ARIA Core Fingerprint Atlas

Version: 1.0 Last Updated: 2025-12-06

---

Overview

The Fingerprint Atlas provides a standardized system for characterizing and comparing ARIA core behaviors. A "fingerprint" is a compressed numeric summary of how a given core behaves under a specific reference scenario — enabling regression detection, cross-core comparison, and behavioral analysis.

What is a Fingerprint?

A fingerprint captures the statistical properties of a core's output trajectories over a reference run:

• Common metrics: Statistics for coherence, stability, intensity, alignment
• Core-specific metrics: Layer-specific measurements (symbols, relations, codes, etc.)
• Behavioral patterns: Dwell times, transitions, entropy distributions

All fingerprint values remain bounded in [0, 1] where applicable, maintaining safety invariants.

Purpose

1. Regression Detection: Compare fingerprints across code versions to detect behavioral changes
2. Cross-Core Comparison: Understand behavioral differences between core types
3. Documentation: Establish canonical behavioral baselines for each core
4. Validation: Verify that parameter changes produce expected behavioral shifts

---

Reference Scenarios

Reference scenarios define reproducible conditions for generating comparable fingerprints.

baseline_quiet

Description: Default configuration, small dt, short run for quick verification.

Use case: Quick sanity checks, CI/CD pipelines, basic functionality verification.

baseline_long

Description: Default configuration, standard dt, long run for stability analysis.

Use case: Long-term stability verification, attractor basin analysis, steady-state characterization.

mild_perturbation

Description: φ-shaped dt pattern for controlled input modulation.

Use case: Testing response to smooth perturbations, measuring adaptation behavior, phase-response analysis.

high_variability

Description: Variable dt schedule within safe bounds for stress testing.

Use case: Robustness testing, edge case detection, stability under varying conditions.

---

Metrics by Core Type

Common Metrics (All Cores)

CFM v2 Specific

CFM v2 uses primarily the common metrics. No additional core-specific fields.

ARIA v0 Specific

ARIA v1 Specific

ARIA v2 Specific

ARIA v3 Specific

ARIA v4 Specific

---

Example Fingerprints

ARIA v4 - baseline_quiet

{
  "core_type": "aria_v4",
  "scenario": "baseline_quiet",
  "num_steps": 200,
  "timestamp": "2025-12-06T12:00:00",
  "common_metrics": {
    "coherence": {
      "mean": 0.582,
      "std": 0.089,
      "min": 0.412,
      "max": 0.738,
      "range": 0.326
    },
    "stability": {
      "mean": 0.724,
      "std": 0.062,
      "min": 0.591,
      "max": 0.843,
      "range": 0.252
    },
    "intensity": {
      "mean": 0.445,
      "std": 0.078,
      "min": 0.298,
      "max": 0.612,
      "range": 0.314
    },
    "alignment": {
      "mean": 0.531,
      "std": 0.054,
      "min": 0.423,
      "max": 0.647,
      "range": 0.224
    }
  },
  "core_specific": {
    "proto_semantic_entropy": {
      "mean": 0.423,
      "std": 0.091,
      "min": 0.256,
      "max": 0.612,
      "range": 0.356
    },
    "proto_semantic_diversity": {
      "mean": 0.387,
      "std": 0.065,
      "min": 0.234,
      "max": 0.523,
      "range": 0.289
    },
    "code_confidence": {
      "mean": 0.577,
      "std": 0.091,
      "min": 0.388,
      "max": 0.744,
      "range": 0.356
    },
    "dominant_code_distribution": {
      "unique_count": 8,
      "mode": 3,
      "mode_fraction": 0.235
    },
    "code_dwell_times": {
      "mean_dwell": 12.5,
      "max_dwell": 47,
      "num_transitions": 16
    },
    "code_activation_mean": [0.045, 0.062, 0.089, 0.134, 0.078, 0.056, 0.067, 0.098,
                             0.072, 0.045, 0.038, 0.052, 0.061, 0.043, 0.031, 0.029]
  }
}

ARIA v3 - baseline_long

{
  "core_type": "aria_v3",
  "scenario": "baseline_long",
  "num_steps": 5000,
  "timestamp": "2025-12-06T12:00:00",
  "common_metrics": {
    "coherence": {
      "mean": 0.612,
      "std": 0.074,
      "min": 0.398,
      "max": 0.782,
      "range": 0.384
    },
    "stability": {
      "mean": 0.756,
      "std": 0.053,
      "min": 0.612,
      "max": 0.876,
      "range": 0.264
    },
    "intensity": {
      "mean": 0.478,
      "std": 0.068,
      "min": 0.312,
      "max": 0.645,
      "range": 0.333
    },
    "alignment": {
      "mean": 0.567,
      "std": 0.048,
      "min": 0.456,
      "max": 0.689,
      "range": 0.233
    }
  },
  "core_specific": {
    "rsv_mean": [0.342, 0.456, 0.523, 0.412, 0.389, 0.467,
                 0.378, 0.445, 0.512, 0.398, 0.423, 0.489],
    "relation_density": {
      "mean": 0.412,
      "std": 0.056,
      "min": 0.298,
      "max": 0.534,
      "range": 0.236
    },
    "clustering_index": {
      "mean": 0.578,
      "std": 0.067,
      "min": 0.423,
      "max": 0.712,
      "range": 0.289
    },
    "asymmetry_index": {
      "mean": 0.187,
      "std": 0.034,
      "min": 0.112,
      "max": 0.267,
      "range": 0.155
    }
  }
}

ARIA v1 - mild_perturbation

{
  "core_type": "aria_v1",
  "scenario": "mild_perturbation",
  "num_steps": 1000,
  "timestamp": "2025-12-06T12:00:00",
  "common_metrics": {
    "coherence": {
      "mean": 0.545,
      "std": 0.112,
      "min": 0.312,
      "max": 0.756,
      "range": 0.444
    },
    "stability": {
      "mean": 0.698,
      "std": 0.089,
      "min": 0.512,
      "max": 0.867,
      "range": 0.355
    },
    "intensity": {
      "mean": 0.423,
      "std": 0.095,
      "min": 0.234,
      "max": 0.623,
      "range": 0.389
    },
    "alignment": {
      "mean": 0.512,
      "std": 0.078,
      "min": 0.356,
      "max": 0.678,
      "range": 0.322
    }
  },
  "core_specific": {
    "symbol_entropy": {
      "mean": 0.356,
      "std": 0.098,
      "min": 0.178,
      "max": 0.567,
      "range": 0.389
    },
    "dominant_symbol_distribution": {
      "unique_count": 6,
      "mode": 2,
      "mode_fraction": 0.312
    },
    "symbol_dwell_times": {
      "mean_dwell": 23.8,
      "max_dwell": 89,
      "num_transitions": 42
    },
    "symbol_activation_mean": [0.087, 0.156, 0.234, 0.145, 0.112, 0.098, 0.089, 0.079]
  }
}

---

Workflow

Step 1: Generate Reference Runs

# Generate all scenarios for all cores
python tools/aria_reference_runs.py --scenario all --output-dir reference_runs

# Or generate specific scenario/core combinations
python tools/aria_reference_runs.py --scenario baseline_quiet --cores aria_v4 aria_v3
python tools/aria_reference_runs.py --scenario baseline_long --output-dir reference_runs

Output structure:

reference_runs/
├── baseline_quiet/
│   ├── cfm_v2_run.json
│   ├── aria_v0_run.json
│   ├── aria_v1_run.json
│   ├── aria_v2_run.json
│   ├── aria_v3_run.json
│   └── aria_v4_run.json
├── baseline_long/
│   └── ...
├── mild_perturbation/
│   └── ...
└── high_variability/
    └── ...

Step 2: Extract Fingerprints

# Batch process all reference runs
python tools/aria_fingerprint.py --batch reference_runs --output-dir fingerprints

# Or process a single run
python tools/aria_fingerprint.py --input reference_runs/baseline_quiet/aria_v4_run.json \
    --output fingerprints/baseline_quiet/aria_v4_fingerprint.json

Output structure:

fingerprints/
├── baseline_quiet/
│   ├── cfm_v2_fingerprint.json
│   ├── aria_v0_fingerprint.json
│   ├── aria_v1_fingerprint.json
│   ├── aria_v2_fingerprint.json
│   ├── aria_v3_fingerprint.json
│   └── aria_v4_fingerprint.json
├── baseline_long/
│   └── ...
└── ...

Step 3: Compare Fingerprints

# Compare two fingerprints
python tools/aria_fingerprint.py --compare \
    fingerprints/baseline_quiet/aria_v4_fingerprint.json \
    fingerprints_v2/baseline_quiet/aria_v4_fingerprint.json

Example output:

============================================================
FINGERPRINT COMPARISON
============================================================

Fingerprint 1: fingerprints/baseline_quiet/aria_v4_fingerprint.json
  Core: aria_v4
  Scenario: baseline_quiet

Fingerprint 2: fingerprints_v2/baseline_quiet/aria_v4_fingerprint.json
  Core: aria_v4
  Scenario: baseline_quiet

Core type match: True
Scenario match: True

Total differences: 12
Significant differences (>5% relative): 2

Significant differences:
  common_metrics.coherence.mean: 0.582 -> 0.621
    (relative diff: 6.70%)
  core_specific.proto_semantic_entropy.mean: 0.423 -> 0.398
    (relative diff: 5.91%)

Regression Detection Workflow

1. Before changes: Generate reference runs and fingerprints
2. Make code changes: Modify parameters, fix bugs, etc.
3. After changes: Generate new reference runs and fingerprints
4. Compare: Check for unexpected behavioral changes

# Before changes
python tools/aria_reference_runs.py --scenario baseline_long --output-dir reference_v1
python tools/aria_fingerprint.py --batch reference_v1 --output-dir fingerprints_v1

# ... make changes ...

# After changes
python tools/aria_reference_runs.py --scenario baseline_long --output-dir reference_v2
python tools/aria_fingerprint.py --batch reference_v2 --output-dir fingerprints_v2

# Compare each core
for core in cfm_v2 aria_v0 aria_v1 aria_v2 aria_v3 aria_v4; do
    python tools/aria_fingerprint.py --compare \
        fingerprints_v1/baseline_long/${core}_fingerprint.json \
        fingerprints_v2/baseline_long/${core}_fingerprint.json
done

---

Interpreting Differences

Acceptable Variations

Small variations (< 5% relative difference) are generally acceptable due to:

• Floating point precision differences
• Timing variations in trajectory sampling
• Natural run-to-run variation in stochastic components

Warning Signs

Investigate if you see:

• Large mean shifts (> 10%): Core behavior may have fundamentally changed
• Std changes: Stability characteristics may be affected
• Distribution mode changes: Dominant patterns shifted
• Dwell time changes: Temporal dynamics affected

Common Causes

---

Safety Invariants

The fingerprint system maintains all ARIA safety invariants:

1. Bounded values: All metrics remain in [0, 1]
2. Non-semantic: No text, labels, or meaning attached to codes/symbols
3. Non-identity: No self-reference or ego modeling
4. Deterministic: Same scenario + seed → same fingerprint
5. Read-only: Tools only read core outputs, never modify behavior

---

Version History