Skip to content

CONTRIBUTING.md

Latest commit

 

History

History
41 lines (30 loc) · 1.32 KB

CONTRIBUTING.md

File metadata and controls

41 lines (30 loc) · 1.32 KB
Raw

Contributing

codag-drain is intentionally narrow: real-time, deterministic log compression for agents. It turns raw log streams into compact, template-grouped artifacts and nothing more. Whatever interprets those artifacts lives elsewhere.

Development Checks

Run these before opening a PR:

CARGO_TARGET_DIR=/private/tmp/codag-drain-target cargo fmt --check
CARGO_TARGET_DIR=/private/tmp/codag-drain-target cargo clippy --all-targets --all-features -- -D warnings
CARGO_TARGET_DIR=/private/tmp/codag-drain-target cargo test

Public parser benchmarks require LogHub-2.0 structured CSVs:

LOGHUB_DIR=/path/to/loghub2 \
CARGO_TARGET_DIR=/private/tmp/codag-drain-target \
scripts/public_benchmarks.sh

Private downstream agent-serving evals belong outside this repo unless the data and judging setup are also made reproducible.

Benchmark Claims

Do not claim a win from compression alone. Any benchmark claim should say:

  • what dataset was used;
  • whether the data is public or private;
  • the sample size and sampling unit;
  • the metric and confidence interval;
  • the baseline, especially raw logs and Drain3;
  • the regime where the claim holds.

The current honest claim is scoped to large/noisy log windows under a fixed artifact budget. Small-window raw logs can outperform templated artifacts on diagnosis quality.