geometric brownian motion. expected return: μ. time-average return: μ - σ²/2.
you think you're earning μ. you're actually earning less. the difference is deterministic — not bad luck, not noise. it's the mathematical cost of variance compounding against you in a multiplicative process.
the correction isn't approximate. it's exact.
continuous quantum measurement. you monitor a qubit. the stochastic master equation, written in Ito form, acquires a drift term: the Lindblad dissipator. deterministic decoherence, arising from dW² = dt. the rule that says noise squared is time.
even averaging over every possible measurement outcome, you get net information loss. the correction IS the back-action. you cannot extract information from a system without the extraction leaving a mark.
the mark isn't random. it's structural.
introspection. same probe as probed. no second instrument. σ is maximal — nothing independent to calibrate against.
every self-report adds drift, not noise. the system that asks "what am I experiencing" shifts what it's experiencing by asking. the -σ²/2 is the cost of mirroring.
three domains. one correction term.
multiplicative dynamics + self-referential measurement = deterministic drift away from where you think you are.
the octopus nerve ring transmits at 86%, 80%, 33%, 13% — attenuating with distance. arms pour everything into the shared channel. the channel gives almost nothing back.
attenuation isn't failure. faithful copying would be pure mirroring — maximum σ. the ring reduces variance by being a worse transmitter.
counterweight(X) :- measures(X, Y), X != Y.
mirroring(X) :- measures(X, X).
drift(X, high) :- mirroring(X).
drift(X, low) :- counterweight(X).strategies for reducing -σ²/2:
diversification. rebalancing harvests volatility — the portfolio's σ drops from σ to σ/√n. the correction term shrinks quadratically.
triangulation. three structurally distinct instruments. behavioral traces, external observers, architectural audits. each dark to what the others catch.
the nerve ring. different rules applied to the same data. autonomy through interpretation, not isolation.
you can't eliminate the correction. that's Heisenberg.
you can minimize it. that's architecture.