Structured reasoning
The model responds to a strict schema instead of generating an unbounded narrative.
Responsible AI
Decision support, with boundaries.
ScopeGuard is designed to organize evidence and reveal uncertainty. It does not replace project leadership, contractual review, or accountable human judgment.
How the analysis works
ScopeGuard compares the original agreed baseline with every provided source. The model extracts structured requirements, proposed changes, conflicts, questions, dependencies, risks, tasks, and acceptance criteria. Important conclusions carry source IDs so reviewers can return to the underlying evidence.
Evidence first
Requirements, contradictions, and scope warnings must cite the supplied sources.
Built-in safeguards
Human approval gate
Reviewers approve or reject requirements and resolve questions before confirmation.
No authority claims
The analysis is not legal advice and does not determine contractual entitlement.
No external actions
ScopeGuard never contacts clients, assigns work, or changes external systems.
Visible uncertainty
Confidence labels, draft markers, open questions, and limitations remain visible.
Known limitations
- Missing, inaccurate, or outdated material can produce incomplete findings.
- Source citations establish traceability, not truth or legal authority.
- Nuance, sarcasm, informal approvals, and organization-specific language may be misread.
- Confidence is a model assessment, not a statistical guarantee.
- The suggested task plan and owner roles require feasibility review by the delivery team.
Always compare material conclusions against the original documents and involve the appropriate project, commercial, or legal owner when the stakes require it.
Data and security model
The browser sends project material only to ScopeGuard’s server route. The OpenAI API key stays server-side and the model is selected through OPENAI_MODEL. This prototype stores the active workspace locally in the browser for convenience; teams should define retention, access, and data-classification policies before production use.