1. Where is the L1 trigger decision made?
A: The information from the front-end chips is put in a buffer and you have many of them saved in the buffer for latency of 2.5 microseconds and you need to make a decision in this time
1. Why trigger at all?
   1. We can’t save 40MHz. In run 1, we saved 300Mbps and in run2  we save at 300Gbps
   2. Luminosity per fill goes down and at the beginning, we save at 1.5KHz at the peak and we save 500 Hz towards the end
1. ATLAS specific:
   1. L1 accept tells you to read the event and also tells you the RoI. We are the first detector to use RoI. You read RoI with full resolution and find a primitive object in this region and then make a decision at next level trigger.
   2. We can read RoI at 100KHz in Run 2, not the full detector though
   3. You do full tracking reconstruction in trigger but not L1
   4. Full FTK idea is to do full tracking at L1. This is important to find PV, tracks, etc. Important because L1 trigger threshold is fixed, eg: 6 muon triggers
1. What would we not trigger on?
   1. Eg: Low track vertex wouldn’t be interesting
   2. Long-lived particles
   3. We don’t write out events randomly? For pile-up, we need to record full data
1. Run 1 vs. Run 2?
   1. L1 remains the same
   2. L2 change
   3. You can use topological triggers and correlate objects in different areas
   4. Phase 1 trigger: Many changes in the topological trigger especially in calorimeter part
1. Cesar’s overview -> Readout buffer is between L1 and HLT
2. Object definitions for same hits in L1, L2 and offline? We will see in performance because if we see something offline then what is the efficiency to get this in L1
3. Overlap between different streams was 10% -> very inefficient. In Run 2, we have one physics stream
4. L1 uses calorimeter and muons, so is it difficult to reconstruct the track or because tracks eventually end up in these subdetector? Turns out it is difficult to do this. CMS has double layered pixels which they may use to trigger on L1? We cannot do this in ATLAS
5. Jets vs electrons: Electrons use 2x2 tower with isolation ring around it. However, for jets it is a varied 4x4 or 6x6 or 8x8 for jets.
6. Trigger towers from L1 are fixed but you unfold the distribution in phi and retain the event if there is signal in neighboring towers. You don’t loose information -> Smart!
7. Muon picture: Why is there a dependence on primary vertex position? Because you use the line joining PV and the hit in the middle layer to give cones in the outer and inner layer to look for hits