The Fauci Prediction, Checked: We Called It and We Were Wrong About the Part That Mattered Most

A prediction dated in advance only means something if you’re willing to stamp both cards. Image: Allison Robbert/AP

Introduction

Five weeks ago, this publication made a specific, dated, falsifiable prediction about how Anthony Fauci would handle his Senate testimony on 29 July. The prediction was precise enough to be checked against a transcript, which is the whole point of making a prediction like that in the first place rather than a vaguer, safer one. Yesterday, the hearing happened. The prediction was wrong, cleanly, specifically and on the exact point that mattered most.

This piece exists to say that plainly, grade it honestly and separate what actually failed from what didn’t.

The Call, Restated in Full

The original piece’s prediction, quoted exactly as published on 25 June: “Fauci will not invoke the Fifth Amendment as his primary posture in July. He will appear, answer at length and present his conduct as good faith scientific judgement under conditions of extraordinary scientific uncertainty and political pressure, rather than as concealment.” The piece named this the “Oliver North” scenario, a witness with a pre-emptive pardon using the hearing as a rehabilitation platform rather than a shield for silence, on the theory that a pardon removes the legal cost of speaking and therefore removes the incentive to stay quiet.

That was the bet. It was specific on purpose.

What Actually Happened

Fauci’s opening statement invoked the Fifth Amendment, in language that made the reversal explicit: “Although it pains me to do so, because of the respect I have for the legislative branch of government, and my decades-long record of cooperating with Congress, under the advice of my attorneys, I will invoke my right under the Fifth Amendment.” That framing sentence was for the opening statement alone. For nearly every question that followed, across a hearing that ran well over two hours, he repeated a tighter, 24-word version instead: “Under the advice of my attorneys, I will invoke my right under the Fifth Amendment of the Constitution to refrain from answering your questions.” He said some version of that sentence more than 100 times. He characterised Chairman Rand Paul’s investigation as an “obvious obsession” aimed at getting him “behind bars,” and said the hearing’s real purpose was to trap him into statements that could support a perjury prosecution. His attorney, David Schertler, was forcibly removed from the hearing room by Capitol Police after calling the proceedings “false and disgraceful,” to applause from the room. Paul, for his part, told Fauci directly and on the record that refusing to answer pertinent questions under subpoena is illegal, citing Title 18, Section 192 of the US Code, and has since announced the committee will vote next week on whether to hold Fauci in contempt, while separately raising questions about the scope of Fauci’s pardon itself. Democratic committee members pushed back publicly during the hearing, arguing Fauci did not deserve the treatment he was receiving, a reminder that the hearing’s reception split along predictable partisan lines even where the underlying facts are not seriously in dispute.

None of that is the North playbook. North was expansive, confident and combative in the way a man behaves when he believes speaking freely helps him. Fauci was the opposite: guarded, repetitive and visibly unwilling to engage with a single substantive question. The prediction called for a performance. What happened was a wall.

Five for five, the transcript ran the opposite direction from the call. All images generated by author unless stated

The Scorecard, Without Softening It

Graded against the original piece’s own stated terms:

Did not decline to testify / did not rely primarily on the Fifth Amendment — FAILED. He appeared under subpoena, which the piece got right, but relying on the Fifth was not merely a posture, it was the entire posture from the first statement onward.

Appeared and answered at length — FAILED. He appeared. He did not answer.

Framed conduct as good faith scientific judgement under uncertainty, rather than concealment — FAILED, in the specific form predicted. He did characterise the hearing itself as unfair and politically motivated, which is adjacent to the reframing the piece expected, but he did not do this by discussing the substance of the allegations at length the way North discussed Iran-Contra. He did it by refusing to discuss the substance at all.

Hearing functioned as a platform for re-characterisation rather than a forum producing clean resolution — PARTIALLY TRUE, for reasons the piece didn’t anticipate. The hearing did not produce a clean resolution. But it didn’t fail to resolve things because Fauci successfully reframed the narrative in his favour, the way North did. It failed to resolve things because he said almost nothing at all, leaving the field open for Paul and public opinion, to fill in the silence however they choose.

Four elements of a five-part prediction, and the honest grade is four clear misses and one technical partial that arrived by an entirely different road than the one the piece described.

What the Piece Actually Got Right, Held at Arm’s Length From the Miss

This is worth stating carefully, because it would be easy to let a genuine, separate insight quietly rescue a prediction that failed and that’s exactly the move this publication doesn’t make. The original piece’s deeper argument was never really about performance style. It was about a specific legal mechanism: a pre-emptive pardon removes liability for the offences it covers, but it does not remove Fifth Amendment exposure for anything outside that scope, meaning the witness retains a real, narrow and genuine incentive for caution wherever the questioning reaches ground the pardon doesn’t cover.

That mechanism held up today, independently of the failed behavioural prediction built on top of it. Paul’s live, on-the-record citation of Section 192, telling Fauci to his face that blanket refusal is itself illegal, is exactly the legal collision the original piece predicted the pardon’s actual scope would eventually produce. And Fauci’s own stated fear, in his opening statement, that the hearing was designed to trap him into a statement supporting a perjury charge, is a direct, unprompted confirmation of the piece’s central claim: the pardon does not protect him from the consequences of what he says today, only from what he already did before it was signed. That fear wasn’t paranoia manufactured on the spot, it had already been stated publicly, in almost identical terms, by the one member of the cabinet with the most direct institutional interest in this exact question. Health and Human Services Secretary Robert F. Kennedy Jr., whose department spent eight months recovering Fauci’s diary from eleven separate servers, told reporters earlier in the week that Fauci could face prosecution regardless of the pardon: “There are many things that he said that we know are lies,” adding plainly, “if he lies again, then he could be subject to perjury prosecution.” Fauci walked into a hearing room where the exact legal exposure he described in his opening statement had already been named, in public, by the government’s own health secretary, days before he sat down. The mechanism was right. What it predicted a rational actor would therefore choose to do with that mechanism was wrong. Fauci, it turns out, judged near-total silence as the lower-risk posture rather than confident reframing, which may say more about what’s actually sitting in that diary than the original piece, working from public information alone, could have known five weeks ago.

One hand has already stopped. The other is still moving.

A Second Dated Prediction, While the First One Is Still Being Graded

One prediction has already been checked and found wanting. Here is a second one, made today, specific enough to check again once there’s a record to check it against: no contempt-of-Congress prosecution will actually proceed against Fauci within the next twelve months, despite Senator Paul’s on-the-record citation of Title 18 Section 192 and despite any formal contempt vote the House may take.

The reasoning is structural, not personal, and it has a precedent already sitting inside this piece: Lois Lerner was found in contempt of Congress by a 231–187 House vote in 2014, and the Department of Justice, months later, declined to bring any criminal charge over it at all. Paul’s committee is now scheduled to vote next week on the same question for Fauci, a concrete, near-term checkpoint this prediction can be graded against far sooner than the original twelve month window suggested, since a contempt vote itself is very likely to happen quickly even if the prediction is that a subsequent prosecution will not. The pattern this suggests is not that contempt findings are toothless in general, it’s that they tend to stay toothless specifically when the person in the dock is not the only one exposed by what a real prosecution would have to establish. If pursuing Fauci seriously requires proving, in a courtroom, exactly how much the intelligence community’s origin assessments were shaped by his own preferred experts, and exactly who else signed off on the gain-of-function funding chain along the way, then the number of currently serving and recently serving officials with an interest in that chain of facts never being established in a court of law is very unlikely to be limited to one 85-year-old former agency director. A prosecution that stays contained to Fauci alone protects everyone standing next to him. A prosecution broad enough to actually resolve the underlying allegations doesn’t stay contained to Fauci alone. That asymmetry, more than any specific legal defence, is the honest favourite to determine what happens next.

This is stated here, dated, before it’s known, the same discipline the original prediction was built on, applied a second time immediately after the first attempt missed.

Two Methods, Both on the Record

There’s a detail worth adding honestly, because it changes what this piece is actually documenting. Before this hearing happened, two separate predictions were made about how Fauci would handle it. This publication’s prediction, built from the pardon’s legal mechanics, publicly available information and an assumed rational-actor model said he would not invoke the Fifth. That prediction failed. A second, independent prediction, made by the same author using a completely different method, said he would invoke it. That prediction was correct.

The second method wasn’t structural analysis. It was a longitudinal character read: years of following this specific story from its earliest public mentions in November 2019, independent knowledge of gain-of-function research and Fort Detrick’s history acquired well before most of the public was paying attention, and a direct, physical reaction to watching Fauci speak that the author describes as involuntary, an inability to hold eye contact, and sometimes to speak at all, in the presence of someone read as being untruthful. That’s not a claim this piece can verify or quantify and it isn’t offered as evidence in the way the sourced facts elsewhere in this cluster are. It’s offered as exactly what it is: a different epistemic method, run in parallel with the analytical one, that happened to produce the correct call this time.

The honest research on demeanour-based lie detection is not kind to this method in general, large studies of even trained interrogators find that judging deception from someone’s face and body alone performs barely better than chance. But that finding describes strangers making snap judgements from limited exposure. It doesn’t describe someone who has tracked one specific case, and one specific man, for years, and combined that sustained exposure with a character read rather than a single glance impression. That’s a different and better supported category of intuition, closer to an experienced investigator’s read on a suspect they’ve worked for months than a stranger picking a liar out of a lineup.

Both predictions are on the record now, made independently, before the hearing happened. The structural method got the mechanism right and the specific behaviour wrong. The character-based method got the behaviour right. Neither result proves its method is generally superior, one correct call and one incorrect call is not a sample size that settles anything. What it does establish is that they are genuinely different tools, capable of reaching different conclusions from the same set of public facts, and that this time, the tool built from watching a specific person rather than modelling a generic rational actor was the one that worked.

The Verdict

What this piece is grading, twice over: One specific, dated, falsifiable prediction about testimonial posture, which failed on its central, headlined claim and one new specific, dated, falsifiable prediction about institutional consequence, made here for the first time and left for the next instalment to grade.

What survives the failure, stated as a separate and distinct finding, not a rescue of it: The underlying legal mechanism the prediction was built on, that a pre-emptive pardon creates real, live exposure beyond its own scope and that this exposure would eventually surface as visible tension in the hearing room, was independently confirmed on the record, by the chairman’s own words and by Fauci’s own stated fear. Getting the mechanism right and the behaviour built on top of it wrong are two different outcomes and this piece is not going to blur them into one softer story.

The method this publication runs on requires publishing calls before they’re checked and grading them honestly afterward, whichever way they land. Today’s grade is a clear miss on the thing that mattered most, and an honestly separate, independently-confirmed hit on the thing underneath it. Both are on the record now.

This piece is a companion to “[The Fauci Prediction — Why He Won’t Plead the Fifth,](http://Five for five, the transcript ran the opposite direction from the call. All images generated by author unless stated)” published 25 June 2026 and available in this publication. It grades that piece’s dated prediction against the 29 July hearing transcript and contemporaneous reporting.