Clinical Scoring Skill

Purpose

Score a biotech company's clinical development program to produce a normalized 0-100 clinical score. This skill encodes the exact rules, thresholds, and lookup tables from Wake Robin's pipeline (Module 4 + PoS Engine) so that any agent or analyst can reproduce the scoring deterministically.

Preconditions

• •All arithmetic MUST use Decimal (never float). Initialize from strings: Decimal("0.40").
• •All dates MUST be ISO 8601 (YYYY-MM-DD). Never call datetime.now().
• •PIT safety: only use data where source_date <= as_of_date - 1.
• •Rounding: ROUND_HALF_UP. Scores to 2 dp (0.01), rates to 4 dp (0.0001).

Step 1: Determine Lead Phase

Map each company's most advanced trial to a canonical phase. Use the highest phase across all PIT-admissible trials for that ticker.

Phase ordering (lowest to highest): preclinical < phase_1 < phase_1_2 < phase_2 < phase_2_3 < phase_3 < nda_bla < commercial

Step 2: Look Up Base Stage Score

Step 3: Compute Phase Score (Module 4, 0-30 pts raw)

Phase Progress Bonus (0-5 pts)

Step 4: Map Indication

Use indication_mapper.py (v2.0.0). Mapping precedence (highest to lowest):

1. •ticker_overrides_v3 (PIT-safe, has effective_from/effective_until) -> confidence 0.95
2. •ticker_overrides (legacy, no time-bounds) -> confidence 0.85
3. •condition_patterns (regex word-boundary matching, 2+ matches) -> confidence 0.80
4. •condition_patterns (single match) -> confidence 0.65
5. •ta_fallback (therapeutic area only) -> confidence 0.50
6. •phase_only (no condition data) -> confidence 0.30

Category Aliases

Step 5: Look Up PoS Benchmarks (BIO 2011-2020)

Phase 1 LOA (Likelihood of Approval)

Phase 2 LOA

Phase 3 LOA

NDA/BLA LOA

PoS Score Conversion

pos_score = LOA_probability * 100

Score range: 0.00 to 100.00.

Step 6: Apply Confidence Gating

Stage-Adjusted Base Confidence

Indication Confidence Modifiers (additive)

Data Quality Confidence Modifiers (additive)

Confidence Bounds

final_confidence = clamp(base + indication_modifier + quality_modifier, 0.20, 0.95)

• •CONFIDENCE_HIGH: >= 0.70
• •CONFIDENCE_MEDIUM: >= 0.55
• •CONFIDENCE_LOW: >= 0.30
• •GATING_THRESHOLD: 0.40 — Below this, PoS contributes 0 weight to composite.

Step 7: Apply Optional Multipliers

Step 8: Recency Scoring (0-5 pts)

Based on days since last trial update (use PIT date field priority: first_posted > last_update_posted > source_date > collected_at).

• •RECENCY_STALE_THRESHOLD: 730 days (2 years). Triggers 20% penalty on score.
• •RECENCY_UNKNOWN_PENALTY: 2.5 (neutral score when date unknown).

Step 9: Trial Count Bonus (0-5 pts, piecewise linear)

Step 10: Indication Diversity Bonus (0-5 pts)

Based on count of unique condition tokens across all trials.

Step 11: Design Quality Scoring (0-25 pts)

• •Base: 12 pts
• •Randomized: +5 pts
• •Double-Blind: +4 pts
• •Single-Blind: +2 pts (mutually exclusive with double-blind)
• •Strong Endpoint: +4 pts
• •Weak Endpoint: -3 pts (mutually exclusive with strong)

Strong Endpoint Patterns

overall survival, OS, progression-free survival, PFS, complete response, CR, objective response rate, ORR, disease-free survival, DFS, event-free survival, EFS, major molecular response, MMR

Weak Endpoint Patterns

biomarker, pharmacokinetic, PK, safety, tolerability, dose-finding, maximum tolerated dose, MTD

Step 12: Execution Track Record (0-25 pts)

• •Base: 12 pts
• •Completion Rate Contribution: completion_rate * 10 pts
• •Termination Rate Penalty: termination_rate * 8 pts (subtracted)

Trial Status Quality Weights

Step 13: Endpoint Strength (0-20 pts)

• •Base: 10 pts
• •Strong endpoint found: +2 pts per occurrence
• •Weak endpoint found: -1 pt per occurrence

Step 14: Compute Total Clinical Score

raw_total = phase_score + phase_progress + trial_count_bonus
          + diversity_bonus + recency_bonus + design_score
          + execution_score + endpoint_score

clinical_score = (raw_total / 120) * 100    # Normalize to 0-100
clinical_score = clamp(clinical_score, 0, 100)

Maximum raw total = 30 + 5 + 5 + 5 + 5 + 25 + 25 + 20 = 120.

Step 15: Commercial Stage Differentiation (PoS Engine)

For companies at commercial stage, apply additional adjustments:

Pipeline Tier Bonuses (LOA adjustment)

Indication-Specific Commercial Risk (LOA adjustment)

Pipeline Diversity Bonus

+0.02 if >= 3 distinct phases represented in pipeline.

Commercial LOA Range

Clamp commercial LOA to [0.82, 1.00].

Severity Classification

Composite Integration

The clinical score enters Module 5 composite with these weights:

PoS Delta Cap: Maximum PoS contribution to composite = 6.0 points.

Pre-Flight Checks

Before scoring, verify:

1. •Trial records are PIT-admissible (source_date <= as_of_date - 1)
2. •pos_benchmarks_bio_2011_2020_v1.json is loaded and contains all 14 indications
3. •Indication mapping covers the ticker (log confidence tier if < 0.65)
4. •At least 1 trial record exists for the ticker (otherwise assign SEV1)
5. •All score components use Decimal arithmetic
6. •Output includes _governance block with score_version, schema_version, pit_cutoff

Source Files