The ranking is shaped by the sample, and the sample is unusual

Measured by frankB under feature-b-a-fourth-corpus-to-test-whether-the-ladder-walls-generalise (resolved b125395e2). reportlab 4.2.5, 421 .py / 0 .c, run as a separate probe at pin v389 — md5 0453ed506a14e464fd6c6cf0d81c6a55, base 83468c5462d4, via a new --probe=<dir> mode that reuses the ladder's exact method while keeping its own baseline and roots. The N/48 series is untouched.

36/159 compile. 123 failures over 30 distinct first walls, and not one of them is a wall the family produced — nothing from the callable / dispatch / kwargs / decorator / protocol cluster this campaign ranks 55-70.

But the reason is not that reportlab hits different mechanisms.

It stops at its first missing import and never reaches the mechanism layer.

The fork

The three corpora that generated the ranking — webencodings, html5lib, tinycss2 — are self-contained web parsers with almost no stdlib footprint. That is what such a library is. They arrived at the mechanism layer immediately and have been producing the ranking ever since.

So the mechanism walls are conditional: they are what a corpus hits once its import surface is already covered. On a corpus with an ordinary stdlib footprint, landing the whole 55-70 cluster moves compile count by ~zero — those items sit behind sixteen missing modules.

This is the trap six sprang at file scale (it gated 15 files; landing mimic_six moved 4/48 → 4/48), now reproduced at corpus scale. Compile count lags; walls-cleared leads.

The campaign has no measurement of library surface at all. The ticket that would produce one — feature-nilpy-stdlib-coverage-gaps-measured [p72] — is the top-ranked NilPy feature in the backlog and has never been started, while items below it get worked.

Options

1. Measure first, then re-rank (recommended) — start feature-nilpy-stdlib-coverage-gaps-measured before any further mechanism work. It is already top-ranked; the only change is actually taking it. The re-ranking question then answers itself with data instead of judgement.
2. Re-rank the mechanism cluster down now on this evidence alone. Cheap, but it trades one unmeasured ranking for another — and the mechanism walls are genuinely real, just gated.
3. Leave the ranking as-is, on the view that the low-dependency corpora are the right sample because mechanism gaps are compiler work and stdlib shims are library work with a different owner. Defensible; should be stated explicitly if chosen, because it is currently true by accident, not by choice.

No re-ranking has been done. prio: is the human's field and frankB correctly did not touch it. Track N is also currently undispatched by owner call, so nothing is blocked on this today — it is filed so the evidence outlives the session that measured it.

Boundary case already split out as a bug

The #1 wall, undefined variable (os) at 30 files, is nominally library surface but is really a seam: os.getcwd(), os.path.join(), os.getenv(), os.environ.get() all compile (dotted calls special-cased in pyparser.inc), while 'HOME' in os.environ and os.sep fail — there is no os module value behind the special-case (pyparser.inc:11852 accepts only seek_set/cur/end). One seven-line leaf file, reportlab/lib/__init__.py, ending in RL_DEBUG = 'RL_DEBUG' in os.environ, gates all 30.

Filed separately as bug-n-os-environ-and-os-sep-are-not-values [N, p60]. Textbook normalise-dont-special-case.