Ledger AI Tax Workbench
Traceable numbers and a work queue ranked by what matters now
Clickable, dependency-free prototype where every value on a return traces to its source document, box and calculation, the dashboard ranks ~208 returns by urgency with visible reasons, and one affordance system shows what each field allows.
Visual evidence
Ranked work queue
Case study
Problem, Build, Outcome
Problem
Preparers cannot trust AI-extracted numbers they cannot trace, and reporting dashboards do not tell anyone what to work on next.
Build
Designed a four-step trace (value, source, location, calculation) with highlighted source documents, a transparent priority score with reason chips and role-aware ranking, and five field states defined once and reused everywhere.
Outcome
A reviewer can verify any number without re-deriving it, managers see their sign-offs first, and the design holds at hundreds of returns with search, filters and pagination.
Architecture
System Flow
Architecture Notes
- Single HTML file with no build step or dependencies.
- Seeded procedural data gives every generated return its own consistent source documents.
- Field state drives color, tag and available actions together, gated by role.
Engineering choices
Technical Decisions
Showed the calculation, not just a source link, so trust does not require re-deriving numbers.
Never used color alone for state; every state carries an icon and a word.
Kept ranking explainable with per-row reasons instead of an opaque priority label.
Evidence
Proof Points
Click any line to see its source box highlighted and the math applied
Role-aware queue: preparer vs. manager ordering
Responsive down to phone width
Next read