Tim Shoppa wrote:
I have been playing around with my homebrew VFO, a Hartley oscillator,
with a tapped inductor in the tank.

Rough parameters: 6AH6 pentode for the VFO, tank resonant at 1.8Mc,
plate circuit resonant at 3.6Mc. Rather low-Q plate tank (on purpose, I
want it to cover 100kc or so). Tank tapped about 1/3 of the way from
the bottom. 150V from an 0A2 on the screen, bypassed by a 0.005 uF
ceramic with short leads at the screen.

While playing around with it I found this weird mode where it wouldn't
necessarily start up in constant oscillation. It would repeatedly
(30000 times a second) start up (starting up very quickly, in just a
few cycles), grid and cathode circuit amplitude would build up to about
10V p-p, then the oscillations would slowly (over the next 30
microseconds) die down. Then it would repeat.

Hand capacitance near the grid or on the tank coil would often break it
out of this mode and into more regular oscillation. Putting a 10x scope
probe on the grid sometimes broke it out of this mode too.

What eventually made the circuit more reliable was putting a few
hundred ohms in series with the grid. But I don't understand exactly
how this helped.

My guess for this squegging mode is that the oscillator would suddenly
start, the tank would ring, the tank would ring hard enough that grid
current flowed, and that the grid current somehow would "latch" on
until oscillation died out, then it would repeat.
(snip)

(my guess)
The grid acts as a rectifier that builds a DC grid bias
voltage from the rectified AC signal, biasing the tube off.
Then the tank dies a natural death. Adding the series
resistor reduced the efficiency of the rectification.

Sounds like you have too much positive feedback, to begin with.