Selective Frank.GameEngine + Frank.ML migration

name: GameEngine ML selective migration overview: Selectively extract library-ready code from Frank.GameEngine (math/geometry) and Frank.ML (neural foundation only) into Novolis scaffolds, while keeping both personal repos active—no archiving—and updating governance to reflect partial migration instead of reference-only/archive policy. todos:

• id: gov-policy

content: Update gameengine-reference-policy, frank-inventory, roadmap, frank-naming-and-structure; add wave-7/8 extraction briefs status: completed

• id: wave7-math-arrays

content: Migrate Frank.GameEngine SubPrimitives/Array2D → novolis-math Novolis.Math.Arrays + TUnit tests status: completed

• id: wave7-math-geometry

content: Migrate Polygon/TriangleMesh/IntPoint/IntRect/Rgba32 → Novolis.Math.Geometry + TUnit tests status: completed

• id: wave8-ml-neural

content: Clone Frank.ML; migrate Foundation Core + Neural.* → novolis-machinelearning (3 packages) + TUnit tests status: completed

• id: registry-ci

content: Wire slnx, .novolis/packages.json, registry entries, CI/release preview for new packages status: completed

• id: personal-readme

content: Add partial-migration README blocks to Frank.GameEngine and Frank.ML (no archive) status: completed

• id: wave7b-obj

content: "Optional follow-up: ObjHelper → novolis-raylib after geometry ships" status: completed isProject: false

Goals and constraints

• Migrate now: renderer-agnostic, packable libraries with tests (or tests that can be ported to TUnit).
• Do not archive: Frank.GameEngine and Frank.ML stay open on your personal account; use partial sunset README blocks (what moved, what remains), not archive banners.
• Follow existing playbook: extract/rebrand via migrate-frank-slice.ps1, no git history transfer, TUnit-only tests (naming.md).
• Respect locked graphics path: no Frank.GameEngine.Rendering.* while novolis-raylib is canonical (gameengine-reference-policy.md).
• Frank.ML scope (your choice): neural stack only in wave 1 — defer Frank.ML.Foundation.AutoMl (Microsoft.ML) to a later wave.

flowchart TB
  subgraph frankGE["Frank.GameEngine (stays active)"]
    Prim[Primitives subset]
    AssetsObj[ObjHelper only]
  end
  subgraph frankML["Frank.ML private (stays active)"]
    MLCore[Foundation.Core]
    MLAbs[Neural.Abstractions]
    MLImpl[Neural.Implementation]
    MLApps[Apps / Domain / AutoML]
  end
  subgraph novolis["Novolis targets"]
    Math[novolis-math]
    ML[novolis-machinelearning]
    Raylib[novolis-raylib later]
  end
  Prim --> Math
  AssetsObj --> Raylib
  MLCore --> ML
  MLAbs --> ML
  MLImpl --> ML
  MLApps -.->|not wave 1| frankML

Phase 0 — Governance and inventory (before code)

Update docs so execution matches “migrate ready parts, keep repos active”:

Personal README (not archive): prepend a short block per repo, e.g. “Core libraries are moving to Novolis; samples/apps remain here.” Link to Novolis repos/packages. Do not run gh repo archive on either repo.

Phase 1 — Wave 7: Frank.GameEngine → `novolis-math`

What is “ready now”

Source is already under bootstrap/scratch/frank-eval/Frank.GameEngine.

What to skip (not library-worthy / wrong lane now)

Target repo work

`novolis-math` is scaffold-only (placeholder Novolis.Math). Replace with:

src/
  Novolis.Math.Arrays/
  Novolis.Math.Geometry/
tests/
  Novolis.Math.Arrays.Tests/
  Novolis.Math.Geometry.Tests/
Novolis.Math.slnx

• Extend `.novolis/packages.json` with both packages.
• Run migrate-frank-slice.ps1 on each slice; add replacements: Frank.GameEngine.Primitives → Novolis.Math.*, namespaces accordingly.
• Port tests from xUnit → TUnit using Novolis.Testing.TUnit from novolis-testing.
• dotnet build / dotnet test; add registry stubs in novolis-registry/packages/.
• Optional aggregate meta-package Novolis.Math referencing Arrays + Geometry.

Overlap note: Frank.Collections also has Array2D with stronger JSON serialization tests. Wave 7 uses GameEngine’s copy to honor your request; schedule a follow-up dedupe issue to merge Collections JSON helpers into Novolis.Math.Arrays if desired.

Phase 2 — Wave 8: Frank.ML neural foundation → `novolis-machinelearning`

What is “ready now” (neural only)

Clone frankhaugen/Frank.ML (private; your gh auth already works) into bootstrap/scratch/frank-eval/Frank.ML.

Tests to port: Frank.ML.Foundation.Core.Tests, Frank.ML.Foundation.Neural.Tests → TUnit.

Explicitly out of wave 8

Target repo work

`novolis-machinelearning` is scaffold-only. Same pattern as math:

• Three packable projects + test projects + Novolis.MachineLearning.slnx.
• Strip Frank.ML-specific ThisAssembly.Project / RepoRoot constants unless needed; align Directory.Build.props with other novolis-* repos (net10.0, warnings as errors).
• Publishing: packages will be public on Novolis-Platform even though source repo stays private on personal GitHub — document that in README.
• Registry entries + 0.1.0-preview.1 release workflow when CI/trusted publishing is configured.

Phase 3 — Wave 7b (optional, soon after math)

Depends on wave 7 geometry types being stable.

Phase 4 — Personal repo README updates (no archive)

For each migrated slice, add to Frank repo README:

> **Partially on Novolis:** `Novolis.Math.Arrays` / `Novolis.Math.Geometry` live in [novolis-math](https://github.com/Novolis-Platform/novolis-math).
> This repo remains the home for samples, experiments, and unmigrated engine code.

Equivalent block for Frank.ML pointing at novolis-machinelearning neural packages. No “repository is archived” language.

Success criteria

• [ ] novolis-math: build + TUnit green; no Frank.* references in production code
• [ ] novolis-machinelearning: neural trio build + TUnit green; AutoML not included
• [ ] Governance docs and extraction briefs updated; inventory/roadmap reflect waves 7–8
• [ ] Frank.GameEngine and Frank.ML remain unarchived with partial-migration README
• [ ] Registry JSON for new package IDs (preview versions)
• [ ] Frank.GameEngine rendering/core/samples untouched; Frank.ML apps/domain untouched

Risk register

Suggested execution order

1. Governance + naming + briefs (phase 0)
2. Wave 7 novolis-math (arrays, then geometry, tests, CI)
3. Wave 8 novolis-machinelearning neural (clone Frank.ML, migrate, tests, CI)
4. Personal README partial banners
5. Wave 7b OBJ → raylib when geometry is stable