diff --git a/README.md b/README.md
index afc1a8d..892bd7d 100644
--- a/README.md
+++ b/README.md
@@ -36,7 +36,7 @@ Hyprland desktop setup on Fedora Linux, styled with Catppuccin Mocha. Built for
 │   │       ├── reload.sh         # Reload compositor, bar, and notifications
 │   │       └── wallpaper.sh      # Per-workspace wallpaper listener
 │   ├── kitty/
-│   │   └── kitty.conf            # GPU terminal with 144Hz input tuning
+│   │   └── kitty.conf            # GPU terminal with configured redraw delays
 │   ├── rofi/
 │   │   ├── config.rasi           # Application launcher
 │   │   ├── scripts/
@@ -109,11 +109,11 @@ https://github.com/user-attachments/assets/e41b7de6-2717-4371-a81f-c4a124757f12
 |---|---|
 | [Hyprland](https://github.com/hyprwm/Hyprland) | Wayland tiling compositor |
 | [Waybar](https://github.com/Alexays/Waybar) | Floating status bar with workspace indicators |
-| [awww](https://codeberg.org/LGFae/awww) | Wallpaper daemon running 144 FPS bezier transitions |
+| [awww](https://codeberg.org/LGFae/awww) | Wallpaper daemon with transitions configured at 144 FPS |
 | [SwayNC](https://github.com/ErikReider/SwayNotificationCenter) | Notification center and quick settings |
 | [hyprlock](https://github.com/hyprwm/hyprlock) | Screen locker with screenshot blur |
 | [hypridle](https://github.com/hyprwm/hypridle) | Idle listener and sleep controller |
-| [Kitty](https://github.com/kovidgoyal/kitty) | Terminal emulator running 6ms refresh and 1ms input latency |
+| [Kitty](https://github.com/kovidgoyal/kitty) | Terminal emulator with 6ms repaint and 1ms input delay settings |
 | [Rofi](https://github.com/davatorium/rofi) | App launcher, power menu, and picker scripts |
 | [Fastfetch](https://github.com/fastfetch-cli/fastfetch) | System information summary |
 
@@ -160,7 +160,7 @@ To link all configuration directories (`hypr`, `waybar`, `kitty`, `rofi`, `swayn
 ./scripts/install-symlinks.sh
 ```
 
-Existing conflicting configuration files or directories will be safely backed up automatically with a timestamp before creating the symlinks.
+Existing conflicting files, directories, and symlinks are preserved in unique backup directories before replacement.
 
 ## System optimizations
 
@@ -206,3 +206,10 @@ deduplication. Git identity, credentials, signing, and differing local overrides
 stay local and take precedence. Symlinked `.bashrc` or `.gitconfig` require manual
 integration. Backups use unique `.backup-*` names; link conflicts are stored under
 `BACKUP/original`. Restore backups to undo installation, then open a new shell.
+
+
+### Performance evidence
+
+See [the performance baseline](docs/performance-baseline.md) for recorded shell
+startup samples, their sandbox limitations, and the remaining host checks. Current
+visual and redraw settings are preferences; they do not establish measured latency.
diff --git a/docs/performance-baseline.md b/docs/performance-baseline.md
new file mode 100644
index 0000000..1e252a4
--- /dev/null
+++ b/docs/performance-baseline.md
@@ -0,0 +1,75 @@
+# Workstation performance baseline
+
+## TL;DR
+
+Preserve the current visuals and target balanced responsiveness. Configuration is
+not a benchmark. The measurements below were captured in a restricted agent
+sandbox, not a complete Hyprland session; no desktop performance gain is claimed.
+
+## Recorded evidence
+
+| Check | Observation | Limit |
+|---|---|---|
+| OS | Fedora 43 KDE edition | Read from host os-release |
+| Memory/storage | About 30 GiB RAM, 8 GiB zram, Btrfs on NVMe | Snapshot, not a pressure test |
+| Shell with current startup | 20 samples; median 39.32 ms; range 38.53–41.25 ms | Non-login interactive Bash, no PTY, output discarded |
+| Shell without Fastfetch | 20 samples; median 33.95 ms; range 32.84–35.54 ms | Temporary rc file; original startup untouched |
+| Fastfetch difference | 5.37 ms between medians | Sandbox observation only; retain current behavior |
+| Codec package | mesa-va-drivers-freeworld installed; mesa-va-drivers absent | Package inventory does not prove application decoding |
+| Power configuration | `/etc/tuned/active_profile`: balanced; EPP: balance_performance | Live TuneD activation could not be queried |
+| Monitor/compositor | Unverified | Hyprland socket access denied |
+| Boot/failed services | Unverified | System and user bus access denied |
+| Desktop idle CPU/power | Unverified | Process visibility excludes the live desktop |
+
+## Repeat shell startup measurement
+
+Run from your normal host terminal. This measures startup without terminal
+rendering, so also compare perceived opening time in Kitty. The temporary variant
+intercepts the direct Fastfetch startup invocation; it never edits dotfiles.
+
+```python
+import pathlib, statistics, subprocess, tempfile, time
+
+original = pathlib.Path.home().joinpath('.bashrc').read_text()
+with tempfile.TemporaryDirectory() as directory:
+    for variant in ('current', 'without_fastfetch'):
+        rc = pathlib.Path(directory) / variant
+        # Intercept only direct fastfetch calls in this disposable child shell.
+        # The tracked fragment and existing rc both use direct calls.
+        rc.write_text(('fastfetch() { :; }\n' if variant == 'without_fastfetch' else '') + original)
+        samples = []
+        for _ in range(20):
+            start = time.perf_counter()
+            subprocess.run(['bash', '--noprofile', '--rcfile', str(rc), '-ic', 'exit'],
+                           stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL, check=True)
+            samples.append((time.perf_counter() - start) * 1000)
+        assert len(samples) == 20
+        print(variant, 'median/range ms:', statistics.median(samples), min(samples), max(samples))
+```
+
+This temporary function does not intercept an explicit `command fastfetch` or
+absolute executable path; inspect customized startup files before using it.
+Retain Fastfetch unless repeated host measurements or your preference justify
+moving it to an on-demand command.
+
+## Complete the host baseline
+
+- Capture `systemd-analyze`, `systemd-analyze critical-chain`, and failed system/user
+  units across three comparable cold boots. Report medians and ranges, not a single
+  longest service: services can start in parallel.
+- Run `hyprctl monitors`, `hyprctl configerrors`, `tuned-adm active`, and
+  `tuned-adm verify`. Confirm actual resolution/refresh and power-profile activation.
+- Count `wallpaper.sh` listener processes after login and after ten `SUPER+A`
+  refreshes. There should be one listener.
+- With the same apps open and no builds/downloads, capture three sixty-second
+  idle samples using `pidstat -u -r 1 60` if installed. Record compositor, bar,
+  wallpaper, CPU activity, and available hardware power readings separately.
+- Inspect `vainfo` and the browser's media diagnostics during the same video test.
+  Confirm the codec, rendering/decoding device, CPU use, and dropped frames.
+
+## Change policy
+
+Keep blur, transparency, animations, VRR, Kitty delays, and wallpaper transition
+rate unchanged in this series. Test one proposed tuning change at a time with the
+same workload. Adopt it only if improvement is repeatable beyond baseline variation,
+and record any visual, latency, or idle-power tradeoff plus rollback.