Prioritized learning backlog

Implemented and bounded

The fictional Northstar workspace already supports evidence search and filters, manual examples, duplicate grouping that retains original quotes, distinct-account display, editable impact/confidence/effort, simulated tag review, rationale and planning state, CSV shortlist export, Undo, local persistence, Reset, and responsive controls. These are demonstrable prototype behaviors, not proof that a team should build a production feedback platform.

Next: resolve decision comprehension

1. Run the proposed five-participant session using F01/F02 import recovery and a saved-report comparison. This is the dependency for all feature expansion because it tests whether people can trace evidence and explain the score.
2. Review consented notes, score the Validation rubric, and revise the evidence layout or language when two participants repeat the same misunderstanding.
3. Test whether users need a group-level view. Do not add group-level scoring unless participants can first distinguish preserved individual signals from unique affected accounts.

Deferred until evidence changes the case

Sanitized CSV import is deferred until the core workflow is understood and a participant comparison shows a material import barrier. Shared state, permissions, CRM connections, customer updates, and real AI tags are later work because each requires access, privacy, retention, cost, provenance, error handling, and evaluation decisions. No assignees, dates, or delivery commitments are implied by this ordering.