initial commit

_bmad-output/implementation-artifacts/13-5-test-suite.md

@@ -0,0 +1,44 @@
+---
+epic: 13
+story: 13.5
+title: "Test Suite"
+status: draft
+---
+
+## Epic 13 — Performance & Reliability
+**Goal:** Instant evaluation, never lose work.
+
+### Story 13.5: Test Suite
+
+As a CalcPad developer,
+I want a comprehensive test suite with > 95% code coverage on calcpad-engine,
+So that regressions are caught early and every function, conversion, and multi-line interaction is verified.
+
+**Acceptance Criteria:**
+
+**Given** the calcpad-engine crate
+**When** the full test suite is executed via `cargo test`
+**Then** code coverage as measured by `cargo-tarpaulin` or equivalent exceeds 95% of lines in the engine crate
+**And** all tests pass with zero failures
+
+**Given** every public function and unit conversion in calcpad-engine
+**When** unit tests are written
+**Then** each function has at least one test for the happy path
+**And** each function has at least one test for error/edge cases (division by zero, overflow, unknown unit, etc.)
+**And** every supported unit conversion pair has a dedicated round-trip accuracy test
+
+**Given** multi-line sheets with variables, dependencies, and mixed expressions
+**When** integration tests are executed
+**Then** end-to-end evaluation of representative sheets produces correct results
+**And** integration tests cover: variable assignment and reference, chained dependencies, circular reference detection, unit conversion across lines, and currency conversion
+
+**Given** the engine's parsing and evaluation logic
+**When** property-based tests are run via `proptest`
+**Then** random valid expressions always produce a result without panicking
+**And** random malformed expressions produce a graceful error without panicking
+**And** evaluation of `a + b` always equals evaluation of `b + a` for numeric values
+
+**Given** the benchmark suite defined with `criterion`
+**When** benchmarks are executed
+**Then** benchmarks exist for: single-line evaluation, 100-line sheet evaluation, 500-line sheet evaluation, DAG rebuild, and unit conversion lookup
+**And** benchmark results are recorded so regressions can be detected in CI