Methodology
How scores are computed and what the data does and does not show.
Scoring formula
Three components, weighted and combined into a single 0–100 overall health score:
- Financial resilience (40%) — reserve months, deficit history, liabilities ratio, earned-income share.
- Governance risk (35%) — compensation-to-revenue ratio (primary signal), control concentration where state AG data is available.
- Program scale (25%) — operating revenue magnitude relative to peers.
Tiers: Exemplary 75+, Stable 60–74, Developing 45–59, Priority Review <45.
Data sources
- IRS Form 990 — revenue, expenses, aggregate compensation, program descriptions. Via ProPublica Nonprofit Explorer.
- IRS TEOS XML (Part VII) — per-officer name, title, hours/week, compensation. 391,713 records across 14,176 orgs in the current database snapshot.
- State AG charity registries — officer and director listings, control-concentration disclosures. Currently NM (COROS), OR (DOJ CT-12), CO (SOS) in depth.
Governance coverage
Of 441,323 scored arts and music nonprofits, 14,176 have at least one year of IRS 990 Part VII compensation data. The remaining 427,147 are either all-volunteer organizations or filed 990-N (e-postcard) returns that omit compensation detail. These orgs receive a neutral governance score of 50, reflecting absence of a compensation signal — not missing data.
Limitations
- Scores reflect the latest fiscal year available; lag varies from 6 months to 3 years depending on filing.
- Control-concentration signals are only fully available for organizations registered in NM, OR, or CO at present.
- ArtMetrics Foundation is descriptive, not predictive of individual organizational outcomes. Signals describe patterns associated with elevated risk.
Data Rights & Dependencies
Our public data foundation is built on the following dependencies:
- ProPublica Nonprofit Explorer: Form 990 financial summaries are sourced via ProPublica, which itself processes public IRS data releases. We do not claim ownership over IRS public records.
- IRS TEOS XML: Part III and Part VII data are ingested directly from IRS XML bulk downloads.
- State Registries: We parse public charity registry datasets made available by the respective state Attorney General offices.
Model Governance & LLM Scope
To address AI accuracy and risk, we strictly govern how Large Language Models (LLMs) are used in our pipeline:
- Segregation from Scoring: LLMs are never used to calculate the 0-100 governance or financial scores. The scoring engine is 100% deterministic, mathematically verifiable, and rule-based.
- Qualitative Scope: LLMs are strictly limited to parsing IRS Form 990 Part III narratives to extract medium, demographics, and community impact.
- Correction: If an LLM incorrectly hallucinates or misinterprets an organization's mission, authorized organization officers can claim their profile to overwrite the AI-generated narrative with a verified statement.
Reputational Risk & Appeals Process
We recognize that assigning a "Critical" or "At-Risk" score can create reputational exposure for a nonprofit. However, our scores are explicitly framed as a "lens, not a shield." They are prompts for deeper conversation between funders and nonprofits, not definitive verdicts of failure.
Appeals & Context: If an organization believes their public score misrepresents their operational reality (e.g., a one-time capital campaign triggered a financial flag), authorized officers can claim their profile and append a public Contextual Rationale. This ensures funders see the organization's explanation right alongside the raw data score.
The Narrative Overlay (Qualitative Data)
Beyond quantitative financial infrastructure, ArtMetrics employs an AI-driven Qualitative Parsing Engine to analyze IRS Form 990 Part III (Statement of Program Service Accomplishments).
- Extraction: We parse gigabytes of raw IRS TEOS XML files to extract self-reported narrative descriptions for thousands of arts and culture organizations.
- Analysis: We evaluate these unstructured narratives using large language models to surface each organization's distinctive mission profile—identifying primary mediums, target demographics, and community impact.
- Purpose: This ensures our platform doesn't just measure financial stability, but actively surfaces and celebrates the vibrant community work happening inside the organization.
Organizational Trajectories
Rather than relying solely on single-year snapshots, ArtMetrics evaluates organizations across specific time horizons (3, 5, and 10+ years) to establish one of 12 Organizational Thumbprints.
- Analysis: We analyze historical revenue, compensation, and program score trends to classify organizations into trajectories like "Resource Divergence", "Governance Lag", or "Institutional Maturation".
- Utility: This longitudinal approach contextualizes current risk profiles, distinguishing between a temporary acute crisis and a long-term structural deficit.
Officer Network & Person Search
We maintain a comprehensive, searchable database of 391,713 historical officer records spanning 705,798 organization-years across the nonprofit arts sector.
- Connection: By linking leaders to their respective organizations and those organizations to specific structural trajectories, we enable unprecedented peer-to-peer networking.
- Shared Experience: Leaders facing a "Founder's Trap" or "Compliance Decay" can find and connect with peers who have navigated the exact same structural challenges, fostering a community built on shared operational reality rather than just artistic discipline.
Technical Whitepaper
For the complete, citable specification—every scoring formula and threshold, the full data pipeline, the twelve trajectory thumbprints, the qualitative narrative overlay, and a plain statement of the engine's limitations—read the technical whitepaper.