cs2df 3.0.0

Reference exporter & validator CLI for cs2-demo-format v3 (CS2 demo → ZIP data package)

cs2df

cs2df is the reference Python exporter and validator for the cs2-demo-format v3 ZIP contract.

It parses CS2 .dem files with demoparser2, writes strict v3 ZIP packages, and validates exported packages with schema plus package-level QA.

Setup

uv sync

The CLI entrypoint is cs2df.

Export

# Standard profile: required files + shots.json + replay.json
uv run cs2df export match.dem

# Research profile: also emit full-tick duels.json windows
uv run cs2df export match.dem --research

# Choose an output path
uv run cs2df export match.dem -o match.zip

The default ZIP compression level is 3, chosen from local benchmark results as a speed/size balance. Use a higher level when smaller ZIPs matter more:

uv run cs2df export match.dem --compress-level 6
uv run cs2df export match.dem --compress-level 9

Batch Export

uv run cs2df export-batch ./demos --workers 8 --descriptive

export-batch scans one directory non-recursively for .dem files and writes one ZIP per demo. It also writes report.json next to the outputs with:

• per-demo success/failure status
• output ZIP size
• source demo size
• compression level
• total duration
• aggregate throughput
• parse/package/write stage timings

Bad demos are reported as failed rows; a single parser failure does not crash the whole batch. Use --fail-fast when you want the batch to stop after the first failed demo.

Validate

uv run cs2df validate match.zip
uv run cs2df validate match.zip --strict

Validation checks JSON Schema and package-level invariants such as cross-file player indexes, round windows, column lengths, weapon dictionary indexes, and formal-round consistency.

Role in the Repository

This is a reference implementation, not the contract itself. The authoritative contract lives in ../schemas/index.ts and the generated JSON Schemas in the repository's spec/ directory. Any producer that emits a ZIP passing strict validation is conformant.

The exporter also serves as the performance baseline for the v3 format: per-frame data stays in pandas DataFrames until the columnar stream builders materialize compact integer arrays for JSON serialization.

Event-extraction logic was originally ported from cs2-demo-analysis-kit (and before that DrEAmSs59/CS2-insight-agent, with the author's permission).