[roandegraaf]

What and why

Symptom. On 0400486024DEC000, entering an area for the first time still builds pipelines at their first draw: task / mesh pipelines have no stand-in while they build, so each one stalls the frame, and render pipelines the list does not name are built in play. Testers report hitches at area entry that line up with pipeline compiles.

Change. Data and a test, no host code.

• pso_manifest.tsv: a sweep over 18 areas of a save on an M4 Pro, started with no M rows, built 62 task / mesh pipelines at their draws in the first area and 0-14 in each later one. Of the 246 it recorded, 84 are not among the list's 164; they go with the other M rows. With all 248 prepared before the first frame, four of the areas checked built none at a draw. The sweep's 1,806 S / R / C rows that no other row names go at the end.
• A #@after_first_frame line after every M row and before the 4,959 first-looks rows. Main's host reads it as a comment (warm_parse skips # lines), so on main every row is still built before the first frame; a host that implements the mark can build the rows after it in the background.
• dm/tools/tests/test_pso_manifest.py (new): every shipped list's rows parse and hold at most one mark; this profile's list has no row twice, every key it names is in its shader map, and every M row comes before the mark.
• The measurement in docs/reference/shaders.md (Native mesh pipelines compiled at their first draw) and the game note.

Overlap. 70 of the 84 new M rows are also in #2324, and 2 in #2280. Whichever lands second should drop its duplicates; test_no_row_twice catches them.

Checked

• Tests: python3 -m pytest dm/tools/tests/test_pso_manifest.py dm/tools/tests/test_docs.py (11 passed) on this branch.
• Games run: 0400486024DEC000 v131072, the 18-area sweep on an M4 Pro, on an older host that implements the mark.

Not checked / open.

• No run on this branch's host. On main the 1,806 tail rows and the first-looks rows are built before the first frame, so a cold first start's prepare step takes longer; not measured.
• No iOS run, where this game has no bundled archive and the list is all a first start has.

Docs

• The finding is in the game note and its docs/reference/shaders.md entry (no new entry, so the symptom index is unchanged)
• A chain tool changed: no
• No game files, keys or data extracted from a game in the change or the attachments