As soon as a subsequent constrained drag begins, the vertical (output/value) component of the step-4 edit is reverted. The keys snap back to the Y values they had *before* step 4, while the X movement of the new drag is applied normally. The lattice frame itself does **not** revert; it stays where step 4 left it.
We’d expect here that the constrained drag should leave the unconstrained axis exactly as the previous, already-committed edit left it, and only move keys along the locked axis. The axis lock appears to be implemented by writing back a *cached* original value for the axis being ignored, rather than by leaving that axis untouched.
Steps to Reproduce
1. Open Sequencer with a track that has several keys and open the Curve Editor.
2. Activate the Lattice tool.
3. Select more than one key so the lattice grid appears.
4. Drag a lattice control point / edge / the cell so the keys are modified in **both** axes (e.g. drag the cell diagonally). Release the mouse — the edit commits and looks correct.
5. **Without changing the selection**, hold Shift and drag the left edge of the lattice horizontally, so the drag is constrained to the X axis.