Metadata-Driven Similarity Conversion

Use this skill when converting existing similarity calculators to the metadata-driven approach with configurable, type-specific rules.

5-Step Conversion Process

Step 1: Add Imports

Add these imports to your calculator class:

import no.cantara.electronic.component.lib.metadata.ComponentTypeMetadata;
import no.cantara.electronic.component.lib.metadata.ComponentTypeMetadataRegistry;
import no.cantara.electronic.component.lib.metadata.SimilarityProfile;
import no.cantara.electronic.component.lib.metadata.ToleranceRule;
import no.cantara.electronic.component.lib.specs.base.SpecUnit;
import no.cantara.electronic.component.lib.specs.base.SpecValue;

import java.util.Optional;

---

Step 2: Modify calculateSimilarity() to Check for Metadata First

// Add field
private final ComponentTypeMetadataRegistry metadataRegistry =
    ComponentTypeMetadataRegistry.getInstance();

@Override
public double calculateSimilarity(String mpn1, String mpn2, PatternRegistry registry) {
    if (mpn1 == null || mpn2 == null) return 0.0;

    // ✅ NEW: Try metadata-driven approach first
    Optional<ComponentTypeMetadata> metadataOpt = metadataRegistry.getMetadata(ComponentType.RESISTOR);
    if (metadataOpt.isPresent()) {
        logger.trace("Using metadata-driven similarity calculation");
        return calculateMetadataDrivenSimilarity(mpn1, mpn2, metadataOpt.get());
    }

    // Fallback to legacy pattern-based approach
    logger.trace("No metadata found, using legacy approach");
    return calculateLegacySimilarity(mpn1, mpn2);
}

---

Step 3: Implement calculateMetadataDrivenSimilarity()

private double calculateMetadataDrivenSimilarity(String mpn1, String mpn2, ComponentTypeMetadata metadata) {
    SimilarityProfile profile = metadata.getDefaultProfile();

    // Extract specs from MPNs
    String resistance1 = extractResistance(mpn1);
    String resistance2 = extractResistance(mpn2);
    String package1 = extractPackage(mpn1);
    String package2 = extractPackage(mpn2);
    String tolerance1 = extractTolerance(mpn1);
    String tolerance2 = extractTolerance(mpn2);

    // Short-circuit check for CRITICAL incompatibility
    ComponentTypeMetadata.SpecConfig resistanceConfig = metadata.getSpecConfig("resistance");
    if (resistanceConfig != null &&
        resistanceConfig.getImportance() == SpecImportance.CRITICAL &&
        !resistance1.isEmpty() && !resistance2.isEmpty() &&
        !resistance1.equals(resistance2)) {
        logger.debug("CRITICAL spec mismatch - returning LOW_SIMILARITY");
        return LOW_SIMILARITY;
    }

    double totalScore = 0.0;
    double maxPossibleScore = 0.0;

    // Compare each spec with weighted scoring
    if (resistanceConfig != null && !resistance1.isEmpty() && !resistance2.isEmpty()) {
        ToleranceRule rule = resistanceConfig.getToleranceRule();
        SpecValue<String> orig = new SpecValue<>(resistance1, SpecUnit.OHM);
        SpecValue<String> cand = new SpecValue<>(resistance2, SpecUnit.OHM);

        double specScore = rule.compare(orig, cand);
        double specWeight = profile.getEffectiveWeight(resistanceConfig.getImportance());

        totalScore += specScore * specWeight;
        maxPossibleScore += specWeight;
    }

    // Repeat for other specs (package, tolerance, etc.)
    // ...

    double similarity = maxPossibleScore > 0 ? totalScore / maxPossibleScore : 0.0;

    // Apply boosts for known equivalent groups
    if (areEquivalentParts(mpn1, mpn2)) {
        similarity = Math.max(similarity, HIGH_SIMILARITY);
    }

    return similarity;
}

---

Step 4: Add Spec Extraction Helper Methods

private String extractResistance(String mpn) {
    if (mpn == null || mpn.isEmpty()) return "";
    // Manufacturer-specific extraction logic
    // e.g., Yageo RC0805JR-0710KL → "10K"
    return "";
}

private String extractPackage(String mpn) {
    if (mpn == null || mpn.isEmpty()) return "";
    // e.g., RC0805 → "0805"
    return "";
}

private String extractTolerance(String mpn) {
    if (mpn == null || mpn.isEmpty()) return "";
    // e.g., "JR" → "5%"
    return "";
}

---

Step 5: Update Tests to Use Threshold Assertions

BEFORE (exact equality):

@Test
void shouldCalculateHighSimilarityForSameResistance() {
    double similarity = calculator.calculateSimilarity("RC0805JR-0710KL", "RC0805FR-0710KL", registry);
    assertEquals(0.9, similarity);  // ❌ Exact match - fragile!
}

AFTER (threshold assertion):

@Test
void shouldCalculateHighSimilarityForSameResistance() {
    double similarity = calculator.calculateSimilarity("RC0805JR-0710KL", "RC0805FR-0710KL", registry);
    assertTrue(similarity >= HIGH_SIMILARITY,  // ✅ Threshold - robust!
        "Same resistance should have HIGH_SIMILARITY (>= 0.9), got: " + similarity);
}

Why: Metadata-driven approach produces more precise scores (0.95-0.99 for very similar parts) instead of fixed thresholds (0.9).

---

Dual-Path Implementation

Pattern: Metadata first, legacy fallback.

@Override
public double calculateSimilarity(String mpn1, String mpn2, PatternRegistry registry) {
    if (mpn1 == null || mpn2 == null) return 0.0;

    // Try metadata-driven approach first
    Optional<ComponentTypeMetadata> metadataOpt = metadataRegistry.getMetadata(ComponentType.OPAMP);
    if (metadataOpt.isPresent()) {
        logger.trace("Using metadata-driven similarity calculation");
        return calculateMetadataDrivenSimilarity(mpn1, mpn2, metadataOpt.get());
    }

    // Fallback to legacy pattern-based approach
    logger.trace("No metadata found, using legacy approach");
    return calculateLegacySimilarity(mpn1, mpn2);
}

Example from ResistorSimilarityCalculator:

• Has metadata → use calculateMetadataDrivenSimilarity()
• No metadata → use calculateLegacySimilarity()

Benefits:

• Backward compatibility (no metadata = legacy behavior)
• Gradual migration (convert one calculator at a time)
• Easy rollback (remove metadata to revert to legacy)

---

Short-Circuit Critical Spec Checks

Pattern: Check CRITICAL specs early, return 0.0 on mismatch.

// Short-circuit check for CRITICAL incompatibility
ComponentTypeMetadata.SpecConfig capacitanceConfig = metadata.getSpecConfig("capacitance");
ComponentTypeMetadata.SpecConfig voltageConfig = metadata.getSpecConfig("voltage");

// Capacitance mismatch
if (capacitanceConfig != null &&
    capacitanceConfig.getImportance() == SpecImportance.CRITICAL &&
    !capacitance1.isEmpty() && !capacitance2.isEmpty() &&
    !capacitance1.equals(capacitance2)) {
    logger.debug("CRITICAL capacitance mismatch - returning 0.0");
    return 0.0;
}

// Voltage downgrade (50V → 5V is incompatible)
if (voltageConfig != null &&
    voltageConfig.getImportance() == SpecImportance.CRITICAL &&
    voltage1 > voltage2) {
    logger.debug("CRITICAL voltage downgrade - returning 0.0");
    return 0.0;
}

Example (Capacitor):

• 100nF 50V vs 100nF 5V → return 0.0 (voltage downgrade)
• 100nF 50V vs 1µF 50V → return 0.0 (capacitance mismatch)

---

Weighted Scoring Formula

similarity = totalScore / maxPossibleScore

where:
  totalScore = Σ(specScore × effectiveWeight)
  maxPossibleScore = Σ(effectiveWeight)
  effectiveWeight = profile.getEffectiveWeight(specImportance)

Example calculation (Resistor):

Spec	Value1	Value2	Score	Importance	Weight	Contribution
Resistance	10kΩ	10kΩ	1.0	CRITICAL	1.0	1.0 × 1.0 = 1.0
Package	0805	0805	1.0	MEDIUM	0.4	1.0 × 0.4 = 0.4
Tolerance	5%	1%	0.8	LOW	0.2	0.8 × 0.2 = 0.16

totalScore = 1.0 + 0.4 + 0.16 = 1.56
maxPossibleScore = 1.0 + 0.4 + 0.2 = 1.6
similarity = 1.56 / 1.6 = 0.975 ≈ 0.98

---

Test Expectation Updates

Old Pattern (Exact Equality)

assertEquals(0.9, similarity);
assertEquals(0.0, similarity);
assertEquals(1.0, similarity);

Problem: Metadata produces more precise scores (0.95, 0.99, 0.703) that don't match exact thresholds.

New Pattern (Threshold Assertions)

// HIGH similarity (≥ 0.9)
assertTrue(similarity >= HIGH_SIMILARITY,
    "Expected HIGH_SIMILARITY (>= 0.9), got: " + similarity);

// MEDIUM similarity (≥ 0.5)
assertTrue(similarity >= MEDIUM_SIMILARITY,
    "Expected MEDIUM_SIMILARITY (>= 0.5), got: " + similarity);

// LOW similarity (< 0.5)
assertTrue(similarity < MEDIUM_SIMILARITY,
    "Expected LOW_SIMILARITY (< 0.5), got: " + similarity);

// ZERO similarity
assertEquals(0.0, similarity, 0.001,
    "Expected ZERO similarity for incompatible parts");

Constants:

private static final double HIGH_SIMILARITY = 0.9;
private static final double MEDIUM_SIMILARITY = 0.5;
private static final double LOW_SIMILARITY = 0.3;

---

5 Remaining Calculators

Calculator	Complexity	Priority	Notes
PassiveComponentCalculator	Medium	HIGH	Generic passive handler, needs multi-type metadata
MicrocontrollerSimilarityCalculator	High	HIGH	Complex specs (flash, RAM, peripherals)
MCUSimilarityCalculator	High	HIGH	Similar to Microcontroller, may merge
LevenshteinCalculator	Low	LOW	String-based, may not need metadata
DefaultSimilarityCalculator	Low	LOW	Fallback, intentionally simple

Recommendation: Start with PassiveComponentCalculator (straightforward, high impact).

---

Behavior Improvements

OpAmp Example

BEFORE (pattern-based):

• LM358 vs MC1458: 0.9 (equivalent group boost)
• LM358 vs LM324: 0.7 (same manufacturer, different config)

AFTER (metadata-driven):

• LM358 vs MC1458: 1.0 (exact configuration + family match)
• LM358 vs LM324: 0.3 (short-circuit on configuration: dual vs quad)

Memory Example

BEFORE:

• 24LC256 vs AT24C256: 0.7+ (equivalent EEPROM)
• 24LC256 vs 24LC512: 0.3 (different capacity)

AFTER:

• 24LC256 vs AT24C256: 0.85 (exact match on type, capacity, interface)
• 24LC256 vs 24LC512: 0.3 (short-circuit on capacity)

Sensor Example

BEFORE:

• ADXL345 vs ADXL346: 0.3 (different suffix)
• DS18B20 vs DS18B20+: 0.9 (known equivalent boost)

AFTER:

• ADXL345 vs ADXL346: 0.703 (same type + interface = MEDIUM)
• DS18B20 vs DS18B20+: 1.0 (equivalent variants boost)

---

Migration Checklist

• Add imports (Step 1)
• Get metadata in calculateSimilarity() (Step 2)
• Implement calculateMetadataDrivenSimilarity() (Step 3)
• Add spec extraction helpers (Step 4)
• Implement short-circuit CRITICAL checks
• Implement weighted scoring loop
• Apply equivalent group boosts
• Update tests to threshold assertions (Step 5)
• Run full test suite (verify backward compatibility)
• Check metadata configuration (SpecImportance, ToleranceRule)

---

Learnings & Quirks

Metadata is Optional

If no metadata exists for a type, calculator falls back to legacy behavior. This enables gradual migration.

Spec Extraction is Manufacturer-Specific

Each manufacturer encodes specs differently in MPNs. Extraction logic must handle multiple formats.

Threshold Assertions are More Robust

Exact equality (assertEquals(0.9, ...)) breaks when metadata produces 0.95 or 0.99. Use threshold assertions (>= HIGH_SIMILARITY).

Short-Circuit Saves Computation

Checking CRITICAL specs early avoids unnecessary extraction and comparison of other specs.

---

See Also

• /similarity-metadata

  - Complete metadata framework documentation

• /component-spec-extraction

  - Spec extraction patterns
• 12 converted calculators - Use as migration templates

• ComponentTypeMetadata.java

  - Metadata configuration

• SimilarityProfile.java

  - Context-aware profiles