I just reran Marissa's calibration.  My original output (from the old
gnuplot-fitted ~/.python/calibrate_cantilever.py) had produced

  $ cat 20081215/20081215101516_analysis_text
  k (N/m) : 0.0471212 +/- 0.0132663
  a**2 (V/nm)**2 : 0.000191658 +/- 1.36722e-06
  T (K) : 295.15 +/- 5.68434e-14
  1/Vp**2 (1/V)**2 : 60334.2 +/- 16980.8
  $ cat 20081215/20081215124856_analysis_text
  k (N/m) : 0.0346675 +/- 0.0118672
  a**2 (V/nm)**2 : 0.000217743 +/- 1.08297e-06
  T (K) : 295.15 +/- 5.68434e-14
  1/Vp**2 (1/V)**2 : 39070.7 +/- 13373.1

Using all the defaults (and only the second take for the first
calibration).  However, once I'd set up the tweakfiles and ran
20081215_calibrate.sh, I got

  $ cat 20081215/20081215101516_analysis_text
  Variable (units)              : mean +/- std. dev. (relative error)
  Cantilever k (N/m)            : 0.0656644 +/- 0.00107743 (0.016408)
  photoSensitivity**2 (V/nm)**2 : 0.000190122 +/- 1.86869e-06 (0.00982893)
  T (K)                         : 295.15 +/- 5.68434e-14 (1.92592e-16)
  1/Vphoto**2 (1/V)**2          : 84756.3 +/- 1113.56 (0.0131383)
  $ cat 20081215/20081215124856_analysis_text
  Variable (units)              : mean +/- std. dev. (relative error)
  Cantilever k (N/m)            : 0.0494809 +/- 0.000855265 (0.0172848)
  photoSensitivity**2 (V/nm)**2 : 0.000217743 +/- 1.08297e-06 (0.00497362)
  T (K)                         : 295.15 +/- 5.68434e-14 (1.92592e-16)
  1/Vphoto**2 (1/V)**2          : 55765.6 +/- 923.129 (0.0165537)

Comparing the two, we see:
  101516:
    k 0.0471212 -> 0.0656644 (with a 92% drop in error)
    photoSensitivity 0.000191658 (little change)
    T 295.15 (no change)
    1/Vphoto**2 60334.2 -> 84756.3 (with a 94% drop in error)
  124856:
    k 0.0346675 -> 0.0494809 (with a 93% drop in error)
    photoSensitivity 0.000217743 (no change)
    T 295.15 (no change)
    1/Vphoto**2 39070.7 -> 55765.6 (with a 93% drop in error)

I don't have hard evidence yet, but I am confident this discrepancy is
due to poor fitting with the older calibration file.