Combined safety harbor: how OpenPlan is governed as open-source software and a managed service.

Internal prototype posture

OpenPlan is currently in supervised early access. Some modules — notably the behavioral modeling runtime — are labeled internal prototype only. Those outputs are shared with caveats, not as planning-grade conclusions.

Open-source license boundary

OpenPlan source code is made available under the Apache License, Version 2.0, unless a file says otherwise. That source license does not grant rights to Nat Ford trademarks, private credentials, client confidential information, third-party datasets, third-party media, or client-specific deliverables.

Managed hosting and services

Nat Ford commercial terms cover managed hosting, onboarding, support, planning services, and custom implementation work around the open-source core. Workspace setup, service scope, and payment setup are gated and reviewed individually.

Screening-grade boundary

Where a module produces screening-grade output, the platform marks it as such on the surface that shows it. Planning-grade consumption is not permitted until the underlying evidence is promoted by a qualified reviewer.

AI-assisted drafting is used throughout OpenPlan for analysis, summarization, scaffolding, and data cleanup. AI output is labeled where it appears so reviewers can distinguish it from hand-authored work.

Final analysis, conclusions, and client-critical decisions must be reviewed by qualified staff before being published, shared externally, or relied upon for regulatory or funding purposes.

AI-drafted content must not be misrepresented as independently verified analysis. The platform is built to surface this distinction, and users of the platform agree to honor it.

Behavioral modeling runtime outputs carry a documented error envelope. The current published ceiling is max APE 237.62% — suitable for screening-grade comparison, not for funding-grade conclusions.

Aerial imagery outputs (orthos, models, surfaces) produced through the Aerial Operations OS modules are measurable but prototype-grade until the QA bundle for that mission is signed off.

Grants and RTP decision records are authoritative once a packet is adopted by the responsible agency — not before. Drafts, preset packets, and preview surfaces are not substitutes for adopted records.

Truth without spin: assumptions, limitations, and uncertainty are labeled rather than smoothed over. Regulatory and technical claims are cited.

Fair exchange: managed-hosting and service pricing aim to be sustainable, transparent, and non-predatory. Published service baselines are not proprietary software-license fees.

Community protection: analysis and recommendations do not shift burden onto disadvantaged or tribal communities for convenience or optics. This is a standing constraint, not a soft preference.

Rapid repair: when something is wrong, the operator communicates early and makes it right through rework, credits, or refunds as appropriate.

Responsible AI use: AI accelerates drafting and QA; it does not replace qualified review on client-critical outputs.

Accountability: final deliverables pass gates for truthfulness, citation support, equity impact, confidentiality, and client-safe readability.