We distinguish two types of fluctuations between observations of a time series, Seasonal (S) and Non Seasonal (NS). Averaged over a stretch of time the S add up to zero, so the change is all NS and it is the same for all periods. The average NS are on a straight line (the g-line) through the overall average with gradient (Yt – Yt-L)/L, where L is the number of observations in the adjustment period.
Figure A: Average components of a quarterly series.

In the initial phase seasonal components (SC) are calculated over a number of full years, by subtracting the NS averages from raw period averages (figure A). After the initial phase average period values are no longer calculated, the SC are updated for every added observation by the changing gradient of the g-line.
If an observation (Yt+1) is added to the time series this has impact on the average change (the NS). To adopt the new observation, the gradient of the g-line is corrected (the g-line turning around its center, the Overall Average). Then all seasonal components (SC) also change, their sum remaining zero.
If the new observation is aberrant we want to reduce its impact on the SC:

1. The difference between the new observation minus the appropriate SC and the extrapolation of the g-line is called e (for extrapolation error). %e is e as a percentage of the Overall Average.
2. Parameter Limit to Error (LE) sets a limit to %e to be not aberrant, the default value of LE is 6%.
3. If %e =< LE then the default “Common Adjustment” length (parameter CA) of 6 quarters is applied to change the gradient of the g-line by (e/L=) e/6 and reset the SC.
4. If %e > LE then, for this one observation, the adjustment period (L) is lengthened to a multiple (parameter M) of %e to reduce the impact of the outlier on the SC.
5. If a shift in the seasonal pattern is detected (parameter T) the adjustment period is shortened to 1 year, to adopt the new pattern at once

Note that the overall average has no part in the procedure after the initial phase but, to complete the picture, visualize that, after the g-line gradient and the SC have been reset, the overall average moves forward by one place along the reset g-line.
Also note that the Camplet procedure is all about change (e and %e) between successive observations and not about values compared to average values. The difference is the updated NS (see figure A).

Publication:
A full description of the procedure can be found in the article
“CAMPLET, seasonal adjustment without revisions”,
that has been published in Springer’s Journal of Business Cycle Research,
https://link.springer.com/article/10.1007/s41549-018-0031-3