backlog: amend #1129 -- a literal before-uncompressing check ships default-on, and cites this cell in its own source

[wshallwshall]

Docs-only. #1129 stays OPEN, on a narrower question. No new number.

The central premise is false

The item's ground is that the engine enforces ceilings incrementally instead of checking before uncompressing. That is the pinned verb's exact distinction -- and it is false for the DICOM path.

Verified against origin/main:

_inflate.py   bounded_inflate_or_error(compressed, *, max_bytes) -> None
              inflates in _INFLATE_CHUNK-bounded chunks, DISCARDING the output,
              raises DicomBombError the instant cumulative size would cross the cap
              guard_part10_deflate(data, *, max_bytes=DEFAULT_MAX_INFLATED_BYTES)
              runs BEFORE any dcmread -- the default argument is what makes it default-ON
peek.py:81    "# ASVS 5.2.3: ... Pre-check the inflate in bounded memory and reject
               an over-cap object ... BEFORE dcmread ever touches it."

So a small compressed bomb with a huge inflated size never materialises in memory, and this is the literal before-uncompressing check the verb asks for.

The module names this very cell in its own source. The code knew about the item while the item did not know about the code -- the same shape already recorded on #1114.

What still stands

The zip_decompress half is unchanged: max_output_bytes=None by default, so a Handler author calling it without a ceiling gets none.

Re-scope: whether that default is an honest gap given that the transport-facing path already does the literal check, and whether "maximum number of files" is answered anywhere. Narrower and harder than "does a before-uncompressing check exist", which is answered.

The pattern, recorded in the item rather than left in chat

This is the second item from the 2026-08-08 sweep with a false central premise, after #1131. Both failed the same way: the absence was asserted from plausible shapes rather than from the state. #1131's regex could not structurally match admin_reset_password; this one looked at the archive reader and never looked for a bounded inflate elsewhere.

If a third turns up, the sweep's method is the defect and wants its own item rather than another amendment. Writing that trigger down now, so the third instance is recognised as a pattern rather than filed as a third one-off.