docs: record how the v2.32.6 archive was actually produced

HISTORY.md

@@ -6,6 +6,9 @@
 - The second finding came from rendering, not reading. The contributor scaled the artwork faithfully by 512/108 everywhere except `strokeWidth`, which landed on 8 rather than 11.852, so the foreground drew the markdown veins at 1.27dp while `ic_launcher_monochrome.xml` drew the same mark at 1.875dp. Nothing in the diff looks wrong; it shows up when you build both revisions and put them next to each other on a launcher. Group scale also went 0.75 → 0.85, which keeps 28.9dp against the 33dp safe radius and stops the icon reading as undersized.
 - Verification: built `main` and the PR head separately and installed them side by side on a Pixel emulator (API 36), which is what surfaced the vein weight. Measured both revisions against the 108dp canvas — `main` reached 34.0dp from centre against a 33dp safe radius, which is the clipping. `launch-smoke` passing after the revert is the direct evidence for the #319 diagnosis.
 - Store notes: 128.txt in seven locales, all comfortably under the 500-character Play cap (433 at the longest). Landing ×7 (three version strings plus the release date) and README ×7.
+- **The `D:\Build` archive came from CI, not from a local export.** `:app:exportReleaseToBuildDrive` crashed the Gradle daemon twice in `:app:minifyReleaseWithR8` — `Native memory allocation (malloc) failed`, at `-Xmx2048m` and again at `-Xmx1536m`, with about 1.7 GB free on the machine. Lowering the heap was the right instinct (the starvation is native, not heap) and it still was not enough; R8 needs headroom this box did not have that night. The tag job had already produced both artifacts, so they were downloaded from run 31891151897 and placed under the same stem: `markleaf-release-aab` → `markleaf-v2.32.6-vc128.aab`, `markleaf-release-mapping` → `.mapping.txt`, with the notes file assembled from the seven `128.txt` in the order the previous releases use. `verify-release-export.ps1` passes on the result, and the AAB's certificate is the release identity (`CN=Markleaf, O=Markleaf, L=Seoul, C=KR`) because CI signs with the same keystore. This is a legitimate substitute rather than a shortcut — the CI artifact is the same task's output on the same commit and key — but it is worth writing down that the local path can fail for a reason that has nothing to do with the release.
+- Also worth writing down: `-Pmarkleaf.requireReleaseSigning=true` must be quoted when invoked from PowerShell. Unquoted, PowerShell splits it and Gradle reports `Task '.requireReleaseSigning=true' not found`, which reads like a missing task rather than an argument that never arrived.
+- `launch-smoke` failed on the release PR and is not a regression: the CI emulator died mid-install (`Broken pipe (32)`, then `Can't find service: package`), which is the flake the job's retry loop exists for. The same job passed on #318 with identical app code; #321 changed version strings only.
 - No hardening issue filed. The one candidate this pass produced is the launch-smoke package-name coupling, and that is already #319; #262 stays the standing tracker.
 
 ## 2026-08-14 - The release docs catch up with the release that just shipped (D068)