According to the errata, the pin definitions for positive and negative differential inputs was corrected in the data sheet, however on revision 2586Q, Page 123 says ADC0 or ADC2 can be selected as positive and ADC0, ADC1, ADC2 or ADC3 can be selected as negative which gives four combinations of inputs. Page 124 further states that if ADC0 or ADC2 is selected as both the positive and negative input then the offset in the gain stage can be measured directly. So it appears the possible combinations are:

ADC0/ADC0 -> offset measurement

ADC2/ADC2 -> offset measurement

ADC0/ADC1, ADC0/ADC2, ADC0/ADC3 -> differential measurement

ADC2/ADC0, ADC2/ADC1, ADC2/ADC3 -> differential measurement

Table 17-4 shows for differential measurement:

ADC2/ADC2 (1x)

ADC2/ADC2 (20x)

ADC2/ADC3 (1x)

ADC2/ADC3 (20x)

ADC0/ADC0 (1x)

ADC0/ADC0 (20x)

ADC0/ADC1 (1x)

ADC0/ADC1 (20x)

With eight MUX combinations taken up by single polarity inputs there are not enough MUX combinations left for the other possibilities. It appears there actually only two combinations (ADC2/ADC3 and ADC0/ADC1) useable for differential pinputs. Page 122 states "the differential input (PB3, PB4 or PB2, PB5) (that is ADC3, ADC2 or ADC1, ADC0) is equipped with a programmable gain stage ...". I presume Page 123 should have stated "ADC3 or ADC1 can be selected as negative input to the differential ampifier", yes?

A secondary question, since one bit is taken up by the sign in a differential measurement, if measuring a bridge output where one side of the bridge is always positive relative to the other (absolute pressure measurement) the 10 bit accuracy could be maintained by using single conversion on both sides of the bridge and subtracting the result, yes?