Beginners

New to ScalarScope? This page walks you through the core concepts and workflows, starting from zero.

What does this tool do?

ScalarScope compares two ML inference runs side by side and tells you exactly what changed. Instead of manually scrolling through logs looking for differences, ScalarScope computes five canonical delta types that fire only when the difference is statistically meaningful. You get a clear answer: “these things changed, and here’s the evidence.”

The tool is built for TensorFlow-TRT inference workloads but the underlying delta analysis works on any pair of time-series traces. Results are exportable as cryptographically verified bundles that anyone on your team can open and see identical findings.

Installation

Microsoft Store (recommended)

1. Open the Microsoft Store listing
2. Click Install
3. Requires Windows 10 (build 17763) or later

From source

# Install .NET 9.0 SDK, then:
dotnet workload install maui-windows

git clone https://github.com/mcp-tool-shop-org/ScalarScope-Desktop.git
cd ScalarScope-Desktop
dotnet restore
dotnet build
dotnet run --project src/ScalarScope

Your first comparison

Follow these steps to compare two inference traces:

1. Launch ScalarScope — the Home tab shows your workspace status
2. Click “Compare Two Runs” — this opens the Compare tab
3. Load the baseline trace — select the TFRT trace from before your optimization
4. Load the optimized trace — select the TFRT trace from after your optimization
5. Read the deltas — the Delta Zone shows which differences are meaningful, with confidence scores and explanations
6. Click “Show Me” on any delta to jump to the relevant view and see the evidence highlighted
7. Export a bundle — click Export Bundle to create a .scbundle file you can share with your team

If you do not have your own traces yet, click Try Example on the Home tab. This loads a built-in demo comparison so you can explore the full interface.

Key concepts

Deltas

A delta represents a statistically meaningful difference between two runs. ScalarScope computes five delta types:

• ΔTc (Convergence Time) — did the runs stabilize at different points?
• ΔO (Output Variability) — is one run more unstable/oscillatory than the other?
• ΔF (Failure Rate) — did one run experience failures the other did not?
• ΔĀ (Average Latency) — do the runs have meaningfully different average values?
• ΔTd (Total Duration) — did structural emergence happen at different times?

Each delta includes a confidence score (0 to 1) and a human-readable explanation. Deltas that do not reach statistical significance are automatically suppressed — you never see noise.

Runtime presets

The TFRT preset is designed for TensorFlow-TRT inference workloads. It maps inference signals (latency, throughput, memory, CPU/GPU load) and suppresses training-only deltas (ΔĀ and ΔTd) that are irrelevant for inference comparison.

Bundles

A .scbundle is a self-contained archive of your comparison results. Every file inside is hashed with SHA-256, and a bundle-level hash detects tampering. When a colleague opens your bundle, Review Mode activates automatically — they see your exact results without recomputing anything.

Alignment

When two runs have different lengths, ScalarScope aligns them before comparison. Three alignment modes are available: step-based (default for same-length runs), convergence-onset (aligns when signals stabilize), and first-instability (aligns at the first oscillation event).

Keyboard shortcuts

Shortcut	Action
Space	Play / Pause playback
Left / Right	Step backward / forward (1%)
Shift+Left / Shift+Right	Fine step (0.1%)
Home / End	Jump to start / end
Up / +	Increase playback speed
Down / -	Decrease playback speed
0	Reset speed to 1x
Ctrl+S or Ctrl+E	Quick export (PNG saved to Documents/ScalarScope Exports)
1–4	Switch tabs (Home, Compare, Guide, Settings)
?	Open Guide tab

Configuration

Open the Settings tab (or press 4) to adjust:

• Theme — System, Light, or Dark
• Playback — default speed, auto-play on load
• Export — default output resolution and export folder
• Accessibility — high contrast mode, color vision simulation (deuteranopia, protanopia, tritanopia, monochrome), screen reader support, adjustable text scale (75% to 200%), large pointer mode
• Session — auto-load last session, recent files limit (5, 10, or 20)

All settings are stored locally in your app data directory and persist between sessions. Click Reset All Settings to return everything to defaults.

FAQ

Q: Do I need a GPU to run ScalarScope? No. ScalarScope is a visualization and analysis tool. It reads inference traces and computes deltas on the CPU. No GPU is required.

Q: What file formats does ScalarScope accept? ScalarScope uses the RunTrace format (JSON-based, schema version 1.0.0). The TFRT connector reads TensorFlow-TRT log directories. You can also open .scbundle review bundles or load the built-in demo.

Q: Does ScalarScope send any data externally? No. ScalarScope has zero telemetry and zero analytics. All data stays local unless you explicitly export a bundle and share it yourself. See the Privacy Policy for details.

Q: Can I use VortexKit without ScalarScope? Yes. VortexKit is published as a standalone NuGet package. Run dotnet add package VortexKit to add it to any .NET MAUI app. See the VortexKit page for API details.

Q: What happens if a bundle has been tampered with? Review Mode verifies every file against its embedded SHA-256 hash. If any file has been modified, the integrity check fails and a warning is displayed. The bundle contents are still shown, but the verification status is flagged.

Q: How do I report a bug? Open an issue at github.com/mcp-tool-shop-org/ScalarScope-Desktop/issues.