Design: based around a blocking oscillator, a complete switching power supply in only one transistor. The 1M resistor provides initial bias, charging the 0.01 base bypass capacitor. When it reaches about 0.7V, the MJE18008 kicks on. Due to positive feedback, it stays on, until collector current reaches about 1.2A. At that point, the capacitor has discharged enough that base current is falling. As a result, collector voltage begins to rise, and by positive feedback, the base is slammed off rapidly. Collector voltage rises to a peak around 500V, clamped by the RCD snubber and slowed by the RC. The flyback pulse is dumped into the output filter capacitor and the base bias supply (0.1 and 1N914, generating about -5V). The transistor turns back on either when the base capacitor reaches threshold, or if it's nearly there already, then when the inductor voltage swings slightly below zero (due to ringing). In this way, the transistor acts as a rudimentary BCM (Boundary Control Mode) switching supply.
To avoid dropping a large current from 160V, a bootstrap bias supply furnishes base current. To control it, an optoisolator shunts the negative supply, thus forcing more bias current into the base circuit, increasing the charge rate of the 0.01 capacitor.
On the output side, a TL431 is used for voltage reference and error amp. Unfortunately, the phase is wrong, so it has to be inverted with the 2N3906. A ring-of-two current sink is used to bias the IR LED instead of a resistor, to maximize startup current. As shown, the error amp starts around 2V, and enters the linear range at 10.3V.
Here's the collector voltage waveform at 2.5A load. Essentially continuous operation. Power input is about 40W, so efficiency is low, around 63%. Most of this is dissipated in the transistor, but the small transformer is getting warm, as is the diode and filter capacitor. The snubbers are also burning a watt or so.
Current waveform for same conditions, measured across a 0.47 ohm emitter resistor.