Mixed Model Covariance Structure

I have a study with a somewhat unique experimental design which I would like to used Mixed Modeling to analyze.

All subjects in the study were tested at baseline. Three repeated measurements (dependent variable) were collected on them under controlled conditions (ie, Pre-Exposure, During Exposure, After Exposure). Subjects were then randomized to one of two treatment group (A or B). Subjects were then tested under two different conditions (control and hazardous) in a randomized order, and under each condition, they had the same three measurements taken (Pre-, During, Post-Exposure).

Given the limitations of using AIC, BIC, etc. to accurately determine covariance structure, I thought it would be best to logically decide upon the structure.

Treatment may influence the error within a given subject, and there may be between-subjects differences in response to a treatment (ie, I may respond strongly to treatment A, whereas you may respond less strongly to treatment A). Error within a condition may also vary between subjects, given that some people may respond to the hazardous condition differently than others. Because individuals respond differently within a condition, and this may differ by treatment, I would expect Pre-Exposure and During/Post-Exposure measurements not to be correlated to one another. This is reflected when I do descriptive statistics and have different variances across each of the 9 time points (baseline plus two conditions x three measurements) for each treatment group. All of this leads me to believe that unstructured would be the correct covariance structure to use, given that the measurements are not longitudinal and that it would be unlikely to relate any one measurement within an individual to any other given measurement within an individual.

Because everybody was assigned to a treatment group, treatment would be a fixed effect. Because everybody underwent all conditions of testing and had three measurements taken within each condition (at the same time intervals), these would also be fixed structures. I would therefore have a three way interaction of fixed effects: treatment x condition x measurement.

Any feedback whether I am correct in my statistical methodology / reasoning?