How We Score

Composite Score (0–100)

Each CDMO receives a weighted composite across four pillars. Scores recalculate as new data arrives — typically daily for FDA actions, weekly for EMA, operational, and capacity signals.

FDA 483 Citation Subsystems & Severity

The Quality Compliance pillar weights every FDA 483 observation by what kind of issue it is (subsystem) and how serious it is (severity 1–5). Source: FDA's public 483 Citations API. We display FDA's verbatim text on every CDMO profile — we do not paraphrase.

Subsystems — citations land in one of nine buckets via deterministic keyword matching on FDA's published descriptions:

Data Integrity — records, audit trails, ALCOA, electronic records, computer system validation.

Environmental Monitoring — aseptic processing, sterility, particulates, cleanroom controls.

Process Validation — qualification, IQ/OQ/PQ, media fills, cleaning validation.

CAPA — investigations, root cause, deviations, OOS handling.

Laboratory Controls — testing, stability, analytical methods, reference standards.

Equipment & Facilities — calibration, maintenance, utilities, HVAC, water systems.

Documentation & Records — SOPs, batch records, written procedures.

Personnel & Training — competency, gowning, supervision.

Other — citations not matching the eight buckets above (typically device or food regulatory citations).

Severity 1–5 uses a rules-as-floor model. Each subsystem has a deterministic floor (data integrity = 5, environmental monitoring and process validation = 4, CAPA and laboratory controls = 3, the rest = 2). For citations where the floor is below 4 and the inspector's description is long enough to merit nuance, we additionally consult Anthropic's Claude with a tight rubric. Final score is the higher of rules and LLM — Claude can promote, never demote. Tiers in the UI: 5 — Critical, 4 — Serious, 3 — Moderate, 2 — Minor, 1 — Trivial.

Validation — the rubric was tested against a 30-citation manual sample with 93% within-±1 agreement and zero LLM bias. The audit harness reruns whenever the prompt or weight table changes.

Repeat observations — flagged when the same site was previously cited for the same regulation under the same subsystem within the prior 36 months. Repeats are weighted 2× because they signal that prior CAPA failed.

Data Sources

How Often Is Data Updated?

What We Don't Score

Self-reported capabilities, survey data, marketing materials, paid directory listings. If it can't be independently verified, it doesn't enter the model.

Score Statuses

Scored — Full composite across all pillars. Requires quality or operational pillar data and sufficient signals for a composite score.

Partial — Scored on 2-3 pillars. Typically missing financial or capacity data. Score shown as "Partial" on CDMO profiles.

Profiled — CDMO profile with modalities and basic data. Scoring data is being gathered.

CDMO Types

Pure CDMO — Dedicated contract manufacturers whose primary business is manufacturing for sponsors.

Hybrid — Companies with both internal pipelines and contract manufacturing arms (e.g., Novartis, Oxford Biomedica).

Platform — Technology platform providers that enable or automate manufacturing (e.g., Cellares, MaxCyte).

Reagent/Material Supplier — GMP reagent, media, or materials suppliers critical to biopharma manufacturing.

National Center — Government-backed or non-profit manufacturing centers (e.g., Cell and Gene Therapy Catapult, CCRM).

Browse CDMO Landscapes by Modality

Frequently Asked Questions