I think two days is too short, and something like 3-3.5 days is better. If there are problems with mulitple feeds (for example due to buggy scripts or failed data sources), which has already happened, it can take hours or longer for them to be fixed (witnesses need to be contacted, scripts may need to be edited, etc.). Likewise for any sort of malicious manipulation (which would likely involve collusion between witnesses, though not necessarily, and not that many). While it may big fixed well before two days, during the time until it is fixed, the feed is still being filled with bad data. To ensure this bad data is not a significant portion of the overall dataset requires that the window be significantly longer than the failure.

The 10%, etc. discount can be reduced and will likely reduce anyway as a consequence of lower inflation, an APR that is more realistic given the risks of storing capital on a risky experimental blockchain without potential for upside (of course we hope this factor can also reduce over time), and not allowing the debt load to grow too large before addressing it. In fact I was already considering a feed discount reduction now that we have a higher APR, once the debt load decreases (which seems to be happening this week). I do think even long term a small discount is probably needed under 'average' market conditions, regardless of feed window (though maybe not in the low-infation model). There is still a time cost to receiving payment 7 days forward vs. trading directly and receiving immediate payment. Under stressed market conditions that cost can be very significant (even for 2-3 days).

I do think that 7 days is a bit of overkill and three days would work pretty well. That means hypothetically 12 hours of bad data would still be only 1/6 of the data set and 24 hours (a more realistic time period to vote out misbehaving witnesses) would be 1/3. That said, 7 days gives a lot of robustness to the system. Despite individual failing feeds, a systemic failure due to the Bitstamp hack, some witnesses failing to update for days on occasion, there has never been a case when the fairness or robustness of the 7 day feed has even been called into question. That says something.

In my opinion, the biggest problem with SBD is not the feed delay and accompanying price volatility (in a well-functioning market even with current high inflation that would still only imply a few percent of discount or less), it is excessive debt load on a still-shaky backing asset with low liquidity making SBD risky and hard to trade (including conversions). Please consider the proposal I have made here. Again, of course hopefully this can improve and indeed this proposal will likely improve liquidity directly and therefore the peg indirectly.