I'm confused about some properties of the sidereal day, in particular whether its duration varies systematically over the course of the year.1 It seems to me that that must be the case, but the details are a bit confusing to me.
I understand that length of a sidereal day is the time interval between successive transits of the line of equatorial longitude corresponding to a specified right ascension, $R$ (usually $0^\text{h}$). Each sidereal day, this line shifts against the earth's direction of rotation by some amount, $\Delta\alpha_{SID}$, to a new position, so that the sidereal day is always shorter than the earth's rotational period on its axis (the "stellar day").
What's confusing me is that the length of the sidereal day is often described as if it were constant; but it can't be, nor can the corresponding changes in the equatorial longitude of $R$, $\Delta\lambda_{SID}$, even assuming a constant rate of precession:2 If the sidereal day has a constant length, $\Delta\alpha_{SID}$ must be constant, but then $\Delta\lambda_{SID}$ will vary (between $\Delta\alpha_{SID}/\cos\varepsilon$ and $\Delta\alpha_{SID}\cdot\cos\varepsilon$) as a consequence of the varying equatorial latitude at which the the line of equatorial longitude corresponding to $R$ intersects the ecliptic. But this is impossible, since equatorial longitude should change uniformly, so that $\Delta\lambda_{SID}$ should be the same for each (same-length) sidereal day. The reverse can't be the case either: that each sidereal day, the position of the line corresponding to $R$ shifts by the same fixed amount, $\Delta\lambda_{SID}$, along the ecliptic. If that were the case, $\Delta\alpha_{SID}$ and thus the length of the sidereal day would vary (between $\Delta\lambda_{SID}\cdot\cos\varepsilon$ and $\Delta\lambda_{SID}/\cos\varepsilon$) resulting in a sidereal day of varying length; but if the length of the sidereal day varies $\Delta\lambda_{SID}$ can't be constant.
So the only conclusion I can come to is that both $\Delta\lambda_{SID}$ and the duration of the sidereal day, and thus $\Delta\alpha_{SID}$, vary, subject to the constraints imposed by spherical trigonometry: $$\Delta\lambda_{SID}\cdot\cos\varepsilon<\Delta\alpha_{SID}<\Delta\lambda_{SID}/\cos\varepsilon$$ and $$\langle\Delta\lambda_{SID}\rangle=\langle\Delta\alpha_{SID}\rangle$$ Is this correct?3 Do both of these quantities vary systematically in this way, with these properties.4 Is "the" sidereal day really a mean sidereal day?5
(1) Though a distinction between a "mean sidereal day" and an "apparent sidereal day" is sometimes mentioned, the terminology I've come across is confusing, and descriptions that do discuss variations seem to be referring to different, unspecified, phenomena. Here, I am asking about the properties of an idealized "mean sidereal day", specifically whether even this "mean" day has systematic variation in duration.
(2) My thinking went like this. Take the plane of the ecliptic as reference. Then precession is just the a rotation of the equatorial coordinate system around an axis perpendicular to that plain. The earth's rotational axis is not collinear with that axis, so that the great circle of the equator intersects the ecliptic plane at two points, one of which can be used as a reference to "track the progress" of precession. Clearly, the progress of that point along the ecliptic — i.e., the change in the intersection's ecliptic longitude, $\Delta\lambda$ — has a direct one-to-one relationship with the rate of precession. What's not clear to me is how this relates to the corresponding change in right ascension, $\Delta\alpha$. But what's being measured in that case is the distance between the reference point and some initial point on the ecliptic, within the equatorial system, an arc that passes through different declinations and thus $\Delta\alpha$ values that vary from $\Delta\lambda$ as described above. (Moreover, this seems to me the only way that $\langle\Delta\lambda\rangle$ and $\langle\Delta\lambda\alpha\rangle$ to be the same, which must be the case for the method described in (3) to work.)
(3) I'm reasonably confident that the first is true, but I'm not sure how to arrive at the second; though it must also be true, since otherwise it would not be possible to determine the precessional period (in sidereal days) from $24^\text{h}/\langle\Delta\alpha_{SID}\rangle$, as is commonly done.
(4) The corresponding (by definition, assuming constant rotational rate for the earth) constant quantities for the stellar day must thus be $$\Delta\lambda_{ST}=\langle\Delta\lambda_{SID}\rangle\cdot\frac{d_{ST}}{d_{SID}}=\Delta\alpha_{ST}=\langle\Delta\alpha_{SID}\rangle\cdot\frac{d_{ST}}{d_{SID}}$$ where ${d_{SID}}$ is the length of a mean sidereal day, and ${d_{ST}}$ the length of a stellar day.
(5) Or perhaps a mean mean sidereal day. (A very mean sidereal day?)