1. Start with the responsible authority
Tax schemes use tax-authority guidance; immigration and citizenship milestones use the responsible government or EU institution. Secondary sources can help discover a rule, but do not become the final numeric source where a primary source is available.
2. Store the rule, year and source together
A number such as a salary threshold is not stored as an orphaned constant. The production data model should keep its country, scheme, tax year, unit, eligibility context, verification date and source URL together.
3. Separate screening from final eligibility
Some conditions are binary enough for a form. Others require legal interpretation or documents. The interface uses phrases such as “potential match” when the tool has only screened a subset of the rules.
4. Prefer “unknown” to fabricated precision
If an exchange-rate call fails, the tax tool stops. If an official nationality page is being updated and a reliable current rule cannot be established, the timeline does not invent a date.
5. Review high-impact rules regularly
Annual salary thresholds should be checked at least annually and after known legislative changes. Pages expose a verification date so stale data can be found and corrected.
6. AI is an assistant, not the source
AI can help write code, transform structured facts into plain language and generate drafts. Calculator decisions come from structured, verified rules rather than free-form generated prose.