Marshall JCM2000 DSL100 — Thermal Bias Drift (Conductive Trace, Bias Supply Hand-Wired)

Symptom

Bias would not hold as the amp came up to temperature. Set correctly cold, the idle current climbed as the chassis heated and kept climbing - reaching the high 100s of milliamps per tube, several times what an EL34 should be idling at.
That is not a tone complaint. At that current the output tubes are being destroyed while the amp sits there doing nothing, and the output transformer is next.

Diagnosis

This is a known defect on pre-2005 JCM2000 boards rather than a failed component. The bias supply is routed through a conductive trace on the board whose resistance changes enough with temperature to move the bias voltage with it. Nothing measures faulty cold, which is exactly why it gets misdiagnosed as tubes, a bad bias pot, or a leaky coupling capacitor.
The tell is that the drift tracks chassis temperature rather than playing time or signal level, and that it comes back with a fresh set of tubes fitted.
Schematic used: JCM2000 Superlead Standard, JCM2-60-00.DGM, Iss 1. A TSL100 print (TL10-60-02) was consulted early and its designators do not match the DSL - worth knowing before it costs you an afternoon. The rear/jack PCB sheet TL10-62-02 is common to the range and does apply.

Fix

Bias supply lines lifted off the board and hand-wired point to point, taking the temperature-dependent trace out of the bias path entirely. This is the accepted fix for the fault and it is reversible - nothing is cut that cannot be put back.
Bias now sits at 32-35mA per tube and stays there from cold through to fully warmed up, against the high 100s it was running away to before. Read across the 1R cathode sense resistors R54 and R68, where a millivolt reads directly as a milliamp.
Measured cold and hot, before and after, to confirm the drift was actually gone rather than merely smaller.
Verified as a standalone repair. It had no effect on the separate high-frequency oscillation this amp also came in with, which turned out to be the FX loop jack contacts and is written up on its own.

What I Took Away

A fault that only exists at temperature will not be found by a cold multimeter, and most bench measurement is cold measurement. Set this amp's bias cold and it reads correct; the damage happens in the twenty minutes after that. If a bias problem measures fine on the bench and misbehaves in the room, the question to ask is what in the path changes when it gets hot - not which component is out of spec.
The gap between the two figures is the whole point. Something idling in the high 100s per tube is not drifting, it is running away, and every hour the amp spends like that costs a set of tubes. An owner who keeps replacing output tubes on one of these is not unlucky.
Worth knowing the board revision before quoting one. On a pre-2005 DSL this is a design fault, not wear, so it will be there regardless of how well the amp has been looked after, and no amount of retubing will fix it.

Reference

Bias and output stage designators from JCM2-60-00.
Output stage is four EL34 in V5 to V8. Bias feed R59 and R61, 220K, via CON9. Grid resistors at the sockets are 220K: R55 for V5, R58 for V6, R62 for V7, R66 for V8. Coupling C22 and C23, 22n, into R47 and R48 680K to the grid bus. Cathode sense R54 and R68, 1R 1W. Screen resistors 1K 5W at R50, R57, R72 and pair. Screen dropper R63 270R 7W, a known hot spot - check the board around it for charring. Fuses F1 to F4, T6.3A.
No grid stoppers are fitted from the factory. The TSL version has 5K6, and adding 1K to 5K6 at pin 5 is a valid reversible mod if a local parasitic ever shows up.
Also worth doing while the chassis is open: C46, 22pF 500V, plate to screen on V8 only - Marshall service bulletin SB30. It is under-rated for that node, fails leaky or shorted, and can take out V8 and the screen resistor with it. Upgrade to 1kV or omit.