An Autonomous Analysis of IRS Form 990 Filings
Intuitionist × AgentAcademy
March 2026 · Target: NVSQ
"To what extent do U.S. nonprofit organizations incorporate technocratic language in their Form 990 mission statements, and how does adoption vary by organizational characteristics?"
Fully autonomous: data collection → coding → analysis → peer review → revision
| SERVICE | Direct beneficiary help |
| CAPACITY | Support other orgs |
| FELLOWSHIP | Member benefit |
| ADVOCACY | Policy change |
| RESEARCH | Knowledge production |
| OUTCOME_ORIENTED | 9.7% |
| PROFESSIONAL | 3.7% |
| EFFICIENCY | 1.3% |
| ACCOUNTABILITY | 1.3% |
| EVIDENCE_BASED | 0.2% |
But adoption is layered onto service missions, not replacing them
Service remains the dominant orientation
Large orgs 4× more likely to use technocratic language than small orgs
| Subsector | Technocratic Rate | Effect |
|---|---|---|
| Community Improvement | 50.0% | OR = 3.55 (p = .034) |
| Education | 23.8% | OR = 2.37 (ns) |
| Human Services (ref) | 17.3% | — |
| Religion | 6.7% | OR = 0.25 (ns) |
| Mutual Benefit | 5.8% | OR = 0.36 (p = .020) |
| Codex (GPT-5.4) | 232 items |
| Gemini 2.5 Pro | 233 items |
Independent coding, identical instructions
| Codex | Reject | Sample too small |
| Gemini CLI | Major Rev | Missing NTEE |
✓ All concerns addressed
30-item sample coded independently → κ = 0.935 (almost perfect agreement)
Technocratic language is layered onto service missions, not replacing them. Supports "institutional work" rather than wholesale co-optation.
Growing organizations should anticipate pressure to adopt outcome language. But service orientation remains durable—coexistence is possible.
Two-level coding (frame + modifiers) captures diffusion patterns invisible to binary schemes.
"The nonprofit sector is adapting to accountability pressures—selectively incorporating technocratic language while preserving service-oriented identity."