Getting Started

Install

1. Visit the Microsoft Store listing (Store ID: 9P3HT1PHBKQK).
2. Click Install. ScalarScope needs Windows 10 version 1809 (build 17763) or later, x64.

Or build it from this repo:

git clone https://github.com/mcp-tool-shop-org/scalarscope.git
cd scalarscope
cargo run --manifest-path rust/Cargo.toml

Your first comparison

1. ScalarScope opens on Welcome. Click Try the sample comparison to load two built-in runs, or Compare two runs to open your own.
2. On Compare, click Open path A and Open path B. Each takes a file of the kinds below. Folder A and Folder B take a run folder.
3. Read the headline: how B compares with A at p50, p90 and p99, each with an interval.
4. Click a tile for Why, then Show me to move to the evidence. Press 1 to 6 to switch views.
5. Save bundle keeps the review as a .scbundle, and Open bundle shows it again exactly as saved.

The repo’s samples/ folder has an inference pair, a geometry pair and a folder of runs for the Workbench. TESTING.md says which to open where. The review explains each part of Compare.

The tabs

Tab	Purpose
Welcome	The sample comparison, recent reviews and what the review gives
Compare	Two runs side by side: the headline, the tiles and the views
Workbench	Many runs that differ in a setting, weighed with a local model (Workbench)
History	A project’s measures across its logged reviews (History)
Guide	What each part means, with search
Settings	Theme, color vision, high contrast, text scale, recent files and the anomaly rule

F1 opens Guide, Ctrl+, opens Settings and Ctrl+H opens Welcome.

Files the review opens

A latency CSV needs a latency column. latency_ms, latency, and time_ms are the names it reads:

step,latency_ms
0,12.5
1,11.8

A benchmark JSON puts samples in results, iterations, or benchmarks. Each sample needs latency_ms or latency:

{
  "results": [
    { "latency_ms": 12.5 },
    { "latency_ms": 11.8 }
  ]
}

A Chrome trace uses traceEvents. A complete ProfilerStep is one inference. dur is microseconds:

{
  "traceEvents": [
    { "name": "ProfilerStep#1", "ph": "X", "dur": 12500 }
  ]
}

A training file is a backpropagate run_history.json. The review draws loss_history. final_loss stays its own number. It is not another sample on that curve:

[
  {
    "run_id": "example",
    "status": "completed",
    "loss_history": [1.2, 0.8],
    "final_loss": 0.8
  }
]

Settings

The Settings tab writes the 2.0 preferences.json keys, so a 2.0 user’s choices carry over. Every key it does not use is kept. It covers:

• theme (Follow Windows, Light or Dark);
• series colors for color-vision modes;
• high contrast and text scale;
• the recent-files limit and Clear recent files;
• the anomaly rule: robust deviations, or 2.0’s 3-sigma rule.

The review’s caption names the rule a review used. About gives the version, the privacy statement, and links to report an issue and to the source.

Accessibility in the Rust review: it applies color vision, high contrast and text scale. It does not offer screen reader descriptions or a larger pointer; 2.0’s settings for those are kept in the file but not applied. If the preferences file cannot be read, ScalarScope leaves that file unchanged and Settings says so.

Running tests

cd rust
cargo test
cd ..

# The 2.0 oracles and the version surfaces
dotnet test tests/ScalarScope.FixtureTests

The .NET app in src/ScalarScope is 2.0. It stays in the repo as the reference that 3.x is checked against.