Initial ideas (suggestions welcome):

* L1:
* Architecture Comparisons: L0 (if applicable), L1, L2 (if applicable) rates and latencies. Justification of differences.
* Performance Comparisons: Predicted trigger rates for various benchmark signals with and without a tracking trigger. Specific rate reductions from tracking trigger.
* Triggering using Pixels.
* Use of timing information.
* Simulation of trigger (e.g. associative memories)
* IBM's Power initiative and in general non-x86 FPGA/CPU hybrids including OpenCL for FPGAs

* HLT:
* Compare input rate, output rates of HLT farm, size of data
* Resources and sharing with offline
* Common Tools
* Tracking and Calo clustering code developments
* Online use of multivariate classifiers : classifiers which are robust against changing detector conditions, multi-class classifiers, classifiers which are explicitly trained with resource constraints (e.g. the farm size/time per event), classifiers which adapt their response to global event properties, etc. 
* How to use HLT to align/calibrate the detector in real time:
   -- Which discriminants are more/less sensitive to the detector conditions, i.e. what can go into the preselection and what needs to wait for the calibration
   -- Resources needed for the buffer space where the events wait while the detector is being calibrated
   -- How much of this can the offline reuse


From Nikos:

A few thoughts for the Friday am session, for 5th Sept. Assuming that
we have ~3h30, I would propose to dedicate ~2h30 for presenting and
discussing the ATLAS and CMS trigger strategies at 5-7e34 (rate
projections and estimates, benchmark channels that are challenging for
the trigger and would justify the proposed thresholds etc etc,
particularly focussing on the step(s) before full detector readout)
and the role/performance/projections for the L1Track capabilities.
That would mean two 25min presentations from each of the experiments,
with discussion before/during/after the coffee break. The remaining
~1h, we could spend it discussing the "trigger-less" architectures of
LHCb and ALICE and the associated challenges, which come a lot sooner
than the ATLAS and CMS ones above!