Evidence Levels#

PhaseLab v0.6.1 introduces evidence levels to classify guide validation strength.

The Three Levels#

Level A: Hardware-Validated

Guides validated on real quantum hardware (IBM Quantum).

Coherence measured on actual qubits
Includes hardware noise effects
Strongest evidence of reliability
R values typically 0.84-0.97

Level B: VQE-Simulated

Guides validated using quantum simulation (VQE).

Full quantum algorithm execution
Simulated noise models
Good evidence of reliability
R values match hardware validation

Level C: Heuristic Only

Guides scored using fast heuristic proxy.

Hamiltonian coefficient variance
No quantum simulation
Weak evidence of reliability
R values cluster around 0.68-0.69

Why Evidence Levels Matter#

The Problem with Heuristic-Only

Before v0.6.1, all coherence scores were weighted equally. This caused issues:

False Confidence: Heuristic R ≈ 0.68 looks good but doesn’t reflect true coherence
Poor Discrimination: All guides score similarly in heuristic mode
Misleading Rankings: Heuristic coherence dominated ranking despite being a proxy

The Solution

Evidence levels ensure:

Honest Reporting: Users know what validation was performed
Appropriate Weighting: Heuristic scores have reduced influence
Clear Recommendations: Experimental validation prioritizes Level B+ guides

How Levels Affect Scoring#

Score Weight by Level

Example Impact

Consider two guides with identical non-coherence scores of 0.70:

# Guide 1: Level B (quantum R = 0.92)
score_1 = 0.70 * 0.70 + 0.30 * 0.92 = 0.49 + 0.28 = 0.77

# Guide 2: Level C (heuristic R = 0.68)
score_2 = 0.70 * 0.95 + 0.05 * 0.68 = 0.67 + 0.03 = 0.70
# Capped at 0.85 max

Guide 1 (Level B) appropriately ranks higher.

Upgrading Evidence Levels#

From C to B

Run quantum simulation:

from phaselab.crispr import compute_guide_coherence

# Upgrade to Level B
r_bar = compute_guide_coherence(guide_seq, mode="quantum")
guide['evidence_level'] = 'B'
guide['quantum_coherence'] = r_bar

From B to A

Validate on IBM Quantum hardware:

from phaselab.quantum import QuantumCoherenceValidator

validator = QuantumCoherenceValidator(backend='ibm_torino')
result = validator.validate(guide_seq)

guide['evidence_level'] = 'A'
guide['hardware_coherence'] = result['coherence']

Recommended Workflow#

Initial Design (Level C)
- Fast screening of thousands of candidates
- Use heuristic mode
- Filter by GO status and basic criteria
Candidate Refinement (Level B)
- Top 10-20 candidates
- Run quantum simulation
- Re-rank by quantum coherence
Final Validation (Level A)
- Top 3-5 candidates
- Hardware validation if available
- Confirm experimental priorities

from phaselab.crispr import design_guides, compute_coherence_batch

# Level C: Initial design
guides = design_guides(sequence, tss_index)
guides['evidence_level'] = 'C'

# Filter
candidates = guides[guides['go_no_go'] == 'GO'].head(20)

# Level B: Quantum validation
quantum_r = compute_coherence_batch(
    candidates['sequence'].tolist(),
    mode="quantum"
)
candidates['quantum_coherence'] = quantum_r
candidates['evidence_level'] = 'B'

# Level A: Hardware (if available)
# ... IBM Quantum validation ...

Displaying Evidence Levels#

Include evidence level in reports:

for idx, row in guides.iterrows():
    level = row.get('evidence_level', 'C')
    r_bar = row.get('quantum_coherence', row.get('coherence', 'N/A'))

    level_desc = {
        'A': 'Hardware-validated',
        'B': 'VQE-simulated',
        'C': 'Heuristic'
    }

    print(f"Guide: {row['sequence'][:15]}...")
    print(f"  Evidence: Level {level} ({level_desc[level]})")
    print(f"  Coherence: {r_bar}")
    print(f"  Score: {row['combined_score']:.3f}")

Best Practices#

Always report evidence level in publications and reports
Prioritize Level B+ for experimental validation
Use Level C only for initial screening
Document validation method for reproducibility
Consider upgrading critical candidates to Level A