Recovery and reorientation after interrupted work

This working paper examines recovery after cognitive work and the problem of choosing the next valid action after an interruption. It brings together an analysis of saturable extracellular glutamate uptake, a secondary analysis of the human workday spectroscopy dataset of Wiehler and colleagues (doi:10.5281/zenodo.6795446), and controlled tests of two frozen Qwen2.5 instruction models. At fixed residual release, applying the classical integrated Michaelis–Menten law predicts a half-recovery time that increases with the starting excess. The human tissue glutamate-to-creatine data show a descriptive difference in workday change between demand groups. They do not measure extracellular glutamate during a break. In 36 synthetic task families, preserving current dependency reports sometimes improved paired action selection, but a reader that simply followed the last visible report outperformed both learned readers on the linked inputs. On full histories the 7B checkpoint often distinguished the two substantive alternatives in its relative scores while still choosing to ask for clarification. Version 1.1 adds kinetic prior art, a baseline-adjusted spectroscopy sensitivity analysis, and a post hoc diagnostic of the archived model scores. The deposit includes the paper and materials for reproducing its calculations, figures, and model analyses. The paper describes the author’s contribution and the use of OpenAI Codex.