Make the two build scripts importable, and read the real prefix lists

[realmarcin]

Closes #97 and #101.

The problem

build_subsets.py resolved every Mech checkout at import — MECHS calls mech_root, which exits when one is missing — then scanned the corpora. build_data.py read both JSON inputs at import and rewrote fleet_data.json. Neither could be imported to look at a constant.

The refactor

Both keep only pure constants at module level. build_subsets grew a prepare() that resolves checkouts and indexes HabitatMech's pages; everything from the scan onward moved into main(). Function bodies are untouched and still refer to the same module-level names, so MECHS, hab_pages and friends are declared empty and filled by prepare() rather than moved into it. build_data grew build() and main().

Importing either is now free and writes nothing.

The payoff: #101 retired, not narrowed

The tests had to recover VOC and PREF from the source text with ast, which cannot see a list rebuilt after its literal. That was the hole #99 closed partly and #101 recorded the rest of. Reading the real objects retires the class:

mutation after the literal	source parser	real import
VOC, _X = [...], 1	passed	FAILED
del VOC[0]	passed	FAILED
VOC[0] += "X"	passed	FAILED
VOC += [...]	FAILED	FAILED
VOC.append(...)	FAILED	FAILED

literal() and its mutator allowlist are deleted along with the parsing they served.

Verified behaviour-preserving

Ran both scripts before and after the refactor over the same checkouts and compared every output file. assets/fleet and subsets_summary.json match semantically: identical per-Mech scan counts, identical edge weights and by breakdowns, identical term lists in identical order, every term entry equal.

One edge file differs by bytes only, and it is not this refactor:

run A  "by":{...,"CAS":1,"ENVO":1}
run B  "by":{...,"ENVO":1,"CAS":1}

Same length, same content. CAS and ENVO are tied, and Counter.most_common() breaks ties by insertion order, which comes from iterating a set of strings under per-process hash randomization. Filed as #107 — it means a refresh churns committed assets for no reason, and the pipeline cannot be used to verify itself.

No committed data changes here; the verification runs' output was restored.

27 tests and assemble_page.py --check pass.

Review fixes

Rebased over #105, carrying its CITATION filter into main(). From the adversarial review: main() now calls prepare() itself, so importing the module and calling main() resolves the checkouts instead of failing with a KeyError inside scan() (closes #117); the unused import ast is gone (closes #116). Left filed: #118 (unclosed handles and non-atomic writes).

🤖 Generated with Claude Code