Content System

Every piece of code you type in Linux Dev Typer flows through a single content pipeline. Whether it ships with the app, comes from a community pack, or is pasted in by hand, all content becomes a Snippet with consistent structure and metrics-based difficulty.

Built-in calibration snippets

The app ships with 168 calibration snippets across 5 languages, covering difficulty bands D1 through D7. These are engine ground truth: their difficulty ratings are derived from structural metrics (symbol density, nesting depth, line count) rather than subjective judgment.

Calibration snippets load from assets/calibration/ at startup. They cannot be modified or disabled by the user.

User snippet packs

Drop a JSON file into ~/.config/linux-dev-typer/packs/ and the app picks it up automatically.

Pack format

{
  "language": "python",
  "snippets": [
    {
      "id": "my_list_comp",
      "title": "List comprehension",
      "difficulty": 3,
      "code": "[x**2 for x in range(10)]\n"
    }
  ]
}

Each snippet requires an id, title, difficulty (1-7), and code. Optional fields:

Field	Purpose
topics	String array of topic tags for filtering.
explain	Factual explanation lines shown on the completion card.
notes	Community tips, alternatives, and perspectives. Anonymous by design.
scaffold	Progressive learning context. Index 0 is a shallow hint; deeper entries add context.
variants	Alternative implementations of the same logic. No ranking or preference signal.
communityDifficulty	Display-only difficulty observation. Never used for selection or scoring.

Packs are validated by PackValidator at load time. Invalid packs are skipped with a warning.

Content-addressed IDs

Every snippet gets a content-addressed ID derived from SHA-256 of its code. This means the same code block is never added twice, regardless of how it enters the system. If you paste code that already exists in a calibration pack, the engine recognizes it and merges metadata rather than creating a duplicate.

Canonical pipeline

All content enters the engine as a CodeItem and flows through the same pipeline:

1. Ingestion — from calibration packs, user packs, paste, file import, or folder import.
2. Deduplication — SHA-256 content addressing prevents duplicates.
3. Difficulty derivation — the DifficultyDeriver assigns D1-D7 based on structural metrics when no difficulty is specified.
4. Normalization — consistent line endings, trimming, language detection.
5. Registration — the snippet enters the pool and becomes available for selection.

Practice profiles

Practice profiles are named parameter sets that tune every engine constant. The Default profile matches hardcoded baseline values. Custom profiles let you adjust:

• XP tuning — base multiplier, repeat decay, sloppy threshold and penalty, completion bonus.
• Rating tuning — K-factor, difficulty base and scale.
• Difficulty tuning — comfort accuracy threshold, minimum sessions, yo-yo window and swing.
• Trend tuning — WPM and accuracy thresholds, minimum sessions.
• Fatigue tuning — break session count, accuracy drop threshold.

All parameters are clamped to safe ranges at load time. The profile comparison (Diff) shows exactly what changed from the default.

Portable bundles (.ldtpack)

The .ldtpack format is a JSON envelope for sharing content between users.

A bundle contains:

• Snippet packs keyed by language name.
• Practice profiles (excluding “Default”).
• Format version (currently v3, supporting scaffold and variants).

Bundles never contain state or history. Import is additive: existing packs and profiles are never overwritten. The MergeInto method returns counts of what was actually imported.

Import and export paths

• Sidebar > Import — load a .ldtpack file.
• Sidebar > Export — save your user packs and custom profiles as a bundle.
• Paste Code — paste from clipboard with auto-detected language.
• Import File — import a single source file.
• Import Folder — import all source files from a directory.

Scaffolds, variants, and community notes

These three optional fields on snippets support a progressive learning experience:

• Scaffold — layered hints. The first entry is shallow (“This uses a list comprehension”), deeper entries add historical or conceptual context. Display-only; never affects engines.
• Variants — alternative implementations showing different valid approaches to the same logic. All variants are structural equals with no ranking.
• Notes — community tips and perspectives. Anonymous by design: no author, no source, no metadata. Imported notes are indistinguishable from local notes.

All three fields are purely informational. They appear on the completion card but never influence selection, difficulty, scoring, or rating.