It is a signal — not a proven link

A reading of vaccine-safety reporting across five national systems, computed from the raw data. The contention is narrow and exact: these are signals, the thing surveillance exists to find — not proof of cause, and not a claim of mass harm.

1. The claim being answered

The recurring move is to say spontaneous reporting “cannot establish a link between the vaccine and the symptom.” True — and beside the point. VAERS and its peers were built by health agencies for one job: detecting safety signals worth investigating, not proving cause. Faulting them for not proving cause is faulting a smoke alarm for not putting out the fire. And as signalers they work: VAERS caught the RotaShield intussusception signal in 1999 (vaccine withdrawn within months) and the mRNA myocarditis signal in 2021 (confirmed by controlled studies, labels updated).

2. The signal converges across five independent national systems

Convergence of four signals across five systemsMyocarditis pericarditis GBS male-skewed everywhere; anaphylaxis female-skewed everywhere.signal \ systemVAERS·USMHRA·UKTGA·AUJADER·JPCVAR·CAMyocarditis6.64.23.15.53.4Pericarditis2.64.32.55.72.6GBS23.11.61.93Anaphylaxis (control)0.40.60.40.40.2values = sex reporting-odds-ratio · red = male-skewed (>1) · blue = female-skewed (<1)within COVID-vaccine reports across five national systems — signals, not proven links
Sex disproportionality (reporting-odds-ratio) within COVID-vaccine reports. Myocarditis, pericarditis and GBS are male-skewed in every system; anaphylaxis is female-skewed in every system — the method discriminates, matching known biology, so it is not flagging everything. A pattern that replicates across five systems with no shared reporting infrastructure is not a US artifact.

3. The signal clusters by organ system

Signal clustering by organ systemCardiac and vascular events male-skewed across systems; anaphylaxis female-skewed.signal \ systemVAERS·USMHRA·UKTGA·AUJADER·JPCVAR·CACardiacMyocarditis6.64.23.15.53.4Pericarditis2.64.32.55.72.6Cardiac arrest3.13.82.2n/a6.9Myocardial infarction343n/a1.8Vascular / clottingPulmonary embolism1.82.61.6n/a1.5Thrombosis1.62.41.41.71.3Cerebrovascular / stroke1.82.72.2n/a1.8Thrombocytopenia1.82.51.21.61.9NeurologicalGBS23.11.61.93Facial paralysis1.21.40.5n/a0.8Seizure / convulsion2.31.72.4n/a0.8SystemicAnaphylaxis (control)0.40.60.40.40.2values = sex reporting-odds-ratio · red = male-skewed · blue = female-skewed · n/a = too few reportscardiac + vascular cluster co-signal; anaphylaxis is the discriminating control — signals, not links
The same disproportionality across twelve events grouped by organ system. The cardiac cluster and the clotting/vascular cluster co-signal across the federation; anaphylaxis remains the female-skewed control. Co-signaling on a shared biological pathway is a hypothesis to investigate — not a proven mechanism.

4. It emerged at one moment in history — in every system at once

VAERS report volume and myocarditis share 1990-2026Stable for decades then a 2021 surge; myocarditis share rose far more than volume and stayed elevated.RotaShield pulledH1N1 pandemicCOVID mRNA rollout19902000201020212026bars: all VAERS reports / year (peak 769k in 2021)line = myocarditis per 100,000 reports — the signal, normalized for volume · rose ~27x in 2021 and stayed up
VAERS ran at 10–50k reports/year for three decades, then surged to 769k in 2021. But myocarditis is not just “more reports”: its share of all reports rose ~27x in 2021 after normalizing for the surge, and stayed elevated as dosing continued — tapering as dosing fell.
Myocarditis share by year in five systems, each own scaleAll five rise sharply at 2021; vertical scales differ and are not comparable.VAERS·USpeak 1082021MHRA·UKpeak 1,2042021TGA·AUpeak 1,5192021JADER·JPpeak 1,3712021CVAR·CApeak 6,7012021myocarditis per 100k reports · each panel on its OWN scale (levels not comparable) · convergent finding is the timing
The same emergence in five national systems independently, each on its own scale. Levels are not comparable across systems (different report mix, coding, volume); the convergent finding is the timing — all five rise sharply at 2021.

5. Signals like this have a track record — and the system tells on itself

Confirmed effect size and regulatory response for seven signalsStrong confirmed signals led to action; a refuted signal led to none; Vioxx acted late, LYMErix over-reacted.no effect (RR 1)RotaShieldintussusception37xPandemrixnarcolepsy12xSwine flu '76Guillain-Barre7xCOVID mRNAmyocarditis5.3xVioxxcardiovascular2xLYMErixarthritis (unconfirmed)1xMMR vaccineautism0.9x1x2x5x10x20xconfirmed relative risk (log scale)confirmed, actedmis-timed: Vioxx too slow, LYMErix over-reactedrefuted, no action
Confirmed effect size (from controlled studies, not VAERS) and the response. Strong confirmed signals drove withdrawal or warnings; MMR–autism, refuted, correctly drove none. The two amber cases are the honest scars: Vioxx, where rationalizing the signal away cost four years; LYMErix, withdrawn over a signal that never confirmed. A chart showing only confirmed harms would be propaganda; one that shows the misses in both directions is evidence.

6. The one thing all five share: no denominator

The civic ask

Every spontaneous system here is a numerator with no denominator. You can see the alarm; you cannot compute a rate — “deaths per million” or “cases per million doses” is structurally impossible from any of them. The denominator exists: CDC dose-administration data, the V-safe active cohort, national health registries — much of it publicly funded. It is simply not surfaced alongside the signal in real time; it is reconstructed years later from private claims, registries, and the Vaccine Safety Datalink.

So the public gets the alarm now and the rate later, if ever. As taxpayers who funded both the products and the surveillance, the reasonable demand is: join the numerator to the denominator in real time — report counts against doses administered, by age, sex, dose and week — so signals can be converted to rates promptly and reproducibly, by anyone. That is what closes the argument in either direction.

7. Protecting trade secrets without hiding the truth

zero-knowledge proof — feasibility

What is already real: the provenance half. Every row carries a content hash (gn_sha256); files carry a manifest; the corpus uses Merkle roots and chain anchors and dated, hash-locked snapshots. That is the commitment layer a zero-knowledge proof builds on — you can already prove which exact data a result came from, and that it was not altered.

What is feasible with real engineering: proving a computed statistic (a disproportionality score, a denominator-adjusted rate) was correctly derived from a committed private dataset without revealing the rows; proving a record is or isn’t in a set without exposing it. This lets a data holder — a manufacturer guarding trade-secret trial data, or an agency guarding identifiable records — publish a verifiable answer while keeping the inputs sealed. So “we can’t share it” stops being a reason to withhold the answer.

What is aspirational, stated honestly: zero-knowledge proofs over large statistical pipelines (millions of rows, regressions) are computationally expensive today, and a proof attests that a computation was done correctly — not that the sealed input was honest, unless that input is itself independently attested. The realistic path: ship the cheap layer now (commit + attest + reproducible recompute), pilot a ZK aggregate/membership proof on one targeted claim, and treat full ZK statistical pipelines as a research track, not a turnkey feature.