Skip to content

SCHEMA.md

Latest commit

 

History

History
235 lines (195 loc) · 6.93 KB

SCHEMA.md

File metadata and controls

235 lines (195 loc) · 6.93 KB

Semiotic Annotation Schema

How to Annotate ARC Puzzles for the Resistance

Overview

Each annotation captures the semiotic structure of an ARC puzzle:

  • What evolved patterns it exploits
  • What conditioned symbols it uses
  • What reality model it simulates
  • What the transformation rule is

One annotation = one puzzle decoded.

JSON Schema

{
  "puzzle_id": "string",           // ARC puzzle ID (e.g., "00576224")
  "annotator": "string",           // Your GitHub username
  "version": "1.0",                // Schema version
  "timestamp": "ISO8601",          // When annotated

  "evolved_patterns": [
    {
      "pattern": "string",         // Pattern type (see catalog below)
      "confidence": 0.0-1.0,       // How certain
      "location": "string",        // Where in grid (optional)
      "notes": "string"            // Explanation (optional)
    }
  ],

  "conditioned_symbols": [
    {
      "symbol": "string",          // Symbol type (see catalog below)
      "referent": "string",        // What color/element it applies to
      "confidence": 0.0-1.0,
      "notes": "string"
    }
  ],

  "reality_model": {
    "primary": "string",           // Main reality model
    "secondary": "string",         // Secondary model (optional)
    "confidence": 0.0-1.0,
    "evidence": ["string"]         // Why you think this
  },

  "transformation_rule": {
    "description": "string",       // Natural language description
    "formal": "string",            // DSL/code representation (optional)
    "edge_cases": ["string"]       // Known tricky cases
  },

  "channel_conflicts": [           // When nature vs nurture disagree
    {
      "evolved_says": "string",
      "conditioned_says": "string",
      "resolution": "string"
    }
  ],

  "difficulty_factors": {
    "ambiguity": 0.0-1.0,          // Multiple valid interpretations
    "complexity": 0.0-1.0,         // Computational complexity
    "novelty": 0.0-1.0             // How novel the pattern is
  }
}

Evolved Pattern Catalog

Pattern ID Description Detection Cue
bilateral_symmetry Mirror symmetry along axis Left = Right or Top = Bottom
radial_symmetry Symmetry around center point Rotation-invariant
enclosure Region inside another Surrounded cells
proximity_grouping Close objects = one group Spatial clustering
edge_boundary Clear boundary between regions Color transitions
face_like Two-over-one arrangement Triangle of dots
continuity Lines/curves want to continue Aligned elements
closure Incomplete shapes feel complete Partial rectangles
small_number 1-4 objects (subitizable) Instant count
motion_implied Arrow-like, trajectory Directional shapes

Conditioned Symbol Catalog

Symbol ID Meaning Common Referent
red_danger Important, danger, stop Color 2 in ARC
blue_calm Water, sky, passive Color 1 in ARC
green_go Nature, proceed, growth Color 3 in ARC
yellow_attention Caution, energy, highlight Color 4 in ARC
black_boundary Background, void, edge Color 0 in ARC
up_positive Good, more, growth Top of grid
down_negative Bad, less, decay Bottom of grid
center_important Focus, main, target Grid center
edge_liminal Boundary, transition Grid edges
one_unique Special, protagonist Single object
two_relation Pair, dialogue, comparison Two objects
three_pattern Pattern established Three objects
bigger_more Important, dominant Larger objects

Reality Model Catalog

Model ID Simulates Transformation Type
gravity Objects falling Move down until blocked
fluid_fill Liquid spreading Fill enclosed region
collision Objects interacting Stop at boundaries
growth Expansion/contraction Cellular automata
reflection Mirrors/bouncing Flip across axis
set_union Combining sets OR of colored cells
set_intersection Common elements AND of colored cells
set_difference Removal Subtract one from another
graph_path Connectivity Path between nodes
graph_tree Hierarchy Parent-child relations
classification Sorting/grouping Organize by property
analogy Consistent mapping A:B :: C:D
attention Filtering/salience Keep important, drop rest
memory Copy/recall Duplicate or retrieve
completion Finish pattern Extend rule-following

Example Annotation

{
  "puzzle_id": "00576224",
  "annotator": "ryancardwell",
  "version": "1.0",
  "timestamp": "2024-12-31T12:00:00Z",

  "evolved_patterns": [
    {
      "pattern": "enclosure",
      "confidence": 0.95,
      "location": "center",
      "notes": "Red rectangle contains blue dot"
    },
    {
      "pattern": "edge_boundary",
      "confidence": 0.9,
      "notes": "Clear boundary between red frame and interior"
    }
  ],

  "conditioned_symbols": [
    {
      "symbol": "blue_calm",
      "referent": "interior dot",
      "confidence": 0.8,
      "notes": "Blue suggests water/fluid that will spread"
    },
    {
      "symbol": "red_danger",
      "referent": "outer frame",
      "confidence": 0.7,
      "notes": "Red frame acts as container/boundary"
    }
  ],

  "reality_model": {
    "primary": "fluid_fill",
    "confidence": 0.9,
    "evidence": [
      "Interior element spreads to fill enclosed region",
      "Respects boundary of enclosing shape",
      "Final state has uniform interior"
    ]
  },

  "transformation_rule": {
    "description": "The color inside an enclosed region floods to fill the entire interior",
    "formal": "FILL(enclosed_region, inner_color)",
    "edge_cases": [
      "Multiple enclosed regions treated independently",
      "Nested enclosures fill innermost first"
    ]
  },

  "channel_conflicts": [],

  "difficulty_factors": {
    "ambiguity": 0.2,
    "complexity": 0.3,
    "novelty": 0.4
  }
}

Annotation Guidelines

DO:

  • Be specific about locations and referents
  • Provide evidence for reality model claims
  • Note edge cases you're uncertain about
  • Include confidence scores honestly
  • Document channel conflicts when they exist

DON'T:

  • Annotate puzzles you haven't solved yourself
  • Copy annotations from others without verification
  • Leave fields empty - use null if truly N/A
  • Annotate in batch without careful thought

PRIORITY:

Focus on puzzles that:

  1. Current systems get wrong
  2. Have clear semiotic structure
  3. Demonstrate rare reality models
  4. Show interesting channel conflicts

Submission

  1. Create annotations in data/annotations/ as {puzzle_id}.json
  2. Run validation: python validate_annotation.py {puzzle_id}.json
  3. Submit PR with your annotations
  4. Get reviewed and merged

Every annotation is ammunition. Make it count.