Gravitational Manipulation Theory v2.8: Residual-Coordinate Reconstruction after Independent Recovery Failure

Gravitational Manipulation Theory (GMT) v2.8 converts the independent recovery failure reported in GMT v2.7 into a pre-registered residual-coordinate reconstruction test. GMT v2.7 evaluated a frozen six-descriptor ridge model on an independent, information-firewalled External grid. The model retained a real but sub-threshold signal (global R² = +0.258) and returned Outcome C under the pre-registered recovery criteria. GMT v2.8 asks whether the remaining residual can be organized by a closed-form apparatus coordinate absent from the frozen six-descriptor basis. Ten target-blind candidate coordinates were defined from frozen geometry and operator information only. Candidate reduction followed a sealed procedure comprising zero-scale and complete-case eligibility checks, a |r| > 0.98 collinearity screen, and a unified leave-one-class-out / leave-one-family-out cross-validation gate against a parameter-matched null. Exactly one orthogonalized coordinate, nchan (operator complexity; observed range {1,2}), survived. The resulting seven-column [6+nchan] augmented recipe and 2,000 matched-null models were frozen on the 36-cell v2.7 External grid. The frozen predictions were then evaluated once on a new 48-cell Confirmation grid containing four pre-registered configuration families, including a previously unseen helical-azimuthal family. No model refitting, candidate reselection, threshold modification, clipping, or cell exclusion was permitted after the freeze. On the one-shot Confirmation grid, global RMSE improved from 1.153 for the inherited baseline to 0.819 for the augmented recipe, while global R² increased from +0.259 to +0.625. The global RMSE improvement was +0.333, exceeding the parameter-matched null95 value of 0.291. Three of four families—azimuthal, differential, and helical-azimuthal—met the pre-registered family criterion; linear-grating worsened. The frozen mechanical A/B/C/D rule therefore returns Outcome B: Partial Structured Recovery. This is not full recovery or a PASS certification. The result is limited to one independent synthetic Confirmation grid and to the frozen [6+nchan] recipe. It does not establish universal or family-independent transfer, does not establish extrapolation beyond the observed nchan range {1,2}, and does not demonstrate experimental gravitational manipulation. All apparatus are synthetic representative classes; none is HUST hardware, and the Confirmation G₁ values are one-shot generated outputs rather than experimental measurements. The deposit contains: the complete 18-page manuscript with Figures 1–8; the 16-page Supplement S1–S13; all eight full-resolution figures; the sealed pre-registration and Stage 1–6 JSON/NPZ records; full-precision frozen model coefficients and matched-null models; the raw one-shot Confirmation G₁ record and run provenance; SHA-256 manifests and canonical self-hashes; figure and supplement generation code; and reproduce.py, which verifies the archived hashes, reconstructs the frozen predictions, recomputes the sealed metrics, and reproduces Outcome B without model retraining or scientific re-adjudication. Manuscript, supplement, figures, and data records are licensed under CC BY 4.0. Source code is licensed under the MIT License. This deposit is the successor to GMT v2.7:https://doi.org/10.5281/zenodo.22012535