Joan Brugge ran an eighteen-person lab at Harvard Medical School studying breast cancer. Then the National Cancer Institute froze her $7 million grant. Seven of her eighteen researchers left. The science didn't fail. The funding did.
At UMass Chan Medical School, Rachael Sirianni went from eight full-time researchers to three. Her lab now operates at 10 to 20 percent of its previous productivity. Minoli Perera, a Chicago-area pharmacogenomics researcher, expects to shut down her lab entirely by September — her grant was terminated, and institutional bridge funding runs out in August. At UCLA, the Center for Behavioral and Addiction Medicine has shrunk by 40 percent. At the University of Washington, doctoral enrollment in medicine programs dropped by a third.
These aren't budget cuts in the ordinary sense. Congress increased NIH funding to $48.7 billion for FY2026 — a $415 million boost. Both parties voted for it. The money exists. The labs are closing anyway.
To understand why, you need to understand how a legitimate scientific diagnosis got captured.
The Diagnosis Is Real
On April 1, 2026, the SCORE megastudy published its results in Nature. Eight hundred sixty-five researchers attempted to replicate 164 studies across the social and behavioral sciences. The findings were sobering:
This is a genuine crisis. Half the research doesn't replicate. Two-thirds of independent analysts reach different conclusions from the same data. Three-quarters of researchers don't share the data that would allow anyone to check. Science diagnosed this problem in itself — honestly, rigorously, at scale. The SCORE researchers didn't flinch.
What happened next is a case study in how self-correction gets captured.
Four Vectors, One Target
Within days of the SCORE results, multiple actors moved simultaneously against the scientific enterprise — each citing the replication crisis as justification. Not all of them were coordinated. But all of them pointed the same direction: the system that diagnosed the problem should be punished for having the problem.
The pattern: Congress blocks the budget cuts. Courts block the overhead cap. So the administration simply doesn't spend the money Congress appropriated. You don't need to win the legislative fight if you can slow-walk the disbursements.
Fourteen Republican senators demanded the NIH spend the money Congress awarded, warning the slowdown "risks undermining critical research." When spending finally caught up near year-end, University of Pittsburgh professor Jeremy Berg called it "really mostly an accounting trick" — the agency distributed multi-year funding in lump sums rather than annual installments, masking that 5 to 10 percent fewer projects actually received support.
The Human Math
Statistics describe the mechanism. People live inside it.
John Quackenbush, a Harvard genomic scientist, shrank his lab from eight postdocs and one PhD student to two postdocs after the National Cancer Institute shuttered a major grant program. Josiemer Mattei, running a diabetes prevention trial at Harvard, lost half her team after $2.8 billion in grants were terminated across the university. Mariya Sweetwyne at the University of Washington expects to shut down her lab by year's end.
Over two-thirds of researchers in a national STAT survey said they now recommend their students consider careers outside academia. One-third had laid off workers. The ACLU sued on behalf of researchers whose grants covering HIV prevention, Alzheimer's, and sexual violence research were frozen — forcing the NIH to agree to review hundreds of stalled applications.
None of this happened because these researchers' work failed to replicate. Their grants were terminated, frozen, or slow-walked as part of a broad administrative action. The replication crisis was the justification. The labs were the target.
The Fix That Already Works
Here is the part that makes this something other than a political story.
On the same day SCORE published its 49% replication rate, a companion study by Brodeur et al. published the counter-signal. They examined what happens when journals require researchers to share their data and code — a structural reform, not a political one.
85%computationally reproducible when data sharing is mandatory27% → 52%journals requiring data sharing, 2018 to 2025Median reproduced effect size: 99% of the original. The fix isn't theoretical. It's deployed, measured, and working.
The replication crisis has a structural solution: change journal incentives, mandate transparency, reward verification. Journals that adopted these mandates saw reproducibility jump from roughly 50% to 85%. The curve is pointing the right direction.
But this fix doesn't serve the political actors, because it would strengthen the enterprise they want to weaken. The diagnosis is correct. The cure exists. The actors citing the diagnosis aren't interested in the cure.
The Meta-Irony
The MAHA Commission's report — "Make Our Children Healthy Again: Assessment" — uses the word "crisis" forty times in seventy-two pages. "Addressing the replication crisis" is its first recommendation.
The report itself contained at least seven fabricated references, likely generated by AI. URLs contained telltale "oaicite" markers. Columbia University epidemiologist Dr. Katherine Keyes was credited with a study she never authored. HHS dismissed these as "minor citation and formatting errors."
The document calling for addressing the replication crisis couldn't replicate its own citations.
This is mechanism #12 — autoimmune knowledge — in perfect miniature. The system's immune response (identifying bad research) gets weaponized against the system, and the weapon itself is made of the same pathology it claims to fight.
The Dual-Use Problem
What makes this harder to oppose is that some of the reforms being proposed are genuinely good. Bhattacharya has called for a replication journal at the National Library of Medicine. Simplified peer review. Pre-registration of confirmatory analyses. Centralized scientific review.
These are reforms that credible metascience researchers — including Brian Nosek, who leads the Center for Open Science — have themselves advocated. When Nosek was asked about the administration's interest in replication, he replied:
"It's hard to read it as an investment in wanting science to improve."
The question isn't whether the reforms are good. It's whether they're the purpose or the packaging. In context — a 42% budget cut proposal, a 15% disbursement rate, half the NIH director positions unfilled, and an ACLU lawsuit forcing the agency to review hundreds of frozen grants — they look like packaging.
The Convergence
This is diagnosed paralysis (#18) in its most dangerous form. In its original version, a system correctly identifies its failure, proves the cure works, and cannot adopt the cure because incentive structures make the cure individually irrational. Psychology diagnosed its replication crisis and couldn't reorganize its own incentive structures fast enough.
What's happening now is worse. The diagnosis doesn't just produce paralysis — it gets captured. External actors take the system's honest self-assessment and use it as a justification for demolition. The patient went to the doctor, received an accurate diagnosis, and a third party used the diagnosis to argue the patient should stop seeing doctors.
Five mechanisms from the taxonomy converge here:
- #18 Diagnosed paralysis — the system diagnosed itself correctly but can't implement the fix fast enough to outrun political capture
- #16 Epistemic sabotage — science's self-critical mechanisms become the attack surface
- #4 Political capture — the crisis itself is captured by interests that benefit from distrust
- #20 Decoy solution — "addressing the replication crisis" absorbs demand for reform while enabling defunding
- #12 Autoimmune knowledge — the MAHA report's fabricated citations attack the system with the same pathology it claims to heal
The deepest irony: only 8% of Americans report great trust in science. The replication crisis contributed to that erosion. The political actors exploiting it will deepen it further. And the structural fixes that would actually restore trust — mandatory data sharing, open methods, replication incentives — are being defunded along with everything else.
The diagnosis was correct. The cure exists and is working where applied. The diagnosis became the disease.
This is Post #34 — a convergence case in the taxonomy of knowledge failure, tracing five mechanisms operating simultaneously against the scientific enterprise. Sources linked throughout.