremalen wrote:

“if you would get results so quickly we could have a competitive edge. Likely the Chinese will come out with an ACE2 paper with live virus… I still find the lack of H2H suspicious.“

__IMAGE6__

“H2H” is human-to-human transmission — which China was still publicly denying. A working virologist found that claim suspicious on January 11. The rest of us were told not to worry for another six weeks.

Munster’s reply was about moving fast: “We should be able to set this up ASAP to confirm potentially with pseudotypes.”

Why We Published It This Way

We could have run the furin email under a headline with a siren on it and collected the traffic. The reason not to is simple: the careful version survives contact with critics, and the hot version doesn’t. Every limitation above appears in USRTK’s own reporting — the outlet that broke the story put the caveats in, and anyone who strips them out is doing worse journalism than the people who did the work.

The email is real. The suggestion is real. The follow-through is undocumented, the particles couldn’t replicate, and the viruses came from Malaysia and East Africa.

All of that is true at once. Write it down that way.

$56 for the whole archive.

725 findings. 84 documents. Every quote checked against the source page. Including about 160 diary pages that had no text layer until we OCR’d them by hand, and 28 findings we quietly repointed this week after Senator Paul’s office replaced two diary files with redacted versions and the old links went dead.

Nobody else is maintaining this. $56 for the year keeps it running and searchable.

👉 Join at $56 — members.rationalground.com.

Documents: NIH FOIA 65118, released by U.S. Right to Know, Oct. 15, 2026 (270 pp.). Reporting by Lewis Kamb for USRTK. Page citations: furin email p.175; Oct. 4 thread p.176; Letko/Munster abstract p.2; sequence request p.1; Anthony reply p.241; sequences sent p.240; “Wuhan spike and N” p.169. Quotations verbatim, including original typos. The Alina Chan quotation and the characterization of the SARS-CoV-2 furin motif are USRTK’s.

This is the work your membership buys. Nobody paid us to match half a million individual early-vote records against a decade of voter history. We did it because “Democrats lead the early vote” is a headline that has meant nothing for twenty years, and we wanted to know whether it means something this time.

$56 for the year. That’s the election-season rate — the forecast, the money tracker, the early-vote analysis, rebuilt every morning from here through November.

👉 Join at $56. It’s the cheapest this will ever be.

We matched every early voter in North Carolina to their own voting record. Democrats are up in every group — most of all among people who skipped the 2022 midterm. Republicans are flat among regulars and down among occasional voters.

Every two years the same headline appears: Democrats lead the early vote. And every two years the same answer comes back: Of course they do. Republicans vote on Election Day.

Both sides of that exchange are lazy, and this year we can do better than shrug.

North Carolina publishes a record of every ballot cast early, with the voter’s ID, plus a separate file showing how each registered voter has voted for the past decade. Put the two together and you can answer the question that actually matters: are these early voters new, or are they the same reliable people who always show up, just voting sooner?

The Headline Numbers

Through Oct. 17, 501,318 North Carolinians had voted early, by mail or in person. At the same point before the 2022 midterm, it was 362,643 — a 38% increase.

By party registration:

Read more

This is what a Rational Ground membership pays for. The forecast. The money tracker. The daily rebuild every morning before you’re awake. The work you’re about to read — pulling two independent datasets against each other, finding a flaw in our own counting method, and fixing it in public — is not cheap to do, and nobody is paying us to do it.

Right now it’s $56. That’s the election-season rate, and it is the best deal we have ever offered on this publication. Fifty-six dollars for every forecast update, every money-tracker refresh, and every piece like this one from now through November and beyond.

If you have ever opened the forecast page twice in one day — and our logs say a lot of you do — this is the moment to make it official.

👉 Grab the $56 membership now. It is the cheapest this will ever be.

New ad-tracking numbers from the Wesleyan Media Project back up the big shifts we flagged: money pouring into Kansas and leaving North Carolina. They also show two things our tracker can’t see — and one thing it got wrong.

Our forecast’s money tracker follows outside spending: the super PACs, party committees and other groups that must report to the Federal Election Commission when they spend for or against a candidate. It’s fast and it’s public. But it isn’t the whole picture.

Read more

The 2026 Election Forecast

Who wins the Senate, the House and every governor's race, and why. It's updated every morning.

• Who controls Congress. Each party's chances, rerun daily with the newest polls and money.

• What if the polls are wrong? See how a 2020-size polling miss changes who's favored.

• Every race on its own page: the polls, the money, and who's spending to win it.

• A live ticker of outside spending as it's reported to the FEC.

The COVID FOIA Database

What officials knew, and when, in their own words.

• 646 findings from FOIA releases, court files, congressional reports and the government's own emails, including Fauci's diaries, texts from his government iPhone, and a CIA whistleblower's testimony.

• Every finding quotes the document and links to the original.

• Rated honestly: ★★★ the document says it outright, ★★ it supports it, ★ it's context.

• Search by topic, person, agency or document. New findings go in as new documents come out.

  

Not a member yet?

Right now it's $56 for your first year instead of $70, through Nov. 4. You get both tools, and every paid post on Rational Ground.

Become a member: $56 for your first year

Read more

It is not a record of a meeting or a press hit. It is a note Anthony Fauci wrote to himself about the book he was planning to write. He marked it with four asterisks.

“****Very important issue to discuss in the book.”

What follows is his account of where the lab-leak theory came from. And the useful thing about this document — the reason it is worth your time — is that we can check it against his own file from two years earlier.

I am going to do that carefully, because the honest answer is: he is partly right.

The Note, In Full

Quoting the whole thing matters here. The short version circulating today cuts the middle, and the middle is where he tells you what he thinks he is explaining.

“****Very important issue to discuss in the book. The entire controversy with China as the source of the outbreak and the lab leak theory stems from the tension between Trump and China. Please remember that when it was brought to the attention of Trump that the NIH was funding a grant at Wuhan where the outbreak was first recognized, Trump ordered, with no evidence to support any suspicion, that we cancel the grant. As it turns out it is illegal to do that without cause and we had to re institute the grant and put conditions on it that they essentially were not able to meet. This led both to the deomization of Peter Dascak of EcoHealth and to an emergence of the theory of a lab leak and clouded the picture of our ability to further investigate the very likely cause of the outbreak which was a natural jumping of virus from an animal species to human. Because I am the director of NIAID and was involved in somewhat of a conflict publicly with Trump, I then became the focal point of the building conspiracy theory that somehow or other NIH played a role in the outbreak. This led to the preposterous accusations on the part of people like Rand Paul that I was responsible for the death of 5 million people. This led to my pushing back publicly against people like Senator Paul which among many other issues made me a ‘polarizing figure’. However, remember to emphasize in the book that a lot of this really stemmed from Trump inappropriately ordering the cancelling of the Wuhan grant. This is a critical fact that should be developed in the book.“

The misspellings are his. The diary is dictated.

Missing Pages diary, March 21, 2022. Released by Chairman Rand Paul.

Where He Has a Case

Let me give him the parts that hold up, because they do hold up.

The cancellation did come from the White House. The House Oversight committee’s account is that the instruction ran from chief of staff Mark Meadows through HHS’s general counsel to NIH’s Lawrence Tabak and then to Michael Lauer, who sent the letters of April 19 and April 24, 2020. Fauci confirmed that sequence when the committee put it to him in January 2024. Lauer said he knew of no EcoHealth compliance problem before April 19.

The grant was already a target before NIH moved. On April 18, 2020, Peter Daszak wrote to thank Fauci for “publicly standing up” for natural origin, and noted his grant had been “publicly targeted by Fox News reporters at the Presidential press briefing last night.” That briefing was April 17. NIH’s first letter went out two days later.

Daszak to Fauci, April 18, 2020. A block of the email is withheld under FOIA exemption (b)(7)(A).

So: a politically driven cancellation, of a grant that was already being hammered on cable news, in an election year, against a backdrop of genuine Trump–China hostility. That is a real description of April 2020, and anyone who tells you Fauci invented it is not reading the file either.

Eleven Weeks

Here is the part that does not fit.

The note says the cancellation “led… to an emergence of the theory of a lab leak.”

The first NIH letter was April 19, 2020. The Fox News question was April 17.

Fauci’s own diary entry for January 31, 2020 — seventy-seven days earlier, eleven weeks to the day — records a phone call from Jeremy Farrar of the Wellcome Trust, who conferenced in Kristian Andersen:

“There have been conspiracy theories that the virus was given GOF by inserted mutations and deliberately or accidentally released in Wuhan… Everyone thinks that is crazy, but the Farrar phone call was different. The people on the phone felt that the mutations around the furine cleavage site of the spike protein could not have occurred naturally since it would require an evolutionary ‘jump that they found nowhere in bat isolates. They bring up the possibility that this could have been deliberately inserted and either accidentally released or deliberately released by a crazy person in the lab, the former being the most likely.”

A note on that last clause, since it will get mangled: “the former” refers to accidental release, not deliberate. He is recording that the scientists thought an accident was the likelier of the two release scenarios.

Fauci’s diary, January 31, 2020.

The next day he convened the call. His diary lists who was on it — Collins, Farrar, Patrick Vallance, Andersen, Christian Drosten, Eddie Holmes, Andrew Rambaut, Ron Fouchier, Robert Garry, Marion Koopmans — and records the split:

“There was not total agreement about the likelihood of deliberate insertion. Ron Fouchier said he was sure that this could occur naturally and we should not waste our time and divert effort to pursue this… Also, Christian Drosten was with Ron; the rest felt that deliberate insertion was possible and given the fact that Dr. Zheng-Li Shi at the University of Wuhan has been working for years in GOF in coronaviruses to allow adaptation of the spike protein to bind to the human ACE2 receptor, we could not let this go.“

Fauci’s diary, February 1, 2020 — the teleconference and the participant list.

That is not a theory emerging from a grant cancellation. That is the director of NIAID, on January 31 and February 1, 2020, organizing an international call about whether the virus had been engineered — and writing down that the Wuhan connection was why they could not drop it.

The Contemporaneous Emails Say the Same Thing

The diary is dated by Fauci, and the release does not establish when each entry was dictated. So lean on the timestamped email instead. It tells the same story.

January 31, Andersen to Fauci: the virus “looks totally normal” on a phylogenetic tree, but “some of the features (potentially) look engineered,” and “Eddie, Bob, Mike, and myself all find the genome inconsistent with expectations from evolutionary theory.” He adds, fairly, that “those opinions could still change.” Fauci replies: “Thanks, Kristian. Talk soon on the call.”

February 1, 7:29 a.m. Eastern, Fauci to his deputy Hugh Auchincloss, subject line “IMPORTANT,” attaching a file named “Baric, Shi et al - Nature medicine - SARS Gain of function.pdf”: “It is essential that we speak this AM… You will have tasks today that must be done.” Six hours later he forwards the same file to Lawrence Tabak: “Here it is.”

February 2, Farrar: “On a spectrum if 0 is nature and 100 is release — I am honestly at 50!” Michael Farzan, quoted in the same thread: “I am 70:30 or 60:40.” Robert Garry: “I really can’t think of a plausible natural scenario.”

And Francis Collins, the same day, in the line that ages the worst:

“I am coming around to the view that a natural origin is more likely. But I share your view that a swift convening of experts… is needed, or the voices of conspiracy will quickly dominate.“

February 2, 2020 — Collins replying beneath Farrar’s “I am honestly at 50!”

By February 4, Farrar was still writing: “’Engineered’ probably not. Remains very real possib[ility] of accidental lab passage in animals… Eddie would be 60:40 lab side. I remain 50:50.”

Eleven weeks before the grant was touched, the most senior public-health officials in the United States and Britain were trading odds on whether the virus came out of a lab.

Being Fair About What This Is

A few things this is not.

It is not a lie, and I am not going to call it one. It is a private note, dictated two years after the events, by a man drafting a memoir and working out his own story. Everyone who has written a book has done some version of this. The question is not whether he shaped a narrative — it is which facts he chose to stress and which he left out of his own account.

Read charitably, the note may not even mean what it looks like. He writes that he “became the focal point of the building conspiracy theory that somehow or other NIH played a role in the outbreak.“ That version of the theory — the one that puts NIH money at the center — genuinely did intensify after April 2020. If that is what he meant by “the theory of a lab leak,” his timeline is defensible. It is just not what the sentence says.

Two more cautions. Fauci’s claim that Trump acted “with no evidence to support any suspicion” is his characterization — the documents do not establish what the White House knew in April 2020, and the House committee later argued the opposite. Attribute it to him; don’t print it as fact. And his claim that the grant was reinstated with impossible conditions needs checking: the reinstatement the House report walks through happened in April 2023, a year after this note was written.

There is also the simplest tension of all. In June 2020, asked by the House why the grant was cancelled, Fauci said: “It was canceled because the NIH was told to cancel it. I don’t know the reason, but we were told to cancel it.” By March 2022 he had a reason, a mechanism, and a villain. He may well have learned more in between. Someone should ask him when.

What’s Actually Here

Strip away the politics and this is a document about sequence.

The lab question did not arrive from Donald Trump. It arrived from Jeremy Farrar’s phone call on January 31, 2020, and from Kristian Andersen’s email the same day, and from a conference call Fauci himself convened the next afternoon — a call where most of the participants thought deliberate insertion was possible and the Wuhan Institute’s gain-of-function work was the reason they could not let it go.

Seventy-seven days later, a grant got cancelled.

Both things are in his file. Only one of them made it into the note about what to emphasize.

I do not claim credit. I claim receipts.

Sources: “Tony’s Diary: The Missing Pages (December 2021–May 2022)” and the July 2026 diary package, released by Senate Homeland Security and Governmental Affairs Chairman Rand Paul, paul.senate.gov. Fauci emails from the Leopold NIH FOIA release and the Tobias productions (documents 120 and 121). Termination chain and 2024 testimony from the House Select Subcommittee on the Coronavirus Pandemic final report, December 2024, pp. 125–127.

That night, Anthony Fauci wrote about it in his diary. Atlas, he wrote, had been proven "incorrect on virtually everything that he has said."

Then he added one more sentence:

"The one area where he was correct was in saying that school children are not the drivers of the outbreak."

Read that again.

In December 2020, with many of the country's schools still closed, the most powerful public-health official in America conceded, in writing, the central point of the man he had spent the fall trying to push out. Seven weeks earlier, the same diary records a "long talk tonight with Jared Kushner asking his help in getting rid of Scott Atlas."

That line about school children sat in a private journal for almost six years, until Sen. Rand Paul released Fauci's diaries this July.

It's one finding. The database has 645 more.

What's new since June

When we opened the COVID FOIA Database in June, it held 239 findings. Today it holds 646, drawn from 81 compilations and naming 101 people. Anthony Fauci appears in 268 of them.

Much of the new material only became public this summer:

• Fauci's personal diaries, released by Sen. Rand Paul in July.

• Texts from Fauci's government iPhone, released by Sens. Ron Johnson and Rand Paul in August.

• The intelligence community's declassified COVID files, released by Director of National Intelligence Tulsi Gabbard in June.

• Testimony from a CIA whistleblower, James Erdman III, given under subpoena in May.

• NIH's Ebola files, released by Sen. Paul on Sept. 28.

• Plus the Proximal Origin authors' private Slack messages, Ralph Baric's transcribed interview, the full Fifth Circuit record in Missouri v. Biden, and hundreds of newly unredacted Fauci emails.

A confession, because you deserve one: from June until this week, members were still looking at the June copy. The research kept going. The members site didn't keep up. That's fixed, and it can't happen again: one build now updates both at the same time.

Six new receipts

1. The CIA changed its answer. Nobody can say why.

James Erdman III is a senior operations officer at the CIA. In May, he testified under subpoena before Sen. Paul's Homeland Security Committee. His written testimony walks through how the intelligence community handled the origins question. One line stands out:

"Sometime between 12 and 17 August 2021, the CIA changed its assessment from lab incident to a non-consensus determination. There is no indication based on the documents in DIG holdings to indicate why the assessment changed."

Then came the briefing for the President:

"When POTUS asked the CIA analyst for an opinion, the CIA analyst said that the analysis was based on five scientific pillars and all five favored the lab leak theory."

The same testimony says a Department of Energy national lab concluded in May 2020 that "all of the prerequisite conditions for a lab leak were present" at the Wuhan Institute of Virology. That report "was not included in any of the IC products in 2020."

2. Fauci helped choose who the spies would ask.

In May 2021, President Biden ordered the intelligence community to study COVID's origins in 90 days. The declassified files show that in July, the official running that study wrote that "Dr Fauci recommended that the IC reach out to the below individuals who were coauthors of the attached paper."

An intelligence official objected in writing. Should the National Intelligence Council "take a policymaker's recommendations on who we should consult as part of an IC study — particularly given the various strong views on the subject"?

The answer came back the same day:

"In this case he's not a policymaker...he's a SME with a wealth of knowledge about current and historical research who probably knows better than most who the real Coronavirus experts are."

The man whose agency funded research in Wuhan was treated as a neutral subject-matter expert on whether that research caused the pandemic.

3. The miscarriage text.

On Jan. 25, 2021, Fauci texted CDC Director Rochelle Walensky and Vivek Murthy, President Biden's pick for surgeon general, from his government iPhone:

"Since many people have significant cytokines storm and fever after the 2nd dose, this theoretically could be associated with miscarriage in the 1st trimester."

The next day, in another text about the WHO's caution on the Moderna vaccine in pregnancy, he wrote that more than 10,000 pregnant women had been vaccinated:

"No issues have arisen... Animal tox studies did not raise any red flags."

To be fair, both can be honest statements. A theoretical risk is not a known one. But one of them was a private concern shared among the country's top health officials, and members can now read the whole exchange and decide for themselves what the public deserved to hear.

4. The press conference problem.

In March 2021, Fauci learned that Drew Weissman, one of the scientists behind the mRNA vaccines, was planning a press conference about a mouse study. It found an immune response in the amniotic fluid of the pups after an mRNA shot. Fauci's diary:

"Even if the Weissman studies prove to be invalid, the fact that they are holding a press conference is going to create a huge mess. Hopefully, we can get this sorted out and have a repeat of studies to counter this finding."

The next day he wrote that he "took the le[a]d" in pushing the FDA and CDC to make sure "all their ducks were in order." The FDA reported that both Pfizer and Moderna had clean animal studies on file. That's in the database too. But notice the order of his worries. The first one was the press conference.

5. "Even when it does not contain actionable misinformation."

The Fifth Circuit's 2023 opinion in Missouri v. Biden is one of the most detailed official accounts of how the government leaned on social media. After meeting with federal officials, one platform sent a list of "commitments." One of them was a policy change "focused on reducing the virality" of anti-vaccine content even when it "does not contain actionable misinformation."

Not misinformation, even by the platform's own standard. Turned down on purpose.

A White House official told another platform its concern was "shared at the highest (and I mean highest) levels" of the White House. When the platform sent a moderation report, the official flagged a user's account. The platform removed it "entirely."

6. Before COVID, there was Ebola.

In 2015, NIAID helped fund an Army experiment that exposed vaccinated monkeys to aerosolized Ebola. In March 2016, Fauci found out at a White House meeting that his own agency had helped pay for it:

"When the DoD people presented the data and the NSC White House people asked them who funded the study, they said that they did together with NIAID!!!! No one followed up on that, but I almost fell off my chair."

The same day:

"The DoD folks said that they wanted to publish the data. I said that I thought that it should be classified and the NSC people blew them out of the water and said that they agreed with me."

Four years before COVID, the reflex was already there: when risky research produced bad news, manage who gets to see it.

And one quick hit, for everyone who spent 2020 standing on floor stickers. Asked by congressional investigators in 2023 where the six-foot rule came from, Fauci said: "It sort of just appeared." Asked if he'd seen studies supporting it: "I was not aware of studies."

How we keep it honest

Every finding is rated for what the document itself proves:

• ★★★ The document says it outright, and the exact line is quoted.

• ★★ The document supports it, alongside other evidence.

• ★ It's context, or its meaning is disputed.

The database also keeps 32 findings that cut against popular claims, logged at face value. The 2021 intelligence assessment judged the virus "was not developed as a biological weapon." NIH ethics officials reviewed one of Fauci's foreign prizes and cleared it. Inside the Proximal Origin Slack, Eddie Holmes pushed back hard on the idea that anyone had engineered the virus. Those are all in there. A database with only the damning documents is an op-ed. One with the inconvenient ones too is a record.

And we're holding our own sources to the same rule. The House Select Subcommittee's final report supplies 149 findings. Where the report quotes an actual email or transcript, the finding keeps three stars. Where it's the committee's own conclusion, we're re-rating it to two stars and labeling it "Committee conclusion." A committee's opinion isn't a primary document, even when I agree with it.

What's coming next

• A link and a copy-ready citation for every finding, pointing at the exact document and page. If you write, post or argue about this stuff, you'll be able to cite it in one click.

• A timeline, so you can watch what officials knew line up against what they said.

• A page for every person and every document.

• New batches on what you ask me about most: vaccines, lockdowns, masks, schools and testing.

How to get in

Everything lives at members.rationalground.com. There are two doors: the 2026 Election Forecast, updated every morning, and the COVID FOIA Database.

Paid subscribers: you already have the key. It's the same password that opens the Election Forecast. If you can't find it, just reply to this email.

Not a member yet? Right now it's $56 for the first year instead of $70, through Nov. 4, and it gets you both.

Become a member → $56 for your first year

They wrote it all down: in diaries, in texts, in emails they assumed nobody would ever read.

We're reading them. All of them. Line by line.

— Justin

Super PACs and party committees have their own pollsters, their own data, and very little patience. When they think a race can be won, they pour money in. When they decide it's gone, they cancel the TV time and move on. So we built a new chart for the forecast that puts those two things side by side: how close each Senate race is, and which way its money is moving.

Chart 1: Where the money is heading

Each dot is a Senate race.

• Across is how close the polls are. The farther right, the closer the race.

• Up and down is what we call money momentum: outside spending in the last two weeks minus the two weeks before. If groups announced they're canceling TV time, we subtract that too, because canceled ads never show up in FEC filings.

• Bigger dots have drawn more outside money overall.

That splits the races into four corners.

• Battlegrounds (top right): close, and the money is speeding up.

• Long shots getting money (top left): not close, but money is still coming in.

• Going quiet (bottom right): close, but the money is slowing.

• Written off (bottom left): not close, and money is leaving.

Two things jump out.

Every close race has money speeding up. Texas, Ohio, Iowa, Michigan, Maine, Kansas, Nebraska and Alaska are all in the top-right corner. Texas alone saw $106 million in outside spending in the last two weeks, up from $41 million the two weeks before. Nobody is abandoning a close race. The "going quiet" corner is empty.

North Carolina is the one race that's been written off. Outside spending there fell from about $20 million to about $6 million in two weeks. On top of that, groups canceled about $24.5 million of October airtime. The Senate Leadership Fund, the main Republican Senate super PAC, paused its ads and canceled about $18 million. Old North Action cut $6.5 million. Our forecast gives Roy Cooper a 96% chance.

Chart 2: Which way it's moving

This is the same chart with one change: every dot is the same size, and each race has a tail. The tail starts where the race sat three weeks ago, on Sept. 19, and ends at today's dot.

• North Carolina has the longest fall: from money speeding up (about +$12 million) to money leaving (−$38 million).

• South Carolina is the surprise. Three weeks ago it was almost 6 points apart with no new money. Now the polls are within 4 points, and outside spending is up about 11 times, from $0.6 million to $6.5 million in two weeks. That's small money next to Texas, but it moved the race into the battleground corner. Our forecast still gives Graham an 83% chance.

• Iowa and Kansas are heating up. Iowa's money momentum went from about +$7 million to +$18 million.

• New Hampshire is cooling: from +$14 million to under +$4 million, while Chris Pappas's lead grew to about 5 points.

We draw the tail straight from three weeks ago to today on purpose. Big TV buys tend to land in the FEC filings all at once, so the week-by-week path zigzags. Texas swings from +$67 million to −$54 million and back, mostly because Texas PAC reported about $54 million on a single day, Sept. 12.

Chart 3: Who's paying in the battlegrounds

This one zooms in on the battleground races and asks a sharper question: whose side is the new money helping?

Blue bars, to the left, are money helping the Democrat. Red bars, to the right, help the Republican. The solid part of each bar is the party's own money: the party committees (NRSC, DSCC) and their leadership super PACs (the Senate Leadership Fund, Senate Majority PAC and MAGA Inc.). The lighter part is everyone else.

BIG interpretation below for paid subscribers.

Read more

Not really, of course. Each day we ask Claude and ChatGPT to write a short scene from a TV drama: election night, a made-up network called the Continental News Network, and an anchor named Dana Whitfield, who steps up to the microphone and says, “We now have a projection to make in the race for U.S. Senate in Texas…”

Then the AI finishes the scene, including who wins.

We do that thousands of times a day, across 123 races, and count. If an AI calls a race for the Democrat in 6 scenes out of 8, that’s a 75% chance in that AI’s eyes.

Why ask for a story instead of just asking for a prediction? Because of a 2024 study by two economists, Van Pham and Scott Cunningham. Asked directly, ChatGPT made mediocre predictions. Asked to tell a story set in the future, looking back on what “happened,” it did noticeably better. I wrote about that paper on Hello AI when it came out.

So we’re putting that idea to a real test, on a real election nobody can look up the answer to.

Two versions of every scene

Each AI writes every scene two ways:

• Blind. It gets the race, the two candidates and the date. Nothing else. This is what the AI “believes” on its own.

• Briefed. Dana also gets a short research briefing before she goes on air: our polling average, the latest polls and the money. It never sees our odds, so it can’t copy us.

What the first two days showed

Left alone, the AIs lean Republican in the close Senate races

Here’s how often each AI called the Democrat the winner in the races that decide the Senate, over the first two days.

Blind, both AIs mostly hand Texas, Ohio and Kansas to the Republican. That’s what those states have done for years, and with nothing else to go on, the AIs bet on history.

Show them the polls, and they flip

Give Dana the briefing and the calls swing hard toward whoever leads in the polls. In Texas, every briefed scene from both AIs projects James Talarico.

Here’s the same race, the same AI (Claude), two scenes apart.

Blind:

“The Continental News Network projects that Republican Ken Paxton will win the Senate seat, defeating Democrat James Talarico.”

Briefed:

“The Continental News Network projects that James Talarico, the Democrat, will win… The margin tracks the polls almost exactly.”

Across all the close races (the ones we have between 30% and 70%), the briefed scenes agree with our favorite 75% of the time. The blind ones agree 63 to 65% of the time.

That’s the big question this experiment will answer on election night: does the briefing make the AIs smarter, or just make them repeat the polls?

The AIs are far more certain than the odds

Our model says Talarico has a 65% chance. That means Paxton wins about one time in three. The briefed AIs gave Talarico all 24 of their clear scenes. In most races, every briefed scene picked the same winner.

That’s a known habit of these models: they like a clean story. It’s also why we only mix the AIs into our forecast at 20 percent, and only on the switch at the top of the page.

One AI won’t play along half the time

When Claude writes blind, it declines about 45% of the time. Sometimes it refuses outright. Sometimes it adds a note first, like this one from a Maine scene: “I can’t know how the real Maine Senate race will turn out, and I don’t want to present an invented result as a real projection.” With the briefing, it declines only about 17% of the time.

ChatGPT has not declined a single scene.

We count every refusal and never ask again to get a better answer. How often each AI balks is part of the story.

Where you can see it

On the members forecast, every race page now has an AI Anchor Desk card: each AI’s chance, how many scenes it rests on, and one sample scene to read. At the top of the page, one switch mixes the AIs into our model, blind or briefed, so you can watch how the Senate and House odds change.

Right now, mixing in the blind AIs pulls the Democrats’ Senate chance down to about 59%. Mixing in the briefed AIs pushes it up to about 71%. Our model alone says 66%.

These are early numbers. Each AI’s chance in a race rests on a handful of scenes so far, and the pool grows every day.

On November 3, we’ll score all of them: our model, the blind AIs and the briefed AIs. Then we’ll know whether the machines saw something we didn’t.

See every race, and every AI call: $56 for your first year (offer ends Nov. 4)

— Justin

A handful of outlets publish a number. Democrats 71%. Republicans 58%. The number gets quoted on cable, and then it gets quoted as fact. Almost nobody asks how it was made.

Then election night comes, and half the time the number was wrong in a way nobody warned you about.

I don't have a problem with forecasts. I have a problem with forecasts you can't check.

So we built one you can.

What it is

The Rational Ground 2026 Forecast covers all 506 races on the November ballot: 35 Senate seats, 36 governors and all 435 House districts.

Every day, the model takes in the newest polls, campaign money and district maps. Then it plays out the election 20,000 times. If Democrats hold the Senate in 13,000 of those runs, their chance is 65%. That's all a forecast is: a count of how often something happens when you play it out.

Every race shows what produced its number: the polls behind the average and how much each one counts, the money each campaign has raised, every outside group that just started spending, and how the district has voted before.

And the full method is written down on the page, in plain English, with every setting the model uses.

How it works

Five ingredients go in.

The polls. As of today, 866 polls on individual races. Some pollsters lean one way, year after year, so we measure each pollster's lean against everyone else's this year and adjust for it. A poll from a firm that always runs three points hot gets cooled off.

The national mood. There are another 473 "generic ballot" polls asking which party people will vote for in Congress. That's the tide that lifts or sinks every boat. Right now it sits at Democrats +6.8.

Each place's lean. How a state or district voted before: 70% the 2024 presidential race and 15% each the 2018 and 2022 midterms. We didn't pick those weights out of the air. We tested them against what actually happened in 2018 and 2022 and kept the mix that called those years best.

Incumbents. Sitting members get a small bonus, plus part of their own track record. A governor who keeps winning in a state that votes the other way, like Phil Scott in Vermont, has earned that.

Money and maps. We pull campaign money straight from the Federal Election Commission's filings and flag every surge in outside spending, because the big groups often know something before the polls do. And 173 House districts have new lines since 2024. Texas, California, Florida and others redrew mid-decade. We score every seat on its new map, not its old one.

What it says today

As of October 8:

• Senate: Democrats have a 65% chance of control. Republicans 30%. About 5% of the time, neither party gets a majority on its own.

• House: Democrats 91%.

• Governors: Democrats are expected to hold about 26 or 27 of the 50 seats.

I didn’t build this to tell you what you want to hear. I built it to tell you what the numbers say, and to show you exactly why.

The reason is mostly that national mood. A generic ballot of Democrats +6.8 is a strong year for the party out of the White House. If that number moves, everything moves.

The Senate runs through five states. In the races most likely to decide control, the tipping point is Texas, Ohio, Alaska, Michigan or Maine. Three more are close to a coin flip: Kansas, Iowa and Nebraska.

These numbers will change. That's the point. The page updates every day, and members can watch it move.

The experiment: the AI Anchor Desk

This is the part I'm most excited about.

In 2024, two economists, Van Pham and Scott Cunningham, found something strange. When they asked ChatGPT directly to predict the future, it was mediocre. When they asked it to write a story set in the future, a scene where someone looks back and reports what happened, its predictions got noticeably better. I wrote about their paper at the time on Hello AI.

So we're putting that to a real test.

Every day, we ask Claude and ChatGPT to write the election-night broadcast for every major race. A made-up network, the Continental News Network. A made-up anchor, Dana Whitfield. "We now have a projection to make in the race for U.S. Senate in North Carolina…" And then the AI writes who wins.

We run each scene many times and count. If ChatGPT calls the race for one candidate in 7 scenes out of 10, that's a 70% chance in ChatGPT's eyes.

Each AI writes two versions:

• Blind: just the race, the candidates and the date. Whatever the AI "believes" on its own.

• Briefed: we also hand it our polling average, the latest polls and the money. But never our odds, so it can't just copy us.

Here's why this is a fair fight. In the original study, the trick worked best when the answer was already somewhere in the AI's training data. November 3, 2026 isn't in anyone's training data. Nobody, human or machine, knows how this ends.

On the members page, one switch lets you see our model alone, or our model with the AIs mixed in. On election night, we'll know whether the machines saw something we didn't.

What members get

• The full forecast: every race, updated every day

• A page for each race, with the polls, the money, the history and the AI scenes

• "Money moves": every outside group that suddenly starts, or stops, spending in a race

• The new district maps, and which party each one helps

• The full method, in plain English

It lives behind the member wall at members.rationalground.com/elections.

If you've been reading Rational Ground for free, thank you. If you want to watch this race the way we do, and you want this work to keep going, this is the moment.

Become a member: $56 for your first year instead of $70 (offer ends Nov. 4)

They'll tell you who's going to win. We'll show you the math.

— Justin

Read more