AGENTS.md

Test data policy

Never add real customer data to tests, fixtures, snapshots, examples, or documentation committed to this repository.
Always use anonymized placeholders for user-identifying and customer-identifying fields such as usernames, organization names, and cost center names.
Prefer obviously synthetic values like test-user, mona, octocat, hubot, example-org, octodemo, and Cost Center A or Octocats so they cannot be mistaken for production data.

When changing parsing, products, SKUs, included credits, or aggregation logic, update the affected parser tests, aggregator tests, and any expectations for product or model breakdowns.
If supported CSV values or user-visible billing behavior change, update the relevant docs in the same change.

Keep App.tsx focused on orchestration and layout; extract substantial view logic into dedicated view or component files.
Prefer precomputed aggregator results over recalculating report data inside views.
Be conservative with React state for large reports: avoid retaining duplicate derived datasets or raw data when aggregated results are sufficient.

Assume uploaded CSV reports can be large.
Avoid changes that significantly increase retained in-memory data, especially duplicated aggregates, large per-view caches, or UI patterns that can hold onto detached DOM trees.