Novolis.Physics — Design Review Findings

Review date: 2026-05-16 Version reviewed: 0.1.0-alpha (net10.0) — remediation shipped in 0.2.0-alpha Scope: Full review per library design review plan (API stability, architecture, consumer ergonomics, correctness, performance spot-check)

Test run: dotnet run --project tests/Novolis.Physics.Unit -c Release — 46/46 passed (967 ms)

Executive summary

Novolis.Physics delivers a coherent force-first core (IForceModel → SimulationPipeline → IIntegrator) with well-factored packages and strong scenario tests. Before stable release, address orphan public API and integration-path documentation. KspLite was removed (2026-05-16): it was an example-only DI shim, not a product package; registration patterns live in examples/dependency-injection.md. No physics blockers were found in the test suite.

Phase 1 — Public API inventory

Package dependency graph

Numerics
  └── Abstractions
        ├── Motion, Gravity, Aerodynamics, Collision.Simple
        ├── Ballistics (+ Collision.Simple)
        └── Orbits (Numerics only)
              Novolis.Physics (meta) → all product packages

Public type classification (49 source files, ~52 public types)

Former `Novolis.Physics.KspLite` package removed — see [examples/dependency-injection.md](./examples/dependency-injection.html).

Internal (non-public): TriangleRay in Collision.Simple.

Orphan and unused-contract flags

Phase 2 — Architecture coherence

Force-first alignment by package

Integration paths (validated)

Four distinct simulation styles coexist:

1. `SimulationPipeline` + `IForceModel` + `IIntegrator` — rigid body and projectile (with adapters).
2. `ProjectileBallisticSimulation.Step` — convenience facade; parity-tested against pipeline (ProjectileDragPipelineParityTests).
3. `BvhStaticSphereIntegrator` — sphere vs static mesh with contact resolution; used in room/billiards tests; not wired through IContactResolver.
4. `CentralOrbitSimulator` / `LeapfrogCentralBodySoA` — central-body leapfrog; scalar/vectorized kernels.

Architecture decision records (recommended)

ADR-1: Orbits remains a separate stack (ACCEPT)

Decision: Keep Novolis.Physics.Orbits as a leapfrog SoA integrator, not folded into IForceModel for v1.

Rationale: Different numerical method (symplectic leapfrog vs semi-implicit Euler), SoA layout for N bodies, and existing energy/angular-momentum tests. PointMassGravityModel already covers pipeline-style gravity for games.

Follow-up: Document when to use Orbits vs Gravity+Motion; consider moving OrbitalTestConstants/OrbitalTestState to test assembly or renaming to OrbitalReferenceOrbit before stable.

ADR-2: Collision stays query-only in v1 (ACCEPT)

Decision: IStaticWorld provides ray/sweep queries only; no full rigid-body contact solver in the pipeline.

Rationale: Matches scope; BvhStaticSphereIntegrator covers sphere-in-room scenarios. XML docs already flag approximate sweeps.

Follow-up: Add consumer doc section on sweep limitations and when BvhStaticSphereIntegrator vs manual pipeline stepping applies.

ADR-3: Ballistics — pipeline is canonical; facade is supported (ACCEPT)

Decision: Recommend SimulationPipeline + ProjectileQuadraticDragModel + ProjectileSemiImplicitIntegrator for extensibility; keep ProjectileBallisticSimulation as ergonomic entry point.

Rationale: ProjectileDragPipelineParityTests proves equivalence for gravity + quadratic drag. Facade reduces boilerplate for cannon-style problems.

Follow-up: README example showing both patterns and when to add custom IForceModel instances.

ADR-4: `IContactResolver` — remove or defer before stable (ACCEPT removal)

Decision: Remove from public API in next breaking window, or move to Novolis.Physics.Abstractions.Experimental if contact pipeline is planned.

Rationale: Misleading contract; actual API is SphereContactKinematics.ReflectWithRestitution inside BvhStaticSphereIntegrator.

Phase 3 — Consumer ergonomics

Persona walkthroughs

Minimal (point mass + Euler)

Path: MinimalSimulationExampleTests — direct construction of FixedStepAccumulator, SimulationPipeline, and PointMassGravityModel.

Friction: Low; see README quick start and INTEGRATION.md.

Game room (gravity + collision + bounce)

Path: BasketballEarthRoomCollisionTests, BouncingBallCollisionTests — build BvhStaticWorld from mesh, loop BvhStaticSphereIntegrator.AdvanceOneStep or AdvanceWithUniformAccelerationAndLinearDrag with UniformAccelerationEnergy for traces.

Friction: Collision is outside SimulationPipeline; gravity/drag applied inside integrator helper, not as IForceModel. Two mental models required.

Ballistics (drag + ground impact)

Path: Most tests use ProjectileBallisticSimulation; parity tests prove pipeline equivalence. BallisticsQueries wraps IStaticWorld sweeps.

Friction: Low for ballistics-only users; unclear which path to choose without reading ProjectileDragPipelineParityTests.

Ergonomics rubric (1 = poor, 5 = excellent)

Documentation gaps

• TestSupport README — references StarConflictsRevolt paths and project names.
• No worked example of sweep failure modes (fast sphere, thin geometry, capsule endpoint sampling).

Phase 4 — Correctness and performance

Test → invariant mapping

Performance spot-check (hot `Step` paths)

Conclusion: Sim-loop paths are allocation-conscious. Avoid calling CentralOrbitSimulator.SimulateFor inside per-frame hot paths without pooling/reuse.

Findings register

Blockers: None identified in this review.

Prioritized post-review backlog

P0 — Before stable (0.2.0) — **Complete**

1. ~~DR-001~~ — IContactResolver removed.
2. ~~DR-005~~ — Orbit fixtures moved to TestSupport.

P1 — Early stable — **Complete**

1. ~~DR-006~~ — Sweep limitations doc + test.
2. ~~DR-010~~ — Ballistics path documentation.
3. ~~Semver policy~~ — VERSIONING.md.

P2 — Nice to have — **Complete**

1. ~~DR-007~~ — timeSeconds documented.
2. ~~DR-013~~ — Orbit simulator reuse API.
3. ~~DR-009~~ — TestSupport README cleanup.
4. ~~DR-011, DR-014, DR-015~~ — Documented in 0.2.0-alpha.

Appendix A — Verified pre-seeded hypotheses

Appendix B — Suggested README example (for DR-008)

using Novolis.Physics.Abstractions;
using Novolis.Physics.Gravity;
using Novolis.Physics.Motion;
using Novolis.Physics.Numerics;

var integrator = new SemiImplicitEulerRigidBodyIntegrator();
var gravity = new PointMassGravityModel();
var pipeline = new SimulationPipeline<RigidBodyState, PointMassField>(integrator, gravity);

var field = new PointMassField([(Vector3d.Zero, 3.986e14)]);
var body = new RigidBodyState(
    new Vector3d(6_771_000, 0, 0),
    new Vector3d(0, 7_500, 0),
    Quaterniond.Identity, Vector3d.Zero, mass: 1.0,
    inertiaDiagonalBody: new Vector3d(1, 1, 1));

body = pipeline.Step(body, field, dtSeconds: 1.0);

End of design review findings.