Previously recorded measurements
These are historical figures carried over from the earlier documentation, not rerun or independently validated during the October 11, 2026 site refresh. Values are preserved, including approximate notation. Missing fixture hashes, exact versions, full logs, and cache conditions limit reproducibility.
Earlier performance table
The old page attributed this table to an M1 MacBook Pro with 16 GB RAM and Rayon parallelization:
| File | Size | Objects | Edges | Phase 1 | Phase 2 + dominators | Total |
|---|---|---|---|---|---|---|
| Small service | 26 MB | 400K | 1.2M | < 0.2s | ~0.3s | ~0.5s |
| Medium app | 212 MB | 3.1M | 12M | ~0.3s | ~0.8s | ~1.1s |
| Large app | 1.5 GB | ~12M | ~50M | ~0.4s | ~0.5s | ~0.9s |
| Large monolith | 2 GB | ~16M | ~65M | ~0.5s | ~0.7s | ~1.2s |
| Very large | 14 GB | ~100M | ~400M | ~2s | ~8.5s | ~10.5s |
The previous Rust guide repeated 1.5 GB/~0.9s, 2 GB/~1.2s, and 14 GB/~10.5s. Its memory tables claimed ~2–3× file size during graph building, ~3–4× peak, and ~1–2× cached usage elsewhere. Those unsupported universal ratios and the 14 GB/16 GB machine attribution require original logs, not guessed replacement measurements.
The old homepage also claimed a 1 GB/~60-second M1 result. Differences cannot be resolved without version/workload/phase provenance. These numbers are not an installation requirement or prediction for another dump.
How to use historical evidence
Preserve the original source record, identify uncertainty, and do not describe a documentation migration as a new validation. Independently recorded JVM/reference-analyzer sizing checks remain distinct from HeapLens-generated results. No submitted/published paper or historical benchmark artifact was modified by this refresh.
For actionable resource guidance use performance. A future reproducible measurement set should record fixture identity, source commit/backend, JVM layout, machine/OS, phase scope, cold/warm cache, trials, RSS, and failures.