Examples gallery

Ready-to-run ModelSpec documents that double as demos, integration fixtures, and starting points for your own models. The canonical copies live in valem-ui/src/examples/ and on the console test classpath. Load one into a running backend with:

curl -X POST localhost:8080/models -H 'Content-Type: application/json' \
     -d @valem-ui/src/examples/insurance-quote.json

(or pick it from the Load an example panel in the management UI).

The specs carry two underscore-prefixed metadata fields, _name and _description, that are not part of the ModelSpec schema. The management UI strips them before POST /models; a raw curl works because unknown top-level fields are ignored on deserialization.

Featured: quoting & eligibility models

insurance-quote.json — Term Life Insurance Quote

The reference model for Valem’s core use case: LLM-driven quoting / eligibility tools over structured data, where a team would otherwise hand-wire the same compute-and-validate layer per project. It exercises every pillar of the runtime on one realistic domain:

Pillar In this model
Constants An actuarial rate table (baseRatePer1000, smokerLoading, coverage bounds) bound as $const.
Chained derivations units → baseAnnual → annualPremium → monthlyPremium, plus age/smoker risk factors, recomputed reactively (change coverage, every downstream value updates).
Constraints coverage-positive (rollback invariant) and coverage-in-range (soft flag).
Meta-derivation Age-dependent maximum coverage overlaid on the field schema.
Effects (all three shell kinds) server fetches a live regional rate multiplier and folds it into the premium; timer expires a priced quote after 15 minutes; caller surfaces a “quote ready” command.
View A vertical quote form bound to the reactive fields.

Certification, not homework. The spec’s embedded tests array is a golden dataset whose expected premiums are hand-derived from the rate table — an independent oracle, not assertions the LLM wrote about its own formula. The suite runs through the full ModelService reactive pipeline in InsuranceQuoteGoldenTest, which is the concrete shape of the draft → review → certify → promote governance lifecycle: a model is not trusted until an independent golden suite passes.

Worked example (from the golden set): a non-smoker aged 35 with $100,000 coverage in a neutral region → units = 100, baseAnnual = 100 × 1.5 = 150, ageFactor = 1.0, smokerFactor = 1.0, regionMultiplier = 1.0 ⟹ $150/yr, $12.50/mo, decision quoted.

benefits-eligibility.json — Benefits Eligibility

A second model in the same family (eligibility over structured data): an assistance-program calculator whose federal-poverty-line threshold scales with household size (constants), whose income-to-poverty ratio drives an eligibility tier and monthly benefit (chained derivations), with a positive-household rollback invariant plus an over-max soft flag, an external income-verification server effect, a caller notification, and a submission-expiry timer. Its golden suite (BenefitsEligibilityGoldenTest) is hand-derived from the poverty table — e.g. a household of 4 earning $45,000 → poverty line 15000 + 3×5000 = 30000, ratio 1.5 ⟹ partial tier, $300/mo.

Together the two featured models split the teaching surface: insurance-quote demonstrates a meta-derivation and a rollback constraint; benefits-eligibility demonstrates a soft-flag constraint and household-driven threshold scaling.

File Demonstrates
car-loan-calculator.json Multi-level derivation chain; a full 60-row amortization schedule via array derivations.
order-items-price-total.json Array wildcard derivations ($.items[*].lineTotal); add/remove rows react.
customer-satisfaction-survey.json Conditional field visibility (relevant meta) + a rollback constraint.
personal-budget-tracker.json Multiple named views; income-vs-spending breakdown.
energy-consumption-heating.json A four-level mathematical derivation pipeline.
savings-growth.json A compound-growth projection ($map over a [0..years] range) rendered as a live line chart (dataChart) plus a year-by-year dataTable — drag the return slider and the curve reshapes instantly.
daily-wellness.json Conditionally read-only fields; per-field gating.
net-salary-estonia-2026.json A gross-to-net pay calculator under Estonia’s 2026 rules (22% income tax over a €700 basic exemption, unemployment + funded-pension deducted first); a clean single-input derivation chain with a keyValueList breakdown.
car-tax-estonia.json Estimates Estonia’s annual motor-vehicle tax and one-time registration tax from a car’s CO₂ and mass — threshold arithmetic (max(0, …)) feeding statTiles and a breakdown.
tic-tac-toe.json A playable game as a pure reactive model: the current player, the winner across all eight lines, and the draw state are derived from nine cell fields — the button grid mutates cells, and the status updates live with no game code.
support-ticket-triage.json All four effect executors together (llm classify, server SLA lookup, caller alert, timer auto-escalate).
world-clock.json A minimal, focused pair: only server (http) and timer effects composing into a self-refreshing poll — the http effect fetches the current date/time for a chosen country from a real public API (timeapi.io), and the timer re-arms every 10s by bumping a tick the http effect is keyed to, so the clock keeps updating itself.
team-offsite-planner.json The component catalog, end to end. Three views navigated by a stepper; a tabs container over currency/percent inputs, a dateRangeField writing two paths, a tagsField, a ratingField, and a collapsible revealed by a relevant meta-derivation. The summary view pairs statTiles with a summaryList and an alert driven by a flag constraint, and the diagnostics view puts explainPanel, auditTimeline and jsonViewer on the model itself.

Each spec ships its own tests; the console module runs them through the real pipeline in ConsoleExamplesIntegrationTest, and BundledExamplesTest in valem-core validates every file in the directory and runs its self-tests — so the gallery is also a regression suite. team-offsite-planner.json additionally has a Playwright suite, valem-e2e/tests/offsite-planner.spec.ts, driving each new component through the browser. world-clock.json additionally has its own end-to-end test, WorldClockExampleTest, which drives it through the real REST API — creating the model, evaluating its viewDefinition form via GET /models/{id}/view, picking a country as a form mutation would, and confirming both the initial time fetch and the recurring re-fetch driven by the timer (the http lookup re-firing each tick).