Retrospective, since this is merged and v0.1.0 is going out on top of it.

Two things I checked in the code rather than inferred, both of which belong in the release notes rather than in a change to this PR.

The gimbal parameter rename is a breaking change for competitors. gimbal_range became max_gimbal_angle with no alias, and that key is in the whitelist agents receive: it is present in both *_given_parameters.yaml on v0.0.2. An agent that reads it raises KeyError on v0.1.0. The shipped example agents do not read it, so the examples still run. I have put it at the top of the draft release notes on #61 rather than proposing an alias this close to the deadline, but the alias is three lines if you would rather have one.

The actuator dynamics path was not exercised when this landed. Both scenarios set all three time constants to null, so the golden masters only covered pass-through. That gap is real and #66 and #71 have since closed it; #71's mutation run showed the earlier tests stayed green with the sampling rate halved and with the roll actuator's filter bypassed entirely.

Separately, while following the TVC integration I found that the flight equations reduce axial thrust for a gimbal deflection and apply the resulting moments, but never add the lateral components of that same thrust vector to the translational equations. At the shipped 15 degree limit on both axes that omits 0.366 * T, and two 15 degree axes reach a combined deflection of 21.47 degrees. Raised upstream as ARRC-Rocket/ActiveRocketPy#21.

I would not hold v0.1.0 for it. It is upstream, it predates this release, every entrant runs the same model so nothing is unfair, and changing the force model five days out would invalidate both golden masters and any submission already made. What it does mean is that the release should not describe the thrust vector model as validated, which the draft notes say plainly.

One more contract question worth its own issue rather than a fix here: the launch branch creates the Flight but does not read tvc, roll or throttle from that same action. Nothing is simulated on the launch step either, so nothing is lost in practice, but accepting a full action and using three of its fields is worth stating one way or the other.