Create shared PolicyEngine household wizard infrastructure

[MaxGhenis]

Context

PolicyEngine apps are starting to duplicate household setup flows. Cliff Watch now has a local guided wizard; Coverage Compass has a household/coverage wizard; California Wealth Tax has reusable wizard patterns; app-v2 has the strongest household model and metadata-driven variable inputs, but its /households/create flow is a dense editor rather than a lightweight standalone-app wizard.

We should create shared infrastructure so PE apps collect households consistently without copying app-specific UI.

Problem

Household setup differs across apps in ways that affect results and UX:

• Some apps set default state or ages.
• County can be free text instead of a state-filtered dropdown.
• Some UIs ask tax filing status when users should be asked marital status.
• Disability, blindness, student, pregnancy, and care-needs fields are inconsistent.
• App-v2 has reusable household semantics, but standalone apps need a lighter guided flow.

Scope

V1 should be US-first. The generic wizard primitives can be country-neutral, but the first household module should explicitly target US household setup and PolicyEngine-US semantics.

Goals

• Add shared wizard primitives: step state, progress, Back/Continue, validation gating, review navigation.
• Add a shared US household draft contract with blank initial state: no default state, no default age, no default marital status.
• Ask marital status in UI; derive filing status only in app/model adapters.
• Support state-filtered county dropdowns with a decided metadata source before implementation.
• Support app-controlled person fields such as pregnancy, disabled, blind, student, and care-needs.
• Provide adapters rather than forcing one UI shell across apps.
• Validate app-v2 compatibility early with round-trip fixtures.

Non-goals

• Do not replace app-v2's full household editor in the first pass.
• Do not move Cliff Watch, Coverage Compass, or wealth-tax-specific fields into shared core.
• Do not migrate every PE app at once.
• Do not design a fake all-country abstraction while only implementing US behavior.

Phase 0: Decisions

• Choose package home and ownership: likely a new shared package, not model logic inside UI kit.
• Decide county metadata source, representation, update cadence, and fallback behavior.
• Define dependency direction so app-v2 and standalone apps can consume the package without circular coupling.
• Decide URL-state ownership: shared draft serialization or app-local serialization.

Phase 1: Audit and contract spike

• Build a field matrix for Cliff Watch, Coverage Compass, California Wealth Tax, app-v2, and other active household apps.
• Create app-v2 round-trip fixtures before locking the draft shape:
  • Single adult.
  • Married adults.
  • Adult plus dependent child.
  • Disabled person.
  • Student.
  • County/state case.
• Verify the draft can map to app-v2 Household groups, members, periods, variables, and entity relationships.

Phase 2: Wizard primitives

• Extract copy-neutral primitives:
  • useWizardSteps
  • WizardProgress
  • WizardNavigation
  • WizardOptionCard
  • WizardReviewList
• Test step validation, skipped/conditional steps, editing earlier answers, review navigation, and disabled states.
• Include accessibility criteria: focus movement, error announcement, aria-current or equivalent step semantics, keyboard option cards, and mobile viewport checks.

Phase 3: US household builder

• Add createBlankDraft, createDraftFromExisting, and normalizeLegacyDraft.
• Add explicit adapters:
  • validate
  • toPayload
  • fromPayload
  • renderExtraSteps
  • renderReviewItems
  • serialize / deserialize where needed
• Keep common flags metadata-backed where possible rather than hard-coded as universal booleans.

Phase 4: First consumers

• Migrate Cliff Watch first:
  • Preserve wage-axis behavior.
  • Preserve SSI/SSDI and fixed adult earnings.
  • Preserve program take-up, costs, URL params, and payload parity.
• Migrate Coverage Compass second:
  • Preserve ESI, coverage fields, pregnancy behavior, URL/share behavior, and calculation payload parity.

Acceptance criteria

• Shared wizard primitives work without household-specific code.
• Shared US household draft starts blank for state, ages, and marital status.
• County dropdown is filtered by selected state and has tested legacy normalization.
• Filing status is not user-facing in shared UI.
• App-v2 conversion tests pass before consumer migration.
• App-specific fields stay in adapters/slots.
• Cliff Watch and Coverage Compass retain calculation payload parity.
• Accessibility and responsive behavior are tested for the shared wizard pieces.