Browser-based graphical shell and control plane for a Linux server you own — real terminal, file manager, live metrics and BYOK AI in one mobile-first workspace. Next.js 16 · React 19 · node-pty · Tailwind v4 · shadcn/ui. Public alpha, self-hosted.
Phase 4/5 comparison: before
Baseline captured on 062fe18, host idle (load average 0.15–0.42), 2026-10-29.
Every number here is measured, not estimated. "After" columns stay empty until the work lands.
Regenerate with the commands named under each table.
1. Bundle (deterministic, from build output)
node scripts/check-bundle.mjs --json
| Part | Raw bytes | Gzip bytes | Gzip KiB | Files |
|---|---|---|---|---|
| initial | 870,887 | 277,248 | 270.8 | 18 |
| entry | 426,778 | 144,578 | 141.2 | 13 |
| runtime (framework) | 444,109 | 132,670 | 129.6 | 5 |
| css | 188,715 | 30,721 | 30.0 | 2 |
| fonts | 69,652 | 69,695 | 68.1 | 1 |
Heavy eager modules (ReactFlow, Konva, Xterm, ONNX) leaking into the first-load graph: 0.
Next version measured: 16.4.2.
Framework share of the initial bundle: 132,670 / 277,248 = 47.9 %.
2. Perceived performance (Slow 4G + 4× CPU, browser)
node scripts/perf-probe.mjs --json, 3 runs. Median shown; range in brackets.
| Metric | Desktop 1440×900 | Mobile 390×844 |
|---|---|---|
| FCP | 1,164 ms (1,160–1,168) | 1,169 ms (1,164–1,172) |
| LCP | 3,152 ms (3,152–3,232) | 2,956 ms (2,952–3,028) |
| TBT | 173 ms (161–186) | 101 ms (90–114) |
| first-load JS | 910.2 KiB | 894.1 KiB |
3. Hydration mark chain
node scripts/perf-hydration.mjs --base=http://127.0.0.1:4005
| Mark | Desktop | Mobile |
|---|---|---|
| FCP | 1,180 ms | 1,160 ms |
| last JS byte | 2,804 ms | 2,746 ms |
| shell mounted | 2,523 ms | 2,529 ms |
| icons painted | 3,078 ms | 2,963 ms |
| LCP | 3,132 ms | 2,996 ms |
| LCP element | SPAN (menu-bar label) |
IMG (app artwork) |
| mount after FCP | +1,343 ms | +1,369 ms |
| LCP after FCP | +1,952 ms | +1,836 ms |
| LCP after last JS | +328 ms | +250 ms |
The chain is the reason LCP is what it is: the shell mounts only after ~910 KiB of JS has arrived and executed, and LCP lands after that. LCP is hydration-bound, not paint-bound.
4. Per-app open cost (4× CPU, cold context per app)
node scripts/perf-app-open.mjs --base=http://127.0.0.1:4005
| App | Open | Chunks pulled | Long task |
|---|---|---|---|
| Image Editor | 618 ms | 8 | 732 ms |
| Video Editor | 566 ms | 4 | 371 ms |
| Terminal | 219 ms | 3 | 63 ms |
| System Monitor | 192 ms | 2 | 68 ms |
5. Code and test surface
| Measure | Value |
|---|---|
| TS/TSX files | 2,688 |
| TS/TSX LOC | 204,905 |
.svelte files |
0 |
lib/ LOC (framework-free since Phase 2) |
100,787 |
frontend/slices LOC |
74,209 |
— of which appshell |
20,028 |
| Route handlers | 117 |
| — of which > 15 LOC | 0 |
Static next/* imports in lib/ |
0 |
| Runtime dependencies | 33 |
| Dev dependencies | 18 |
node_modules |
1,435 MB |
— of which next |
210 MB |
| Test files | 873 |
| Test wall time | 34.82 s, 877 files / 5,730 tests, 380 MB RSS |
6. What this baseline says about the remaining work
From docs/MSO-MIGRATION.md §5, with the appshell inventory measured in §4 Phase 5:
framework floor now 132,670 B gzip (129.6 KiB, 47.9 % of initial)
framework floor if swapped 33,070 B gzip (§2.1, measured in a real build)
first-load JS now 910.2 KiB (browser-measured)
first-load if swapped ~150-165 KiB (projected, app code 115-130 KiB)
appshell to re-express 20,028 LOC => 166-239 days
slices to re-express 54,181 LOC => 452-677 days
That is the "after" target this comparison will be measured against. It is 618–916 days of work, so this document tracks increments, not a single leap.