Help me evaluate and, only if it matches, stage the experimental native tiled-5K Aquamarine build documented at: https://aquamarine.youens.com/ My goal is one real 5120x2880@60 Hyprland output from a dual-stream 2x1 DRM TILE display. I do not mean the small c8dd371 connector-suppression patch. Do not change my installed system yet. Start by downloading and reading the page's native5k-preflight.sh, verify its published SHA-256, and run it without sudo. Independently confirm the report: distribution and display manager, Hyprland and Aquamarine packages, Aquamarine SONAME and linked ABI, GPU and driver, and the DRM topology. Require two connected connectors on one DRM device forming one complete 2x1 TILE group, with matching panel identity and a compatible 2560x2880@60 mode on each tile. The guided hardware path is the LG UltraFine 27MD5KA on amdgpu PCI 1002:1900 (Radeon 780M class), with no other connected monitor. Stop if the topology or GPU differs, the system already exposes a true native-5K output, or a tested live SSH recovery session with working sudo is not available. Record the live Hyprland PID and the Aquamarine file in its /proc maps; do not confuse a clean ldd result with the library the current compositor actually mapped. A TTY tested under stock is useful but is not sufficient by itself because VT switching is unproven in the candidate. The guided source cut is exactly: - base f3d1804205e8158c15595cdda1b566f93349ffae - source head 01cad213c76ab1cc4adfe80ad9b7093f9e4c721b - resulting tree 08f6997bc876ea459d1fae4876cd25eac474e835 - 21-commit mbox SHA-256 e4d1493cdb65dc90f7cdb4b03120fa23ca549ddc57ff8b14c67e255652eeed5c It is the last hardware-proven native-5K startup cut. Do not combine it with the divergent 45167f7 VT/hotplug branch, the later acd258c CTM/DPMS line, the 8fec9bb auto-activation patch, or c8dd371. The original physical proof used Omarchy 4.0.0-1 and Linux 7.1.8-arch1-3. The current guarded build path requires Omarchy 4.0.1-1, Hyprland 0.56.2-1, Aquamarine 0.14.0-2, libaquamarine.so.13, UWSM 0.26.7-1, and Linux 7.1.9.arch1-2 on x86_64. The guided parser and login path also pin SDDM 0.21.0-7 and v4l-utils 1.32.0-2. The later acd258c runtime proves the inherited startup core still works on that current matrix; it is not a fresh 01cad21 candidate-login proof. Do not downgrade, force a patch, or silently port it when the matrix does not match. If the preflight matches, inspect native5k-stage.sh before running it. It must clone the exact base, verify the mbox, reproduce the published tree, build and test with LD_LIBRARY_PATH removed, and stage the result only under my user data directory. It must not use sudo, overwrite /usr/lib, replace a package, edit my normal Hyprland configuration, start another compositor on the active seat, or restart anything. Invoke both downloaded scripts with BASH_ENV and ENV removed and with `/usr/bin/bash --noprofile --norc`; a shebang cannot undo startup code Bash already sourced. The candidate desktop Exec must likewise clear BASH_ENV, ENV, loader, Omarchy, and stale Hyprland variables before invoking the wrapper with the absolute Bash path and no profile or rc files. Before activating a separate login session, show me: 1. every file that would be installed under /usr/local; 2. the staged library checksum and SONAME; 3. proof that LD_LIBRARY_PATH selects that exact staged library; 4. proof that the full Hyprland loader closure has no missing dependency and ignores inherited LD_PRELOAD and LD_AUDIT; 5. build and test results; 6. the stock session I can choose at the login screen; 7. the exact SSH recovery commands and the secondary TTY fallback; 8. the complete candidate Lua config and proof that it imports no Omarchy or personal module, shell, hook, autostart, lock/idle, CTM, DPMS, or monitor control; and 9. the effective SDDM Autologin User, Session, and Relogin values, including every non-hidden file SDDM loads from its .d directories; and 10. the account login shell and all login-profile code SDDM executes before the desktop Exec. The guided path requires Bash. If .bash_profile, .bash_login, .profile, or the noninteractive path through .bashrc has side effects, stop and explain them before activation. Require the launch wrapper to resolve libaquamarine.so.13 to the hashed real file on every login, hash the isolated config, then rerun ldd and refuse startup if that mapping changed or any dependency is missing. It must execute /usr/bin/Hyprland directly rather than activating UWSM or the restart watchdog in start-hyprland, and it must refuse to overlap an existing compositor or managed graphical session. Clear LD_PRELOAD, LD_AUDIT, OMARCHY_PATH, stale Hyprland socket/config variables, every inherited AQ_* and GPU-selection override, and use PATH=/usr/bin before any wrapper validation or launch. Use a standalone recovery-first Hyprland configuration, not an Omarchy module allowlist that can grow later. It should read only /etc/vconsole.conf, publish the 5120x2880@60 monitor only on the topology-derived tile at location 0,0, record that connector in the manifest, disable XWayland, VRR, hardware cursor planes, session-lock restoration, and input-triggered DPMS wake, and define only these three bindings: Super+Return for `/usr/bin/alacritty --config-file /dev/null -e /usr/bin/bash --noprofile --norc`, Super+Q to close a window, and Super+Shift+Escape to exit to SDDM. Never use an empty catch-all monitor selector and never hard-code an example connector name. Do not load any personal or default Hyprland input, binding, look-and-feel, shell, hook, autostart, lock, idle, Night Light, DPMS, monitor, hotplug, or lifecycle module. Wait for my approval before using sudo, installing the separate session, logging out, or restarting the display manager. Preserve the stock Omarchy session and normal configuration. Discover DRM card, connector, CRTC, and plane IDs dynamically; never copy DP-5, DP-6, card1, plane 150, or plane 203 from the example machine. Never configure SDDM to autolog directly into the candidate. If effective autologin is configured, allow the manual route only when Relogin=false, its Session explicitly names a non-candidate stock session, and the live helper shows that the first autologin has already been consumed. Then use a normal logout only—no reboot or SDDM restart. Empty Session is unsafe because SDDM can fall back to the last manually selected session on the next boot. If Relogin=true, Session is empty/unknown/candidate, or effective configuration is uncertain, stop the manual route. With my separate approval, disable autologin transactionally so I can choose the candidate by hand; put the exact root-owned backup outside every sddm.conf.d directory, show the restoration command in advance, and restore it before reboot. SDDM loads non-hidden backup-looking files from its .d directories, so never create an adjacent `.bak` there. Do not claim that the wrapper prevents login-profile side effects: SDDM runs the account login shell before the desktop Exec. The wrapper must still use `#!/usr/bin/bash`, reset PATH immediately, and clear inherited loader, Aquamarine, GPU-selection, Omarchy, and stale Hyprland variables before any validation. It then refuses overlap before starting the bare compositor. After the candidate login, verify that Hyprland maps the staged library, configerrors is empty, exactly one enabled monitor is 5120x2880 at about 60 Hz with scale 2, both TILE connectors remain active, and the DRM plane state uses one nonzero framebuffer with complementary 2560x2880 source rectangles. If any check fails, press Super+Shift+Escape and return to the stock session. Also verify that the candidate exposes only the three documented bindings and that omarchy-shell, hypridle, and hyprsunset are absent. After returning to stock, prove that the candidate AQ_* and LD_LIBRARY_PATH values are absent from the user-manager environment and that stock Hyprland maps the distro Aquamarine under /usr/lib. Do not call rollback complete until both checks pass. Be explicit about the limitations. Only native 5K at login is proven. This startup cut does not support Night Light/CTM or grouped display sleep/DPMS. Hotplug/replug, monitor reload, VT switching, suspend/resume, GPU reset, multi-GPU, docks, and multi-monitor use are not certified. Do not quietly disable features and call that a fix; if I need one of those features, advise me not to use this experimental build. The separately named verification session must have no lock, idle blanking, Night Light, or automatic output management at all. The panel remains on until I exit. A manual `/usr/bin/hyprctl` lifecycle command can still bypass policy, so do not issue DPMS, monitor keyword, reload, or CTM commands in that session. Never describe the absence of those features as working DPMS.