.NET MAUI · Windows

ML inference runs, compared with rigor.

Stop eyeballing logs. ScalarScope loads two TFRT traces side by side, fires canonical delta analysis only when differences are statistically meaningful, and exports cryptographically verified bundles your team can reproduce.

Get from Microsoft Store Read the Handbook

Install

# Microsoft Store (ID: 9P3HT1PHBKQK) # https://apps.microsoft.com/detail/9P3HT1PHBKQK # Or use VortexKit in your own .NET app: dotnet add package VortexKit

VortexKit

using VortexKit.Core; // Shared timeline across all canvases var player = new PlaybackController { Duration = 10.0, Loop = true }; player.TimeChanged += () => { trajectoryCanvas.CurrentTime = player.Time; eigenCanvas.CurrentTime = player.Time; scalarsCanvas.CurrentTime = player.Time; };

Export bundle

var exporter = new ExportService(); await exporter.ExportComparisonAsync( baseline, optimized, outputPath: "results/run-42.scbundle" ); // SHA-256 verified, frozen deltas, // full provenance metadata inside

Replace eyeballing with structure

Every comparison is reproducible. Every delta is earned.

Apples-to-apples comparison

Load two TFRT inference traces side by side. The TFRT preset auto-suppresses irrelevant metrics so your analysis stays focused on what actually changed between runs.

Canonical delta analysis

Five delta types — ΔTc, ΔO, ΔF, ΔĀ, ΔTd — fire only when differences are statistically meaningful. No noise, no false signals, no manual threshold tuning.

Reproducible .scbundle exports

Export cryptographically signed bundles with SHA-256 integrity, frozen deltas, and full provenance metadata. Open in Review mode and results are verified, not re-derived.

Five canonical delta types

Each delta fires only when the difference clears its statistical threshold.

Delta

Measures

ΔTc

Compute time — wall-clock latency difference between runs

ΔO

Output divergence — numerical difference in model outputs

ΔF

Feature activation — changes in intermediate layer activations

ΔĀ

Average metric shift — mean performance across the evaluation set

ΔTd

Data throughput — tokens or samples processed per second

Quick start

Install from Microsoft Store

# Store ID: 9P3HT1PHBKQK
# https://apps.microsoft.com/detail/9P3HT1PHBKQK

1. Click "Compare Two Runs"
2. Load baseline TFRT trace (before)
3. Load optimized TFRT trace (after)
4. Review deltas in the Compare tab
5. Export .scbundle for reproducible sharing

Add VortexKit to your app

dotnet add package VortexKit

// Reusable visualization framework:
// - Time-synced playback across canvases
// - Animated SkiaSharp rendering
// - Comparison views + annotation overlays
// - SVG/PNG export
// - Semantic color system

Custom canvas with VortexKit

public class MyTrajectoryCanvas : AnimatedCanvas
{
    protected override void OnRender(
        SKCanvas canvas,
        SKImageInfo info,
        double time)
    {
        // SkiaSharp rendering at current timeline position
        // Automatically synchronized with other canvases
        // via shared PlaybackController
    }
}

Build from source

git clone https://github.com/mcp-tool-shop-org/ScalarScope-Desktop
cd ScalarScope-Desktop

# Requires .NET 9 + MAUI workload
dotnet workload install maui-windows
dotnet build ScalarScope.sln

# Run tests
dotnet test

Designed for serious ML workflows

Scientific rigor without the spreadsheet.

Review mode

Open any .scbundle and results are cryptographically verified against the embedded SHA-256 hash — not re-derived from raw data. Share bundles and know both parties see identical results.

Privacy first

Zero telemetry, zero analytics, zero network calls. All comparison data stays on your machine unless you explicitly export a bundle. No accounts, no subscriptions.

VortexKit for your own tools

The visualization framework powering ScalarScope is a standalone NuGet package. Build time-synced animated canvases, comparison views, and export pipelines into any .NET MAUI app.