CosmoCoffee

I have a model that varies the CMB in the thermal history only and when I plot the CMB residuals (i.e., [math]\Delta C_\ell/C_\ell) the high-[math]\ell contributions to the CMB spectra blow well out of proportion. I show the residuals here and you can see the response just shoots up where actually its quite a simple variation (take my word for it!)
Any ideas on why this would be happening in CAMB and how we can change it?

I don't know what spectra you are plotting, or whether this is stable to changes in accuracy parameters?

So this is the TT power spectra (i.e., [math]\Delta C_\ell^{\rm TT}/C_\ell^{\rm TT,fid}) and using an [math]\ell_{\rm max} = 8000. The particular model variations are difficult to explain without a) discussing a paper in preparation too much or b) taking up lots of time – main point is that it is a variation that exclusively affects the ionisation history. I tried changing but that didn't fix anything either.

lensed or unlensed? The silk tail is presumably quite sensitive to the timing of reionziation since falling so steeply.

Thanks Antony. This is specifically lensed, however the impact between the two is fairly similar. The model variations used in this case were also centred around the recombination epoch with no knock on to reionisation. I'm just trying to ascertain whether the CMB should be so erratic at higher [math]\ell (obviously since this is a residual, there is every chance that this is just blown up by how small the CMB spectra is).