Interesting couple of polls. According to YouGov the No vote has plummeted 16 odd points in a fortnight, while Panelbase showed a modest boost for Yes post after the second debate and then showed a NC in the relative gap in this poll. I think it comes down to the relative composition of both panels. Originally, Panelbase was always the kindest pollster to yes, and Peter Kellner did like to point out that the PB panel was 'nat heavy' - Given the fact that most 2011 SNP voters were already Yes, and that the Yes campaign has been all about getting converts from the other groups, it means that PB polls tend to show good base results for Yes, but tend to require huge shifts in other groups to translate into a shift in the headline figures, i.e. they are possibly under sensitive to shifts to Yes. You can observe that in the last big shift to Yes post currency-gate when the likes of ICM, TNS and Survation all recorded dramatic shifts to Yes, but PB's shift was modest at best. Since then, they've modified their methodology to spit weighting between 2011 and 2014 Euros, which has the effect of slightly downgrading both SNP AND Labour, making it quite hard, I think, for them to see relative shifts in the polls.
YouGov, on the other hand maybe overly sensitive to shifts from the Labour group. Kellner's panels were always a bit Labour heavy, and the dreaded Kellner correction was designed to root out shy Nos by splitting the 2011 Nat vote into 2010/2011 Nat/Nats and 2010/2011 Lab/Nats, the latter group is then upweighted quite dramamtically. The early effect of this was to surpress the Yes vote due to that group of 'passing nats' being mostly nos at the time (it also had the effect of mutliplying a large statistical uncertainty into those figures). Now that Yes is making inroads into the lapsed Labour voters, that group is movign to Yes and ironically the Kellner correction is having the effect of amplifying the effect.
So we have two polls, from one pollster who's methodology makes it hard to see any shift to Yes and the other who's massively sensitive to it. The two show a yes vote between 48-51 percent. Both are statistical ties. I tend to think the truth is somewhere between the two, and that's before factoring my concern that BPC pollsters all based on prior vote methodologies will struggle to pick up all those lapsed or first time voters.