Essential session storage is required to sign in. Optional analytics remain off unless you allow them.
Build a useful AI workflow that answers, drafts, classifies, or prepares an artifact from approved knowledge. Requirements: (1) define source ownership, freshness, metadata, and access boundaries; (2) retrieve only eligible, relevant context and show citations or evidence in the user experience; (3) separate trusted instructions, trusted context, and untrusted input; (4) request a structured output and validate its schema and business constraints; (5) design human review with edit, rejection, escalation, and feedback capture; (6) create a versioned evaluation set with normal, edge, no-answer, stale-source, and adversarial cases; (7) define release gates and show evaluation results.
Source ownership, authorization, metadata, retrieval, citations, prompt/context boundaries, output validation, and no-answer behavior form a coherent trustworthy path.
Representative held-out cases, severity definitions, deterministic and qualitative checks, release thresholds, and regression handling make quality claims inspectable.
Users can inspect evidence, understand uncertainty, edit or reject output, escalate difficult cases, and generate structured feedback without unnecessary workflow friction.
Use this README structure to make the work reviewable.
DevPath fetches a read-only, size-capped snapshot. Submitted code is never executed.
# A grounded AI workflow with evaluation and review ## Goal Build a useful AI workflow that answers, drafts, classifies, or prepares an artifact from approved knowledge. Requirements: (1) define source ownership, freshness, metadata, and access boundaries; (2) retrieve only eligible, relevant context and show citations or evidence in the user experience; (3) separate trusted instructions, trusted context, and untrusted input; (4) request a structured output and validate its schema and business constraints; (5) design human review with edit, rejection, escalation, and feedback capture; (6) create a versioned evaluation set with normal, edge, no-answer, stale-source, and adversarial cases; (7) define release gates and show evaluation results. ## Acceptance criteria - [ ] Grounded system design: Source ownership, authorization, metadata, retrieval, citations, prompt/context boundaries, output validation, and no-answer behavior form a coherent trustworthy path. - [ ] Evaluation and quality gates: Representative held-out cases, severity definitions, deterministic and qualitative checks, release thresholds, and regression handling make quality claims inspectable. - [ ] Reviewable product experience: Users can inspect evidence, understand uncertainty, edit or reject output, escalate difficult cases, and generate structured feedback without unnecessary workflow friction. ## Verification - [ ] Document installation and run commands. - [ ] Record automated checks and their exact commands. - [ ] Add representative output, screenshots, or a short demo where useful. - [ ] Confirm failure paths and known constraints. ## Decision log ### Decision title - Context: - Choice: - Alternatives considered: - Tradeoffs: ## Evidence - Link each rubric criterion to the file, test, or artifact that demonstrates it. ## Known limitations - List what is intentionally out of scope and what should be improved next.