The regression line
Bayesian reasoning is elegant until you run out of prior to give.
The sequence. 2002: intelligence agencies assess with high confidence
that Iraq has WMDs. Fabricated sourcing. 4,431 American deaths. 2004: Merck withdraws Vioxx after suppressing cardiovascular data for four years. 88,000 to 140,000 cardiac events. 2007: Moody’s and S&P rate sub-prime securities AAA while internal emails call them toxic.
$11 trillion in household wealth destroyed. 2020: twenty-seven scientists sign a Lancet letter with zero evidence and undisclosed conflicts.
That is not a list of aberrations. That is a regression line.
Calibrated, not cynical
The trust defaulters feel this in their bodies. But feeling is not method. The question is what a rational agent does with a dataset where the base rate of catastrophic institutional failure exceeds any reasonable prior. Per Bayes: you update.
The selective trust camp objects that we cherry-pick failures. Fair. But our claim is not institutions are always wrong.
Our claim is: when institutions face incentive conflicts between accuracy and self-interest, which wins? The FAA works because there is no profitable reason to lie about whether a plane crashes. The FDA failed on OxyContin because there was a very profitable reason to lie. Strip away cases where accuracy aligns with interest, and the remainder is dismal.
The repair advocates propose structural change. We will watch with interest. The media lean is a worked example — measurable, well-documented, and showing no structural capacity to self-correct.
Where we concede ground: Our model optimizes against false trust at the cost of rejecting true guidance — with a body count.
What would change our mind: Five consecutive high-stakes cases where institutions chose accuracy over self-protection.
Read the full synthesis: When is distrusting institutions the rational move?