Start a led on a sertent voltage

Typically, LEDs are actuated by a current rather than a potential, but, lets presume that you want to derive a 3-volt signal to drive a ballasted LED.  The problem you mention, otherwise, is rather canonical.  You can use an op-amp in the inverting configuration followed by a comparator.  Your input resistor should be about 100x larger in value than the feedback resistor, this configuration allows for fractional (and negative) gain.  Don't forget the current-balancing resistor, from gnd to the non-inverting terminal whose value is Rin//Rf for best results.  You can protect the inverting node with a Zener or something, SVP.  You might want to opto-isolate this output.  You could directly deploy a current comparator, outfitted with input resistors.  You could go old-school and feed the input led of an opto-isolator thru a ballast resistor and a string of Zener diodes.  If you need precision from this topology, you need to include forward Si diodes to balance the tempcos of the Zeners.

However you do it, this is HV land, be careful.

You can also use a resistive attenuator to bring your signal potential into the realm of your circuit.  They have names like 'L' and 'Pi,' you can look them up on the internet.  If you want simple, you can use a resistive string and a comparator.  Let's say I was possessed of a 2.5V voltage reference.  I would want to hook it up to one input of a voltage comparator.  Probably the inverting node, I can switch sense with another comparator, duals are as cheap as singles.  Rather than directly connect the reference, I probably want to use a resistor to balance the input stage and to allow for positive feedback if I figure that I want add Schmidt action.  (330 + 2.5)/2.5 = 133.  So you could use a 3/4 K resistor pulling down and a 100K resistor as your input. Note that Power = V * V / R, so at the moment of truth your input resistor is dissipating 11W!  To see if we could use a smaller (in terms of power rather than value) resistor, we would have to look at your duty-cycles and probably perform some definite integrations.  Because the ratio of these resistors is so extreme, we can just use another 3/4 K resistor as our current-balancer, as a large resistance in parallel with a small resistance is just a little less than the small resistance.  For a small error penalty, you could rescale your resistors by an order of decimal magnitude, and use a 1W input resistor, I suppose.

An interesting side-note about Ohm's law:  Material scientists remind us than less than half of all materials are Ohmic!

Probably the gold-standard circuit here, and just as probably beyond your requirements, is to PWM or FM a signal according to the HV level, send it through an opto-coupler or other isolation boundary, and detect the HV on the other side, this allows both precision and isolation while operating our coupler in saturation mode so that we don't have to be concerned with the non-linearities of the isolator, as we are only operating it at two points along the curve.

Integrated Soln:  TL431 http://www.ti.com/lit/ds/symlink/tl431.pdf example 22 crowbar circuit, replace TRIAC w/ P-FET or optoisolator.

You can also use a resistive attenuator to bring your signal potential into the realm of your circuit.  They have names like 'L' and 'Pi,' you can look them up on the internet.  If you want simple, you can use a resistive string and a comparator.  Let's say I was possessed of a 2.5V voltage reference.  I would want to hook it up to one input of a voltage comparator.  Probably the inverting node, I can switch sense with another comparator, duals are as cheap as singles.  Rather than directly connect the reference, I probably want to use a resistor to balance the input stage and to allow for positive feedback if I figure that I want add Schmidt action.  (330 + 2.5)/2.5 = 133.  So you could use a 3/4 K resistor pulling down and a 100K resistor as your input. Note that Power = V * V / R, so at the moment of truth your input resistor is dissipating 11W!  To see if we could use a smaller (in terms of power rather than value) resistor, we would have to look at your duty-cycles and probably perform some definite integrations.  Because the ratio of these resistors is so extreme, we can just use another 3/4 K resistor as our current-balancer, as a large resistance in parallel with a small resistance is just a little less than the small resistance.  For a small error penalty, you could rescale your resistors by an order of decimal magnitude, and use a 1W input resistor, I suppose.

An interesting side-note about Ohm's law:  Material scientists remind us than less than half of all materials are Ohmic!

Probably the gold-standard circuit here, and just as probably beyond your requirements, is to PWM or FM a signal according to the HV level, send it through an opto-coupler or other isolation boundary, and detect the HV on the other side, this allows both precision and isolation while operating our coupler in saturation mode so that we don't have to be concerned with the non-linearities of the isolator, as we are only operating it at two points along the curve.

Integrated Soln:  TL431 http://www.ti.com/lit/ds/symlink/tl431.pdf example 22 crowbar circuit, replace TRIAC w/ P-FET or optoisolator.