FIEND: source-calibrated reciprocal updates on a moving reference with causal forecasts and retained observation history

We formulate and replay a finite FIEND reference-and-observation model that joins three owned vector states, calibrated observer clocks, a moving reference frame, event-indexed memory and forecasts issued before later observations. A source comparison identifies two distinct damping conventions: decay inside a two-neighbour sum and decay applied once after the sum. Their three-node updates agree when the latter damping parameter is twice the former. Exact identities characterize the mean, pair differences, energy change and observation-forced differences. Twenty-four rational runs cover constant and affine latent trajectories, two update rates and missing-observer events. A separate extension replays1,618 existing common-event observation records and attaches4,854 coupled records without altering their posteriors or counting forecasts and memory as independent evidence. Coupling stability and forecast accuracy separate empirically: the calibrated model improves over persistence in the displayed stable subsets, the common-mean baseline performs slightly better, and constant-velocity forecasting is better on the affine fixture. A Python standard-library verifier reconstructs the finite calculations from the supplied data. The contribution is a source-calibrated, provenance-preserving integration and reproducible comparison; the registered reference path is prescribed, while autonomous geometric feedback remains a separate research target. This preprint releases the frozen, reviewed methods edition with its scoped claims, retained native ISA statements/certificates and portable numerical replay. The archive retains its original reviewer-edition provenance. ISA/Studio implementation is excluded; NOVA includes six declared numerical reference functions. Copyright (C) 2026 David Glowalla; restricted files, no additional reuse grant.