@@ -29,6 +29,16 @@ A review of `45b0d53` raised ten findings, all confirmed against the code and fi
 
 Verification after the review (fixes on `077930b`; no VM running and load below 8 before every build and gate; Node 22.23.2, because the default `node` here is 22.11, older than the 22.12 the scripts need, and `test-native-cli.mjs` then stops at `node:sqlite`; window and Node gates with a private `HOME` and `PATH` reduced to that Node and the system directories): `cargo fmt --check` and `clippy --workspace --all-targets -D warnings` clean; `cargo test --workspace --locked` (the `native-source` run) 1475 passed / 0 failed / 14 ignored, with 16 `window_size` tests (run with the real home: a private home on a long path makes `sandbox_isolation`'s socket path too long for a Unix socket); `native-behavior` (`test-native-cli.mjs` 21 scenario groups, `test-native-stress.mjs` 100/100, `test-native-tui.mjs`, the MCP server over stdio and HTTP); `glib-backport` 9/9 and 6/6; `release-tests` 176 + 2 and `release-auth` 59 with no secrets found; `built-project-cleanup` 1/1 (15,504 entries); `native-window` under `run-native-x11.mjs`: `test-native-desktop.mjs` 31 checks (first window 1296x920 on the 1440x1100 Xvfb screen at the saved place moved inside, 132 px from the top; the same after the relaunch) and `test-native-markdown.mjs`; `native-endurance` 100/100 with 32 checks; and `run-native-wayland.mjs node scripts/test-native-desktop.mjs` 31 checks on a private Weston (1380x873 content on 1600x1100, the same after the relaunch). Probes of three launches each from a planted 1536x1170 state (resized to 1000x700 before the second quit): a private Weston at 1600x1100 gives 1380x873 content every time, one at 1366x768 gives 1229x644 (a 1229x691 window), and Xvfb gives 1296x920 at the same place each time. Not run here: `managed-runtime`, `packages`, `clean-host-packages` and `packaged-behavior` (they need the Vulkan llama.cpp build and Docker; left to CI). The one remaining physical-desktop check is KDE Plasma (Wayland) in the Shadowfetch Linux 5.0 VM with the packaged 1.0.1.
 
+### 1.0.1 on KDE Plasma
+
+Checked on September 30, 2026 with a local, unsigned 1.0.1 `.deb` built from `8ffbd69` (the `.deb` leg of `build-native.mjs`; `lintian` reports no errors), installed with `apt` over the preinstalled 1.0.0 in the published Shadowfetch Linux 5.0.0 live ISO (KDE Plasma on Wayland, QEMU with 4 vCPUs and 6 GiB, throwaway disks). Each launch is `systemd-run --user /usr/bin/shadowcode`, 20 s, a screenshot, the window-state file, then `systemctl --user stop`. Window sizes come from KWin's frame geometry and, independently, from the screenshots' pixels; both agree. Evidence is in the Shadowfetch QA tree (`work/qa-5.0.0/evidence/shadowcode-1.0.1-proof`).
+
+- **Control, 1.0.0 on a 1920x1080 screen:** 1380x920, then 1432x1019, then 1484x1118, off the screen and under the panel.
+- **1.0.1, five launches from a fresh profile:** 1380x920 at the same place every time, 57 px above the panel; the file saves no size (0x0) on each exit.
+- **A grown state:** the 1536x1170 that 1.0.0 left in the control, and a planted 1700x1150 (two launches), each open at 1380x920.
+- **A 1366x768 screen** (a second VM with a 1366x768 EDID): 1229x691, 16 px above the panel, on two fresh launches, after opening a project (the composer and the status bar are above the panel), and from a planted 1700x1150 state.
+- **A saved maximized window** opens maximized, filling the work area (1920x1034).
+
 ## 1.0 program
 
 Checkpoint (September 29, 2026). The owner asked for 1.0.0: research the pain points people have with coding agents and fix them, "easy for anyone to use but advanced for real work". A survey of 32 pain points (8 already solved, 21 partly, 3 missing) set the scope. Work ran in two waves on `flagship/reliability-foundation`, each change with tests: