Architecture

Separate the storage engine from the presentation layer.

The long-term direction is a compact local data layer, immutable/snapshot-style reads, incremental updates and a Native AOT desktop shell.

Pipeline

MFT → normalized IDs → local snapshot → query planner.

Each boundary exists for a reason: filesystem parsing stays low-level, the data layer owns persistence, and the UI consumes stable projections.

No SQL types leaking into WinForms
Set-based snapshot loading
Immutable read path during reload
Provider-specific details contained in Data

Layers

Four responsibilities, four clean boundaries.

I/O

Filesystem ingest

MFT, USN and file metadata are parsed without UI assumptions.

Persistence

Compact identities, parent chains, volume metadata and reusable hashes.

Query engine

Predicates, ordering, projections, grouping and virtual-folder semantics.

Native shell

Explorer-like navigation with dense grids and storage-specific actions.

Offline by design

A disk does not disappear just because it is unplugged.

Volume presence is a state, not the identity of the catalog. Search metadata remains available while content-dependent actions clearly report that the source is offline.

Data contracts

Example query flow.

scan volume:disk21 --mode=mft --incremental
index --fields=name,parent,size,mtime,attributes
query "ext:pdf size:>100MB path:*forensic*" --sort=size:desc
dedup --scope=query --hash=blake3 --verify=exact

DB row navigation test:

Home row

Docs row