Tracking play is set by gauge − back-to-back − flange thickness. If a wheelset’s back-to-back is too narrow, a wheel can pick the points of a turnout; too wide and it can drop into the flangeways. We expected a narrow sweet spot: small changes in back-to-back should swing turnout reliability sharply.
One 40′ boxcar, one #6 turnout (diverging route), pushed through at a steady slow speed by the same loco. We set the car’s single test wheelset to three back-to-back values with a gauge and calipers: 0.040″ under the NMRA nominal, nominal, and 0.030″ over. 50 passes per setting (25 each direction). A “derailment” is any pick, drop, or climb at the points or frog. Reported as derailments per 50 passes.
| Back-to-back setting | Derailments / 50 | Failure mode seen |
|---|---|---|
| 0.040″ under nominal | 31 / 50 | picked the points |
| NMRA nominal | 0 / 50 | none |
| 0.030″ over nominal | 12 / 50 | dropped at frog flangeway |
The result is stark: at nominal back-to-back the car ran the turnout 50 times with zero failures; a few hundredths narrow and it picked the points on 62% of passes; wide, it dropped into the frog flangeway on 24%. The two failure modes are exactly the ones the geometry predicts — narrow lets the flange find the wrong side of the point rail, wide lets a wheel fall into the gap. This is a threshold effect, not a gentle gradient: the nominal window is small and the penalty for leaving it is immediate.
Back-to-back gauge is decisive for turnout reliability — arguably the single highest-leverage thing to check on a car that derails at turnouts. A cheap back-to-back gauge pays for itself. Caveats: one car, one turnout, one frog; absolute numbers vary with your specific points and frog, and “nominal” here is the NMRA target checked with a gauge, not a lab metrology bench. The lesson — set back-to-back first — generalizes; the exact failure rates are specific to this rig.