Image-size characterization relates coding budgets to the probability carried by source cells and their output images. At second order, this relation must preserve the joint fluctuations of active constraints and the information available for choosing an auxiliary description. We develop simultaneous posterior and conditional-composition image bounds, together with cell-dependent reference measures for reconstruction constraints. The reference measures retain both distortion deficits in one likelihood and permit the same analytical description cost to cancel between two image comparisons. A Gaussian evaluation gives matching second-order rates for Heegard--Berger coding with one uninformed decoder in the strict both-active regime, its common-reconstruction counterpart, and the corresponding two-stage refinement corner. The framework also gives compatible helper-image intersections, reconstruction-cell bounds, and conditional-type privacy amplification. An agreed-reproduction image explains when a scalar Gaussian coefficient replaces an optimized information--distortion profile. Gaussian Wyner--Ziv and Gray--Wyner characterizations provide attributed benchmarks and complementary evaluations of the image constraints. For arbitrary helper sections, general reconstruction policies, and reliable-key generation, the finite-block bounds identify the remaining likelihood or partition constraint. The rate applications use the same observation pattern and joint performance event in the converse and construction.

翻译：暂无翻译