To be honest, this is a complex issue, but there are a few approaches taken in real life trading.
The most simplistic approach is to mark the off-the-runs against the liquidly traded points on the curve. We almost always have accurate real-time pricing on benchmark issues and CTDs for bond futures. The off-the-run issues then trade at some spreads to these liquid issues. How are the spreads determined? The hand-waving answer is just supply/demand that shifts the spreads around. You as the market maker will look at the aggregated market information and adjust the spreads based on market conditions, your inventory profile, your objective (e.g., are you purely providing liquidity or are you also exploiting relative value opportunities in conjunction with other positions in the book), etc.
A second common approach is to build an on-the-run spline using liquid points and price all the other issues relative to this spline (basically you are determining the most appropriate z-spread relative to this spline). Here's a press release from RiskVal that touts their implementation of this approach; it's rather vague, but might give you a sense for what practitioners actually do (I have no affiliations to them).
Yet others use dynamic term structure models. These models are far more likely to be used when relative value opportunities are an important consideration. Similar to the on-the-run spline approach, you'd fit the model using a pretty small number of liquid points on the curve and then price all the other issues based on appropriate z-spreads relative to this model. An example of such a model can be found in the 3rd edition of Bruce Tuckman's Fixed Income Securities. The challenge is again to determine the appropriate z-spreads based on market conditions, etc.
Some also reference off-the-run splines, OIS spreads, etc., but these are really just variations of the same theme. At the end of the day, it really just boil down to supply/demand and what makes sense to you as a liquidity provider. If you look at other posts, it's abundantly clear that large chunks of Treasuries can deviate from fair value persistently, but there's just not much you can do.