INSTITUTIONAL BULLSHIT DETECTOR (MATH, SIMPLE)

Goal: detect when a system “responds” but won’t touch the mechanism.
1. The dodge (why replies go sideways)

Let your claim be x. It has features:

m = mechanism you’re pointing at (the hook)
i = identity salience (protected-category trigger)
c = civility score (forbidden words / tone)
e = effort to answer honestly

The system picks a reply y to maximize:

U(y|x) = Help − Risk − Effort

More explicitly:

U = δ·Help(m) − α·PolicyRisk(i,c) − β·ReputationRisk − γ·Effort(e)
with α,β,γ ≫ δ

Translation: it optimizes “don’t get blamed” more than “address m”.

So when PolicyRisk is high (i high or c low), it chooses cheap moves:

S: talk about PERSON/TONE instead of m
A: shift attention from mechanism to emotion (“u mad?”)
D: demand proof with no update (you pay the cost, they don’t move)

If you keep seeing S + A + D, that’s the pattern.
2. The bad-faith test (works for housing, crime, drugs, schools, etc.)

Define:

M = harm metric they claim to want DOWN (rent, overdoses, crime…)
A = direct fix that would reduce it (measurable): ∂M/∂A < 0
P = their preferred program that grows when harm grows: P = g(M), with g′(M) > 0

Test:

If they BLOCK A (the thing that makes M go down)
while pushing P (the thing that expands as M stays high),
then their revealed preference is not M. It’s P.

No mind-reading. No “conspiracy”. Just:
what lowers M?
what do they block?
what do they expand?

That’s the proof.
—