Comprehensive analysis of server/src/ covering:
- Boot sequence and entry points
- 8 design patterns (Singleton, Manager, Registry, Observer, Controller/Strategy, Composition, Serializable, Fixed Timestep)
- Event system with typed events and 3 emission scopes
- Dual transport networking (WebSocket + WebTransport)
- Entity/World system with physics integration
- Player lifecycle and management
- Module dependency graph
- Error handling (3-tier severity)
- Performance optimizations (IterationMap, packet caching, update thresholds)

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Comprehensive analysis of client/src/ covering:
- Multi-scene Three.js rendering with 6-pass post-processing pipeline
- Chunk/voxel system (16x16x16 chunks, 2x2x2 batch grouping)
- Web Worker architecture for geometry generation
- Entity system (~1700 line Entity.ts with 7-pass per-frame update)
- 22+ Manager pattern instances with Game as service locator
- Dual transport networking with msgpack serialization
- glTF instanced mesh rendering pipeline
- Working variables pattern for GC pressure reduction
- Manual matrix management throughout
- Mobile support with virtual joystick

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Comprehensive analysis of protocol/ covering:
- 26 packets (2 bidirectional, 6 inbound, 18 outbound)
- Tuple wire format [id, data, tick?] for minimal msgpack overhead
- 44 schema files with extreme property name minification
- Co-located TypeScript types + JSON Schema with JSONSchemaType<T>
- Custom binary framing protocol (4-byte big-endian, 32MB cap)
- Load-time Ajv schema compilation via shared singleton
- Schema composition via object spread (no \$ref)
- Delta encoding via all-optional properties
- Fail-fast duplicate packet ID detection

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Exhaustive style guide extracted from 25+ files covering:
- Naming: PascalCase classes with role suffixes, camelCase methods with verb prefixes, _underscore private fields/methods, UPPER_SNAKE_CASE constants
- Import/export: default exports for classes, named for enums/types, import type usage
- File organization: consistent 9-section class structure
- One-liner getters, named setX() methods (no property setters)
- TSDoc: comprehensive on server (with Category tags), minimal on client
- Error messages: ClassName.methodName(): Description format
- Arrow functions for event handlers
- 7 documented inconsistencies between server and client conventions

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

36 findings across 40+ files analyzed:
- 4 Critical: NetworkSynchronizer God Class (1561 lines, 60+ identical handlers), Client Entity (2900 lines), 70+ identical setter+emit patterns
- 18 Major: RigidBody/Collider ~1750 lines each, GLTFManager 1812 lines, tickWithPlayerInput 225 lines, monkey-patching applyImpulse, event listener leaks, fatalError for validation, 60+ client TODOs
- 14 Minor: as any casts, console.log leftovers, hardcoded URIs, empty catches
- 14 concrete standards for new code to prevent these smells

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Overall score: 6.6/10 across 13 principles:
- Strongest: LSP (8), KISS (8), YAGNI (8), Composition over Inheritance (8)
- Weakest: DIP (4/10) - heavy singleton usage, no DI, no abstractions
- Coupling (5/10) - fully connected singleton dependency graph
- Law of Demeter (5/10) - deep chains through managers
- SRP (6/10) - Entity ~1200 lines mixing physics/visuals/animations
- Top recommendations: introduce interfaces + constructor injection, decompose Entity, break up long methods

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Master guidelines document for reviewing AI-generated pull requests,
compiled from server architecture, client architecture, protocol layer,
style conventions, code smells, and SOLID principles analyses.

Includes: hard rules, naming conventions, code organization, type safety,
error handling, architecture patterns, performance patterns, formatting
rules, and a comprehensive PR review checklist.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Three identical patterns is enough to warrant a mapping table
or generic helper. Five was too lenient.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

New sections from mega-sappy guidelines + ArkDev code analysis:
- PR requirements (testing, AI review, backwards compat assessment)
- Multi-layer review process (static checks, AI, human, game regression)
- Backwards compatibility rules (defaults are sacred, opt-in vs automatic)
- Data-driven design principles (no magic numbers, config over hardcode)
- Quality gate stack (typecheck, lint, unit, perf, AI, human, manual, regression)
- Extended PR review checklist with new categories

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Strip out damage/weapon/catalog examples that are specific to
mega-sappy game logic, keep rules SDK-generic.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

CODING_STANDARDS.md now focuses purely on code quality (naming,
types, architecture, performance). Process/governance content
extracted into new CONTRIBUTING.md (PR requirements, backwards
compatibility, review layers, testing expectations).

Hard rules consolidated from 16 to 13. Section numbering cleaned up.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

…chable

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Section 6 covers: extract method over comments, explanatory
variables, parameter objects, guard clauses over nesting,
and when comments are appropriate (why, not what).

Checklist updated with Code Clarity category.
Sections renumbered 6-13 to accommodate.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Covers server/desktop/mobile/high-player-count considerations,
what to ask yourself before submitting runtime changes, and
when before/after measurements are required in the PR.

Checklist and PR requirements updated to include performance.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Contributors should list which games they tested against
(SDK examples, their own game, etc.) and on what targets.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

PR should specify which games, which devices (desktop/mobile/OS),
and what actions were taken in-game to exercise the change.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Layer 5 is now general AI review — bugs, logic errors, edge cases,
merge readiness. Use multiple tools (Claude Code, Codex) for
independent perspectives.

Layer 6 is standards compliance against CODING_STANDARDS.md.

Both require fresh context. Review stack renumbered to 9 layers.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>