What Predictive Averaging Can and Cannot Establish: A Standard Decision-Theoretic Note for KWM Research v0.3

Explains why the conditional-mean squared-error identity does not certify physical identification, calibrated posterior weights or control optimality.

Reviewed TeX was found but no existing PDF was located; the catalogue provides notes and source rather than inventing a PDF.

Why we tried it

Clarify what squared-error properties of predictive averaging can establish.

What the evidence shows

The companion note states the standard conditional-mean squared-error identity.

Where the work stands

The identity is not a new theorem and does not certify physical identification, posterior calibration or optimal control.

Possible next steps

Test physical parameters, belief calibration and task decisions separately rather than substituting a predictive-mean identity.

Suggested next steps are untested research questions, not new experimental results. The archive records manuscript provenance, licensing and status.