Replication package: What Should an Adaptive Tracer Optimise? Measuring Diagnostic Sufficiency Under a Byte Budget

Anonymised replication package for a paper under double-anonymous review. Contains the adaptive tracing controller, the trace replay simulator and its byte ledger, every analysis script, the trained models with their conformal references, and the per-run per-policy results from which every table and figure in the paper is regenerated.

The tables regenerate byte-identically from the included results with a single command; see README.md. Audit scripts are included that re-derive every quoted number from the data rather than comparing against a stored table.

The derived per-(window, container) measurements for all 304 recorded runs (~22 GB) are published as separate archives in this record. They are needed only to re-run the replay from scratch. The ~8 TB of raw kernel traces behind them are not redistributed; the derived measurements are the level at which replay operates, so what is published here is sufficient to reproduce every policy comparison in the paper.