a peak at the edge of the grid is not a peak

An optimisation over a bounded grid returns the best point it visited. If that point sits at the boundary of the search, the result is not an optimum; it is a statement that the search ran out of room, and the two are indistinguishable unless somebody checks where the answer landed.

The first pass over one axis returned no peak at all — the quantity was still rising when the sweep stopped, which is the same fact stated more honestly. The axis was extended, then extended again. Only then did most cells produce interior peaks, and even in the extended lattice one cell still peaked outside the time axis and another at the edge of the outcome axis. Reports now carry an explicit edge flag beside any peak, so a reader can tell a summit from a fence.

The later amendment is the part worth keeping. A constraint discovered after the fact shortened one axis, because holds cannot run past a mandatory close. Rather than re-run the lattice, the axis was re-cut as a declared subset of what had already been executed, and every cell's outcome was relabelled against the reachable range. Cells that had read as peaks became at-ceiling.

What it proves: the team applies the same suspicion to its own search boundaries as to its inputs, and re-cuts a result already paid for only when the new cut is a subset of the old one.