Finish the truncation change: the log, the final frame, the README and the changelog

[thc1006]

Finishes the split between terminated and truncated. That split was right; several places still read terminated from when that one flag also covered the clock, and none of them moved with it.

I swept for all of them rather than fixing the two that were reported.

The environment

if _timeout:
    logger.info("Terminated: Reached max time")

Running out of horizon is truncation now, and this line is the one thing an operator reads to find out why a run stopped.

if _remainder == 0 or terminated:
    self._render_frame()

Frames are drawn every 0.1 s of simulation time, or when the episode ends. While the clock counted as termination, an episode running out of horizon always drew its final frame. Afterwards the gate only fires on terminated, so a truncated episode draws one only if its last step happens to land on the cadence. Scenario 0 truncates at step 9999 against a cadence of 10, so it does not. Scenario 1 ends this way normally.

The README, which is where a competitor writes their loop from

It described env.step() as returning "the new observations, reward, termination flag, and additional info", which is four values and one flag, and said the episode ends when the time limit is reached or the rocket hits the ground, as though those were one outcome. It now names both flags, says which cause each is, and shows the loop.

The changelog

Had no entry for any of this. Added under Changed as breaking for every agent loop, alongside the observation copies from #99, and the four behaviour fixes from this wave under Fixed.

What I checked, and what was already fine

place	state
evaluate.py	correct since #94
doc/examples/run_env_agent.py	correct since #94
doc/examples/test_navigation_agent.py	correct since #102
evaluate_scenario_colab.ipynb	drives no loop of its own, so nothing to change
"Terminated: Rocket flight finished"	correct, that one really is termination

Tests

The two environment fixes are covered in tests/test_episode_lifecycle.py, driving a no-launch episode to the horizon. The frame test also asserts that the final step misses the render cadence, since otherwise the old gate would have drawn it anyway and the assertion would have held against either version.

The README snippet is not a .py file, so the invariant test added in #102 could not see it. It now reads the fenced python blocks out of the README and puts any loop it finds through the same guard and unpack checks as the runners, plus a check that it found one at all, since a snippet that stopped being tagged python would otherwise leave that holding over nothing.

The existing console-logging test pinned the old wording, so it moves too.

Verified by mutation, all five caught:

mutation	caught by
the old log wording	the lifecycle log test
the old render gate	the final-frame test
README shows while not terminated:	the README loop test
README discards truncated	the README loop test
the snippet stops being tagged python	the "found a loop at all" test

Local CI green on a clean tree: ruff check, ruff format, uv lock --check, 315 passed with BPC_RUN_SLOW_TESTS=1.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.