1. To estimate the effect of the recently discovered bug in the status code on the pi0 cross section in 2005 p+p data, I propose the following case study.

Looking manually at one SMD status table from 2005, chosen at random, I found 860 strips marked with status = 3 (cold) or 4 (hot), consisting of 395 groups of consecutive
strips. I ony looked at one half-barrel.
Because of the bug, each of those two status codes was misplaced and assigned to the next strip (SoftId+1).
Therefore, for each such string of strips, one good strip was unnecessarily masked (both in the data and simulations, so doesn't count), and one bad strip was
exposed (in the data only).
About half of those groups consisted of only one strip and had one good strip in between, the rest were strings of multiple consecutive bad strips separated by
multiple good strips.

Assuming that one bad strip does not produce a spurious pi0 signal, I naively estimate the effect of this bug to be the apparent loss of acceptance of 395/9000
= 4.4%, in each SMD plane.
This applies to the HighTower-2 data, where the SMD clusters were allowed to contain one strip.

In the HighTower-1 data, the SMD clusters were required to contain at least two strips, therefore the loss of every other strip in some regions did not produce
an additional loss of acceptance. In those regions, every other strip was already masked anyway. So, I estimate the loss of acceptance in the HighTower-1 data
to be about half of that in the HighTower-2 data, ~2%.

In the MinBias data, the SMD information was not used at all.

In summary, the bug is estimated to cause an apparent loss of SMD acceptance of 2*4.4% in the HighTower-1 data (two SMD planes combined, for two photons) and 2*8.6% in the
HighTower-2 data.

2. From Frank Simon:

I think that your estimates are an upper limit of the real effect: You assume that the two planes of the SMD are not correlated, plus you assume that every problematic strip will actually cause a problem. The faulty channels are hot and cold channels, so they are not completely dead, they just see above or below average count rates. So they might also yield usable signals from time to time. Plus, in many cases a SMD cluster will have more than one strip, which also reduces the impact of the wrong status.

It might be possible to get a better estimate of the impact by a comparison of distributions (Pi0s vs phi, eta) in data and MC, to see if any unexplained holes show up.

On the other hand, if you add the additional systematics in quadrature, the effect is probably not that big, since we have quite large systematics from other sources as well.

3. I looked at the eta and phi distributions of the reconstructed BEMC points in the real data and in the simulations. The ratio data/simulation is flat within chi2/ndf = 5. Bringing chi2/ndf to unity would require an additional uncertainty of 5% on the acceptance. For the pions, the corresponding number should be doubled (two photons).

This is in agreement with Frank's observation that data and simulations match well on the hit level.

4. Disclaimer

I made two separate estimates, and both are suffering from uncertainties that could easily reduce the final number by a factor of several:
a) In the first estimate, I assumed SMD-eta and SMD-phi planes to be independent. I reality, there are failing modules, where the failures in two planes are completely correlated.
b) In the first estimate, I assumed that a loss of a hit leads to a loss of a cluster. In reality, many (about a half) clusters have two or more strips, so an impact of one bad strip is not so severe.
c) In the second estimate, I assumed that all observed difference between data and MC is due to this bug. This gives an upper limit, at best.

Therefore, the estimates may be reduced by some factor, which I just take to be 2.

4. Conclusion

I propose to assign a systematic uncertainty of 5% to the HighTower-1 and HighTower-2 pi0 cross section.

5. Cross check with simulation

In the following two plots, I show the pi0 acceptance and efficiency correction factor for HighTower-1 and HighTower-2 simulations, respectively. Solid symbols correspond to the status tables used in the main pi0 analysis. Open symbols correspond to the "corrected" status tables. The lower panels in both plots show the ratios and the pol0 fits.

As expected, the HighTower-1 data are less affected than HighTower-2, and in both cases the deviation from unity is within 5%.