Plan 9's critical invariant: each process has its own namespace view.

Process A sees: /dev/draw -> rio window 1
Process B sees: /dev/draw -> rio window 2
Global /dev/draw does not exist

This eliminates an entire defect class:

• No global DOM pollution
• No "which window has focus" race conditions
• No iframe/shadow-DOM boundary confusion

Everything is a file: drawing, input, network, UI. One protocol enforces:

• Composability invariant: pipe any output to any input
• Auditability invariant: all operations are file operations (logged)
• Security invariant: authentication at protocol boundary, not per-app

Acme proves accessibility can be intrinsic:

• Commands are always text-selectable
• Mouse chords are combinations, not sequences (eliminates timing bugs)
• Undo is compositionally independent (each action reversible without coupling)

New invariant candidates:

• Namespace isolation: Each interactive context must have its own namespace
• Protocol purity: Single protocol for all resource access
• Focus-before-event: Focus state and event delivery must be atomic

ICCCM (Inter-Client Communication Conventions Manual) is one of the few systematic UI invariant specifications in computing history.

        ┌─────────────┐
        │  Withdrawn  │ ← Client controls entry/exit
        └──────┬──────┘
               │
    ┌──────────┴──────────┐
    │                     │
    ▼                     ▼
┌────────┐          ┌─────────┐
│ Normal │ ◄──────► │  Iconic │
└────────┘          └─────────┘
    WM controls transitions

Critical invariant from ICCCM 2.0:

WM must NOT directly SetInputFocus if WM_TAKEFOCUS advertised. Must send ClientMessage and wait for client to request focus.

Modern frameworks violate this constantly (focus stealing).

DPS enforced invariants that CSS cannot:

NeXTSTEP's event model was NOT like DOM:

NSApplication
  └─ NSWindow (first responder context)
      └─ NSView tree
          └─ [First Responder] -> Super Chain -> App Delegate

Key difference: "Super" meant responder chain ancestor, not DOM parent. You could skip ancestors in the chain.

THE critical invariant NeXTSTEP enforced:

First responder state is immutable while any event is in dispatch.

Web violates this:

button.addEventListener('click', () => {
  input.focus();  // Changes focus mid-dispatch!
});
// Later listeners see different focus context

IB enforced at compile time:

• Outlet/action typing (IBOutlet creates compile-time reference)
• Connection validation (signal must have receiver)
• Lifecycle contracts (outlets initialized before awakeFromNib)

Modern web has no equivalent - event listeners can target nonexistent elements.

New invariant candidates:

• Focus immutability during dispatch: Focus must not change mid-event
• Rendering determinism: CSS lacks portable IR; frameworks compensate
• Design-time connection validation: All handler connections must resolve at mount

Xerox Alto (1973) and Star (1981) established what we now take for granted:

"Don't mode me in" - avoid modal state that hides available options.

Modern web has modal dialogs everywhere. Is "modeless" still enforceable?

Star had features that didn't survive:

• Property sheets: Bounded invariant spaces (vs modern unbounded dialogs)
• Universal undo: Every action reversible (modern: app-specific, inconsistent)
• Document-centric: Files are first-class, not windows (modern: app-centric)

These are the ORIGINAL invariants. Many goldberry clusters trace back to PARC:

• focus_state: First responder concept
• device_contract: Mouse/keyboard integration
• temporality: Universal undo as temporal invariant
• totality: Direct manipulation requires complete state visibility

OSF/Motif Style Guide (1993) was the first attempt to formally specify a complete GUI standard. It codified:

• Keyboard navigation model (fallback bindings required)
• Widget behavior specifications
• Dialog and window conventions
• Focus semantics (foundational to WCAG 2.1)

CDE (1993-1996) added:

• Session management (reproducible sessions)
• Drag-and-drop protocol (type-safe transfers)
• ToolTalk messaging (inter-app communication)

OPEN LOOK: "Menus are ephemeral by default; user can override with pushpin" Motif: "Menus follow keyboard & mouse conventions from other systems"

Neither was wrong. They encoded incompatible invariant claims. Sun abandoned OPEN LOOK because HP+IBM had more deployment reach.

1. Invariants are contestable: Two implementations can both claim correctness
2. Ergonomics beats formalism: English + examples won over Z notation
3. Licensing matters: CDE failed at licensing, not at specification
4. Standards ossify: Living documents (WCAG updates) survive; frozen specs die

Oberon forbade overlapping windows by construction:

• Tiled layout only
• No z-index, no stacking context, no depth sorting bugs
• Focus model simpler (no modal stacking pathology)

The invariant: "No two windows may overlap" eliminates an entire defect class.

PostScript-based window system (like DPS). Key difference from X11:

• Drawing commands sandboxed before transmission (security invariant)
• Richer protocol (not "dumb" like X11)
• Lost to X11 because X11 was simpler, not because invariants were wrong

Document embedding precursor to OLE/COM/iframes:

• Compound documents with embedded interactive objects
• "Which component owns focus?" becomes hard question
• Modern web faces same problem: iframe/shadow-DOM boundaries

"Overlapping windows" is a CHOICE, not an invariant. Alternative models:

• Tiled (Oberon): simpler focus, guaranteed layout
• Document-embedded (Andrew): focus ownership complexity
• Network-transparent (NeWS, X11): security at protocol boundary

• Pike et al., "Plan 9 from Bell Labs" (1995)
• Pike, "8½, the Plan 9 Window System"
• Pike, "Acme: A User Interface for Programmers" (1994)
• 9P Protocol Documentation

• Inter-Client Communication Conventions Manual (ICCCM)
• Extended Window Manager Hints (EWMH)
• X Window System Protocol Specification

• NeXTSTEP AppKit Reference Manual (Apple, 1989)
• Display PostScript documentation
• Interface Builder User Manual

• "Designing the Star" (IEEE Computer, 1982)
• Tesler, "The Graphical User Interface" (1981)
• Kay & Goldberg, "A Personal Computer for Children of All Ages" (1972)

• OSF/Motif Style Guide Revision 1.2 (1993)
• CDE Infrastructure documentation
• IEEE P1295 (Motif standard)

• Wirth, "The Oberon System" (1988)
• Sun NeWS documentation
• Andrew Project papers (CMU)