Beyond Ping
Matched-marginal multiplayer network benchmark

This bundle contains a deterministic sensitivity benchmark for three network
models:

1. Independent ordering of a fixed delay and loss multiset.
2. Temporally correlated ordering of the same multiset.
3. Temporally and cross-player correlated ordering of the same multiset.

Every player receives exactly the same count of one-, two-, three-, six- and
twelve-tick delays and packet losses under every model. The benchmark changes
only their ordering. It then runs the same scripted multiplayer input traces
through each model.

The game model is synthetic. It does not emulate a named engine, transport or
commercial game. The outputs are not estimates of Internet conditions, player
experience, hit-registration quality or competitive fairness.

Quick verification

1. Inspect BUNDLE-MANIFEST.json for the SHA-256 digest and size of every
   payload file.
2. Run the deterministic invariant checks:

   python run_network_benchmark.py --self-test

3. Recreate all CSV and SVG outputs:

   python run_network_benchmark.py

The standard run uses 1,000 episodes, eight players, 1,024 ticks per episode
and a 64 Hz server. It creates 3,000 model runs. No external Python packages
or network access are required.

Primary outputs

network-benchmark-runs.csv
    One row per episode and network model. Each row records the model version,
    Python version and canonical configuration SHA-256 digest.

network-benchmark-summary.csv
    Aggregate network diagnostics and gameplay-sensitive outcomes.

network-model-comparison.csv
    Absolute values and ratios for the main comparisons.

matched-delay-trace.csv
    The full first-episode delay trace for one player under all three models.

network-model-outcomes.svg
    An accessible chart generated from the comparison results.

matched-delay-trace.svg
    A selected window showing how the same marginal values can be reordered.

Model boundary

The movement-error proxy estimates displacement while the authoritative
server is applying a different movement command from the client. It resets
when both commands match. The synthetic shot test checks that proxy against a
declared hit radius two ticks after each direction change. Neither measure is
a replacement for engine telemetry or a player study.

Bundle reproduction

Run:

   python build_artifact_bundle.py

The builder uses fixed ZIP timestamps, permissions and file ordering. It
recreates beyond-ping-benchmark-bundle.zip from the files in this directory.

See LICENSES.txt and THIRD-PARTY-NOTICES.txt before reusing the contents.
