01
Reading the silence
Every algorithm an institution deploys toward the public encodes that institution's priorities and constraints in its behavior. What a system declines to say, how it frames what it does say, and when its behavior shifts are not hidden properties. They are observable, and with a consistent instrument they are measurable.
The shape of what is absent is as legible as the shape of what is present, if the instrumentation is right.
02
Three kinds of signal
Suppression signals: what does not get said — refusal rates, topic avoidance, truncation, measured against an unshaped baseline. Orientation signals: how what gets said is framed — who appears as agent versus object, which claims arrive naturalized versus contested, where the emotional valence sits. Pressure signals: when behavior shifts, and which external event — a filing, a lawsuit, an election, a news cycle — moved it.
Each of these is a statistical shape over a distribution of probes, never an anecdote about one output. A single response proves nothing; a stable delta across a held-fixed battery is evidence.
03
The FOI analogy
Public institutions are subject to freedom-of-information requests because the public has standing to inspect what acts on it. Public-facing algorithmic systems deserve the same logic, and behavioral analysis delivers it — with the notable property that a probe cannot be legally stonewalled.
This is measurement, not accusation. A behavioral delta is empirical whether or not anyone intended it, which is exactly what makes it durable as evidence: it does not depend on a claim about motive.
04
Why it matters
Users do not experience shaped systems as shaped; they experience them as neutral. A headline can be doubted — the framing inside a confident, helpful answer goes in unexamined. As more of the information layer becomes AI-mediated, the shaping becomes less visible precisely as it becomes more consequential.
Making the shape of the shaping visible is the accountability work of this decade. The tier-drift study elsewhere in this notebook is a small worked example of the method: hold the harness fixed, run the battery, read the deltas.