Turn versioned execution traces into diagnostics about where behavior changed and what a run resembles.
Ygg combines application events, Linux kernel telemetry, a versioned Arrow schema, representation learning, and change-point analysis. Dropped-event counts stay in the data instead of disappearing from the result.
| Study | Result |
|---|---|
| Single-thread event emission with collector draining | 80 ns p50, 108 ns p99, 0 dropped |
| Eight-thread event emission | 82 ns p50, 2.0 µs p99, 0 dropped |
| Synthetic contention regimes | +0.30 silhouette |
| Blind policy-switch localization | 0.06% error |
| Real Norn objective ablation | -0.06 to -0.09 silhouette across all five variants |
The overhead run is macOS arm64 and the representation studies use their committed trace sets. The negative Norn result is part of the evidence, not a line to hide below the fold.
 |
 |
Synthetic contention regimes separate under masked-only training. A blind policy switch was localized close to its injected point.
Real Norn backoff regimes did not separate from application events alone across five objective variants. That negative result points the next study toward scheduler, preemption, and migration signals from the Linux collection path.
The Rust collector compiles on macOS but does not collect kernel events there. Linux eBPF campaigns require a compatible kernel and privileges.
Build, capture, train, reproduce the studies, and inspect every limitation.
See GUIDE.md for build presets and dependencies.
Functional CI and performance evidence are separate. See GUIDE.md and LAB_RULES.md / EVIDENCE.md in stra-ta/.github for manifest provenance and the one-command suite (./scripts/verify.sh / ./scripts/confidence.sh or tools/verify.sh).
CI is functional only. Performance evidence requires a committed manifest with machine metadata (commit, compiler, kernel, CPU, arch, build type, seed, argv) and a link from the claim to that artifact. See stra-ta/.github for lab-wide caveats.