It has been reported that SL17d_embed indicates a significant reduction in hits-on-tracks for low absolute η (i.e. near mid-rapidity) tracks. See, for example, the N_fit plots starting on page 57 of this embedding QA. It seems to be fine at high absolute η (matching the real data very well), then gets worse as η approaches 0, by which there's a clear ~3 hit deficit for the simulated tracks. I confirm this finding and tried to look for whether hits are missing from tracks.

First, from looking at the code that generates the embedding QA plots, I do believe that the quantity N_fit only includes TPC hits.

These plots are the fraction of time a track topology map has a bit set to 1 for the tracks examined. Looking at Run 16 dAu200 data with HFT in tracking. The track topology map indices are explained here. I exclude bit 0 (which is simply 1 for primary tracks and 0 for globals). The 45 TPC pad rows are bits 8-52.

Hits on simulated primary tracks at low eta, and high eta:
 

Hits on real primary tracks at low eta, and high eta:
 

Hits on simulated global tracks at low eta:
 

The above plots were made with the default TPC distortion smearing for Run 16 of 5%. In case this smearing is larger than it should be for dAu in Run 16, I tried changing the distortion smearing to 2.5% and 0%, but it seemed to have no notable impact.

-Gene