AllasCode TestsCreator

Automatic deterministic test generation for AllasCode function/Action contracts and behavior trajectories.

TestsCreator generates:

• one fully valid synthetic case;
• boundary cases from declared constraints;
• isolated invalid cases, changing exactly one parameter/property at a time;
• reproducible values from a seed;
• trajectory tests that traverse modeled workflow branches to terminal outcomes;
• edge-coverage reports that identify transitions not exercised from the declared start;
• ephemeral test execution metadata so payloads and business events are never persisted.

1. Contract-case generation — one valid synthetic case, boundary cases, and isolated invalid cases derived from declared constraints, using a reproducible seed.
2. Semantic suite compilation — the PureCore DSL and scenario config compile into deterministic canonical JSON for five distinct test types: unit, load, stress, chaos, and security.

The semantic compiler validates matching suite/agent/intent identities, required test kinds, recognized configuration keys, and bounded non-zero load/stress settings. DSL/config are data, never executable assertions. The runtime executor remains the authority for observed pass/fail outcomes.

The Entity Driven Testing parser also preserves zero-padded identifiers such as CPF values, rejects duplicate input names, and validates that each Run argument refers to a declared Input before lowering.

Canonical test types

Unit/contract tests

For a function such as CreateUser(name: String(min=1,max=80), age: Int(min=18,max=120)), the generator creates valid, boundary, and isolated invalid cases. Every negative case keeps all other arguments valid.

Trajectory coverage

The structural path generator is in src/trajectory.zig, with its skill contract in skills/trajectory-coverage/SKILL.md and details in docs/TRAJECTORY-COVERAGE.md.

Given a directed graph with a declared start and explicit terminal states, it enumerates reachable terminal trajectories, handles cycles with a bounded policy, and reports uncovered edges. This is structural model-based path generation; guard satisfiability and runtime execution still require adapters.

Validation-Driven Synthetic Data Generation

The validation feedback layer emits structured diagnostics and applies deterministic, constraint-directed repairs while preserving the original failing case. See the VDSDG design and src/validation_feedback.zig. The real validator remains the pass/fail oracle; repair is not proof of validity.

Native Smith semantic fuzzing