What the precision comparison found

All six fixed endpoints completed, using 84 integration evaluations and zero archived-data optimizer starts. All eleven arithmetic/derivative controls and all six subsequent exact-motion solver controls passed.

Wider internal arithmetic improves the stringent stopping checks without changing the physical equations, calibration bounds, observations or packet-clock convention:

• The two unfinished background-field fits pass their original full probe checks with wider arithmetic. Their cost-change ratios fall from approximately 1.06 and 4.01 times the limit to 0.0095 and 0.0615 times the limit.
• One previously completed neighboring endpoint also passes the newer full probe check under wider arithmetic, after failing that check under ordinary arithmetic.
• The September disc endpoint remains outside the vertical stability limit: its full correction moves vertical predictions about 3.77 mm RMS, versus a 1 mm numerical tolerance. More arithmetic precision does not resolve that case.

Overall, five of six full probes pass under wider arithmetic, compared with two under ordinary arithmetic. Corrections recalculated from the wider derivatives also pass at the same five endpoints. This identifies arithmetic precision as a contributor to the cost-only stopping failures; it is not a new instrument correction.

These are fixed-point diagnostics, so the published findings remain 30 globe preferences, zero flat and 23 unresolved. A failed primary cannot be promoted without its complete comparison and repeated-start qualification. The paired trial tests that next.

Full precision report and individual probes and exact-motion solver controls retain the evidence and its scope.