There are two unusual features of the charge asymmetry in top pair production that are often misunderstood: the lack of acceptance corrections and the lack of reconstruction corrections. Since CDF's preliminary measurements came out, these stand out even more since CDF's analyses correct for both effects. Things are particularly confusing when both CDF and D0 results are presented in one talk, as the results are not comparable. We must start by reviewing the theory in more detail than is available in the PRL draft: Since QCD production dominates top pair production and is indifferent to electric charge, any Feynman diagram predicts charge symmetric top pair production. But interferences between diagrams produces asymmetries, and this starts at NLO (alpha_s^3). NLO includes both 2-->2 and 2-->3 processes. The NLO predictions are 3.8% [2], 4-5%[1]. References are as in the paper: [1] Rodrigo & Kuhn, [2] Bowen,Ellis,Rainwater, [3] Dittmaier, Uwer, Weinzierl The 2-->2 process is predicted to have an asymmetry of 6.4%[2]. The 2-->3 process an is predicted to have an asymmetry of -6.9% [2] or -7 to -8% [3]. NNLO predictions are available only for the 2-->3 processes in [3], they are: 0 to -3%. This differs greatly from the NLO predictions, showing that the perturbation series has not yet converged and that the NLO predictions are not a reasonable approximation of the SM. Another hint in that direction is that softer higher orders, as approximated in Pythia's showering model, may also play a role. More on the Pythia predicted asymmetry below. More generally, small asymmetries are difficult to calculate since they are the difference between two large numbers (the forward and backward cross sections) which magnifies any uncertainty on those numbers. 1) The main conclusion, which completely shaped our measurement, is that the shape of the asymmetry is unknown. Therefore we should not correct for acceptance and we should keep the acceptance simple. This is in stark contrast to the situation in other top measurements, as explained in the conference note's appendix (subsection 3, page 10): http://www-d0.fnal.gov/Run2Physics/WWW/results/prelim/TOP/T50/T50.pdf Note that the decision not to correct for acceptance is due to the state of theoretical knowledge, not due to any experimental issue. Having designed a measurement with a simple acceptance, we examined the MC@NLO particle-level (detector-level numbers are given in the paper) asymmetries within our acceptance: 2-->2: 3.4% (+/- 0.4%) (i.e. 4 jet final state) 2-->3+: -7.6% (+/- 0.7%) (i.e. 5+ jets) inclusive: 1.1% (+/- 0.3%) Just like in [2], we see that in the integrated asymmetry, the 2-->2 and 2-->3(+) contributions partially cancel out. All our numbers are smaller though, due to the acceptance which mainly biases the phase space towards higher top pTs. Therefore we ask that our results for the 4 and 5+ jet bins are also presented in talks: those are interesting asymmetries and the (accidental?) fact they are so different in our (statistically limited) data is intriguing. 2) The second issue is that of reconstruction effects. The key is the dilution plot you can see in the Arxiv. - If the asymmetry is at low |dy| values (as is the case for the light Z' resonances we examined), less than half of it will be reconstructed. - If the asymmetry is spread out (as in MC@NLO and for the heavy Z' resonances we examined), about 50-65% of it will be reconstructed - If the asymmetry is at high |dy| values (as is the case for Pythia's showering model), about 80% of it will be reconstructed. An obvious partial solution is to quote the SM corrections. - it is partial in the sense that it prevents the use of the measurement for BYSM limits. - must be derived from simulation. The best one we have is MC@NLO, but unfortunately NLO results are of dubious accuracy (see above). One can imagine several solutions to this issue, e.g., 1) Ignore events with low |dy| 2) unfolding the asymmetry as a function of |dy|. However, we found the present statistics insufficient for this. Note: these solutions cost statistical strength, it's not clear that even with full Tevatron statistics we will be willing to pay that cost. Therefore we chose not to correct the measurement for reconstruction effects. Instead we ask the model builder to use our stated dilution to make his model predictions for D0 data. For completeness, we note that the same arguments we present against correcting the measurement while integrating over the |dy| distribution can be made against our solution, which integrates over everything else. The counter arguments is that since the |dy| dependence dominates, we were able to consider all other dependencies as systematic uncertainties. Those are tiny for SM production, but for BYSM production whose kinematics are very different, we quote an additional 15% uncertainty on the dilution (i.e. a multiplicative uncertainty). BTW: It is easiest to understand the charge asymmetry without using Feynman diagrams: each incoming particle is surrounded by a cloud of virtual particles that polarize the vacuum according to its charge. It's easiest to start with an electric charge and production, but the picture is the same for strong production such as top pair. When the incoming particles interact, the virtual particles keep going in their original directions, so that the decay products match them best if their charges correspond to the original charges. Hence in qqbar-->g-->ttbar the top tends to be produced in the direction of the incoming proton. In ttj production, the color charges are different, and this simple picture provides no prediction.