novolis-governance / imports-todo/frank-ml-remainder.md

Import: `Frank.ML` remainder (post-foundation)

Policies that keep the org coherent

dotnetgovernancenovolis

Source: D:\frankrepos\Frank.ML

What is already in Novolis

Frank	Novolis
`Frank.ML.Foundation.Neural.*`	`novolis-machinelearning`
`Frank.ML.Foundation.AutoMl`	`novolis-machinelearning`
`Frank.ML.Foundation.Core`	Partial / evaluate
`Frank.ML.Domain.Racing` (simulation)	`novolis-simulation` (`Novolis.Simulation.Racing`)

What remains on disk

Project	Suggestion
`Frank.ML.Domain.Legacy`	Defer — uses `Frank.Mapping`, `Frank.Libraries.Csv/Json`; mine algorithms only
`Frank.ML.Domain.MlPipeline`	Partial — pipeline abstractions → `novolis-machinelearning` if generic
`Frank.ML.Domain.MlAssets`	App/private — asset bundles stay out of platform
`Frank.ML.Presentation.*` (Avalonia, ConsoleSpectre, Adapter)	Apps — `novolis-dogfooding` or private; not platform
`Frank.ML.App.*`, Aspire host	Skip platform — product apps
Tests / Playwright	Port patterns to `Novolis.Testing` only

Why

wave-8-machinelearning-neural.md scoped building blocks, not domain apps.
library-boundaries.md: NN evolution on racing lives in apps (e.g. NeuralRacing), not Novolis.MachineLearning.*.
Legacy domain would pull Mapping and Libraries monolith — import Mapping first (frank-mapping.md).

How

Finish frank-mapping.md.
Audit Domain.MlPipeline for types that are truly framework vs product.
Port only packable, domain-agnostic libraries; leave Avalonia/Web/CLI in Frank.ML or dogfooding.
Add partial-migration block to Frank.ML README (per frank-partial-migration.md).

Acceptance

No Frank.Libraries.* PackageReference in novolis-machinelearning production code.
Racing sim stays in Simulation; ML training glue stays in apps.