docs(should_simulate): the point-source "not in this gap" note is repo-specific

[Jammy2211]

Docstring only — no behaviour change. Refs #470. Paired with PyAutoHands#253, which adds the guard that enforces what a docstring cannot.

The wrong sentence

should_simulate's "Known gap" section currently ends:

Note that point-source datasets are not in this gap: they write a top-level data.fits alongside their JSON and are covered normally. The original issue text grouped them with weak lensing as "JSON with no FITS"; that is true of weak lensing only.

That holds in autolens_workspace. It is false in autolens_workspace_test, whose .gitignore lists data.fits under "Generated artifacts — never check in":

$ git -C autolens_workspace_test ls-files dataset/point_source/simple
dataset/point_source/simple/point_dataset_positions_only.json
dataset/point_source/simple/point_dataset_with_fluxes_and_time_delays.json
dataset/point_source/simple/tracer.json

No data.fits ⇒ no SMALLDAT stamp ⇒ _is_capped_at_the_current_cap returns False ⇒ the capped branch deletes the directory.

That is precisely the directory #470 was about, so as written the paragraph reassures the reader about the one case that actually bit us — a committed, allowlist-protected dataset destroyed on every smoke run and replaced with degenerate output (PointSolver.solve short-circuits to a fixed position pair under the cap).

What it says now

The generalisation is replaced with the rule that actually governs: whether a dataset family is reachable by the stamp is decided per repository by that repo's .gitignore, never by the family name. "Point-source is safe" becomes "point-source is safe wherever a top-level data.fits is actually written", with the counter-example named.

It also points at PyAutoHands' check_dataset_allowlist guard (PyAutoHands#253), which fails a pre_build run when a should_simulate call site that has not released PYAUTO_SMALL_DATASETS would delete git-tracked files.

Verification

test_autoarray/util — 88 passed. -k "should_simulate or small_datasets" — 8 passed.

🤖 Generated with Claude Code

https://claude.ai/code/session_01F11sMzmaVWfU6NCz1PKVVb