[jeany55]

Between step 15 and step 16 of the way off eudplib: the Open list's bugs, as the plan's new section "The Open list's bug pass (2026-10-23)" has them.

The emitter's fold. (v & 0xffff) − 1025 came out as v's low half under −1025's high half. The lone-place fold puts a scratch cell straight into the one dword that takes it, and when that dword starts from a constant it turned every set 0 act of the scratch cell into a set of the constant — the &'s own set 0 under ~mask among them. The fold is not made when a masked write works the scratch cell. Eight shapes of it in tests/emitter.test.ts.

Taken: return at the top of a program (front end, emitter's END, the Python side's top coroutine end); memory()/setMemory() of the deaths table's dwords 12–26 named as the deaths of player e mod 12 of unit e div 12 in the front end, so no build reads a player group; disabled() in a program refused, preserve() and a dropped displayText(…, false) said as remarks; build warnings (editor notice + log line) for hyper triggers left out and triggers in no list; the false errors (n++ as an index — and the index before the value —, a comparison as a boolean argument, a switch on a known text, this in an arrow of a method, first(…)?.order(…), u?.hp ?? 0, sleep(…).then(() => 7) as a value, new P(4).x, && of conditions in a const, rows[i].pop() ?? -1, a build-time throw inside a try); ~, **/Math.pow by a known power, Math.sign, Math.log2, fill(v, start, end); first()/randomUnit() of a set of units; the cutscene header example compiles as written (a test reads it).

Root cause of first(…)?.order(…): libraryName took any symbol declared in the declarations file for the library's, a unit's order() method included, so the hoister treated the call as the library's order() action and never registered its object. A member of one of the library's interfaces is no longer a library name (members of its objects — on.died, local.selected — still are).

Decisions recorded (the user, 2026-10-23): no map protection on the own build; Magenta's rows go on our backend after step 17 (until then a map with Magenta's member builds through eudplib).

Ledger: 13 new rows, all takes — steps 0 (+2), 1 (+3), 3 (+1), 4 (+2), 5 (+1), 6 (+1), 9 (+1), 10 (+1), 11 (+2). No row flips.

Tests: npx vitest run 74 files / 1382 pass (+ tests/ledger.test.ts with the new rows), tsc clean. Not played in Remastered: nothing here changes a played probe's layout except the masked-fold case, which the walker covers.

🤖 Generated with Claude Code

https://claude.ai/code/session_01En4SHqafb1QFJ4xgbUkAe5