• WCAG is a W3C Recommendation with ISO designation (ISO/IEC 40500:2025)
• Legal enforcement is real and growing (8,800+ ADA lawsuits in 2024)
• WCAG criteria are testable and falsifiable

• "Codified" suggests binding specification; WCAG is voluntary guidance with legal reference status
• "Adoption" is misleading: 96.3% of top 1M websites fail WCAG (WebAIM 2024/2025)
• Only 3.7% pass automated testing; true compliance is lower
• WCAG 2.2 achieved only 12% adoption in first year

• Native HTML/browser defaults: 100% adoption, invariants enforced by spec
• Section 508 / EN 301 549: legally binding with stronger enforcement
• Component libraries (Radix, React Aria): more precise behavioral contracts

WCAG provides "strong" coverage on only 1 of goldberry's 10 clusters (a11y_contract). 6 of 10 clusters are weak or absent:

• temporality: weak (time limits only)
• event_graph: none
• lifecycle: none (framework-era specific)
• layout_contract: weak (source order only)
• boundaries: weak (reflow only)
• security_contract: tangential (authentication only)

• APG provides design patterns and example code for accessible widgets
• Published by W3C WAI
• Used as guidance by major component libraries

• "Reference implementations" is wrong terminology
• APG is explicitly INFORMATIVE, not normative
• Examples are "illustrative, not prescriptive" (W3C's own language)
• You can achieve WCAG compliance without following any APG pattern

1. Keyboard interactions are partially implicit (ARIA mandates focus behavior but not which keys to use)
2. Touch interaction is "severely underspecified" (APG slider pattern admits difficulty)
3. Support tables cover only 4 basic patterns with 2 browsers and 3 screen readers
4. Complex patterns have no assistive technology support data

• React Aria: "implements semantics according to APG" but ADDS real-world deviations
  • Touch accessibility beyond APG scope
  • Cross-browser/AT normalization beyond APG scope
  • Hidden dismiss buttons not in APG
• Radix, Headless UI: claim APG adherence but diverge for practical reasons
• None implement APG exactly

• Khourshid does present the "unreachable states" framing
• XState implements modern interpretation of Harel's 1987 formalism
• The FaceTime bug is a real example of implicit state machine problems
• Statecharts help identify states that should not be reachable

• "Unreachable states" is NOT the primary framing in Khourshid's work
• Primary emphasis is: implicit state machines -> scattered logic -> invisible state graph
• The real value proposition is making state explicit and visible, not specifically unreachable states
• The 24ways article's login form example doesn't demonstrate unreachable states

• State space explosion: finite state machines become unwieldy quickly
• Performance tradeoffs in XState-based architecture
• CSS is orthogonal (declarative rules, not state transitions)
• Race conditions still possible at system level

• Bruce Hauman created Devcards (June 2014, not ~2015)
• "Corpus-as-development-environment" is accurate characterization
• The CLJS stack does enforce invariants at multiple layers:
  • ClojureScript: immutable by default
  • Closure Compiler: dead code elimination
  • Figwheel: hot reload preserving state
  • re-frame: declared side effects (coeffects)

• "~2015" is off by a year (June 2014)
• "Corpus-as-development-environment" is goldberry's framing, not Devcards' native terminology
• Immutability "forbids temporality.race_resolution" is too strong:
  • Prevents data races
  • Does NOT prevent async ordering bugs, temporal sequencing errors, deadlocks
• The "can't retrofit" limitation is a design choice, not fundamental architecture

Correct but incomplete:

• Clojure is a "deliberate paradigm choice," not gateway language
• 200-400 active Clojure job postings vs 100,000+ JS/TS
• The problem is Clojure ecosystem size, not just parens

• Z, VDM, Petri nets, CSP, UAN, CTT, Interactors all attempted UI formalization
• Z required 200+ lines to specify button behavior (ergonomics barrier real)
• Notation required specialist training

Ergonomics was 10-20% of the failure. The full failure breakdown:

• Airbus integrated formal verification for avionics safety-critical logic
• Z specifications used for Rolls Royce RB211-524G fuel control
• BUT: They don't use formal methods for UI specification
• Safety-critical logic is verified; UI behavior is tested empirically

1. Invariants as primary abstraction
2. Temporal reasoning for async systems
3. Composition and modularity
4. Separation of concerns (Seeheim model)
5. Falsifiability as requirement

1. Code generation as goal -> goldberry: build corpus FROM code
2. Completeness before iteration -> goldberry: phase gates
3. Notation as progress -> goldberry: minimal YAML + prose
4. Separation of spec from code -> goldberry: corpus built from code
5. Hiring specialists -> goldberry: readable by senior engineers