Using VCO instead of PLL (Drift of a VCO)

I have a very system-level question: If, in my application, the synthesized frequency does not to be precise but it would be sufficient to know the actual frequency (e.g. by measuring it): Can I get away without PLL (VCO only) or is a PLL always strictly required?

For example:

1.) Does phase noise improve with PLL? (I guess no because PN of PLL should even be higher due to PD)

2.) A PLL definitely accounts for temperature drift of a VCO ... but how critical is this? Is it possible to build a VCO (LC Oscillator) in a modern CMOS process that is sufficiently temperature/"long term" stable? With long term ... hours range. That means: If my control signal is exactly 1.234V and after 2 hours I apply the same 1.234V - can an LC oscillator be built that will produce the same frequency?

If you want a fixed frequency, then you want a TCXO / OCXO.
VCOs are meant to be bossed around and have no great P, V, T
stability, this is Somebody Else's Problem (those somebodies
being the reference clock, and the PLL).

Now you may find that the available 10MHz-range XOs are
highest accuracy, and rolling them up with a PLL and VCO
gets you better overall phase noise and frequency accuracy
than a natural higher-frequency compensated oscillator.
The PLL /N ratio knocks down jitter quite a bit.

A simple LC VCO isn't going to have great temperature
stability and probably will have fairly high phase noise.
The realizable Q is not that great, using spiral aluminum
inductors and MIM caps. And the lowest practical
frequency of such a design is probably hundreds of MHz
because component values can't get real big before
they eat the whole chip.

As to your example questions, "same" as in identical to
an arbitrary number of decimal places, won't happen (by
definition). If you had a tolerance / stability quantity in
mind you could analyze to it, or look at what scholarly
papers and maybe even silicon products have achieved.
But my gut says temperature excursions alone will kill
you, even at 10% allowed deviation (and good luck
with your "make" tolerances, getting that close at t=0
and all environmentals dead-center.

Ok, thank you. To summarize: An LC-oscillator based LCO has no frequency stability in practice due to temperature drift.

Reason: As mentioned, the exact frequency does not matter in my application (I can "calibrate" the system at startup for whatever frequency the VCO will give me for a certain control input, hence PV do not matter; voltage will be regulated too). However, after startup calibration, it needs to be "long term" stable in terms of temperature and good phase noise. Now somebody wants to tell me a VCO is sufficient but I think I really need a PLL. Your response would confirm this.

Phase noise of the VCO almost ALWAYS gets better, within say +/- 50 KHz of the carrier, when you phase lock it. That is because the phase noise of the crystal oscillator clock, even when multiplied theoretically by 20 LOG N, is much better than the free running phase noise of the VCO.

If you look at a free running varactor tuned microwave vco, it is quite common to see it jumping around /- many megahertzes.

You ignore device noise.

If you apply constant voltage as cnontrol voltage of VCO, you see fairly large fluctuation of frequency due to device noise.
This has no relation to PV at all.

PLL is required.