WMAP observations have accurately determined the position of the first two peaks and dips in the CMB temperature power spectrum .
These encode information on the ratio of the distance to the last scattering surface to the sound horizon at decoupling .
However pre-recombination processes can contaminate this distance information .
In order to assess the amplitude of these effects we use the WMAP data and evaluate the relative differences of the CMB peaks and dips multipoles .
We find that the position of the first peak is largely displaced with the respect to the expected position of the sound horizon scale at decoupling .
In contrast the relative spacings of the higher extrema are statistically consistent with those expected from perfect harmonic oscillations .
This provides evidence for a scale dependent phase shift of the CMB oscillations which is caused by gravitational driving forces affecting the propagation of sound waves before recombination .
By accounting for these effects we have performed a MCMC likelihood analysis of the location of WMAP extrema to constrain in combination with recent BAO data a constant dark energy equation of state parameter w .
For a flat universe we find a strong 2 \sigma upper limit w < -1.10 , and including the HST prior we obtain w < -1.14 , which are only marginally consistent with limits derived from the supernova SNLS sample .
On the other hand we infer larger limits for non-flat cosmologies .
From the full CMB likelihood analysis we also estimate the values of the shift parameter R and the multipole l _ { a } of the acoustic horizon at decoupling for several cosmologies to test their dependence on model assumptions .
Although the analysis of the full CMB spectra should be always preferred , using the position of the CMB peaks and dips provide a simple and consistent method for combining CMB constraints with other datasets .