loadbearer v0.6.1

• loadbearer compare now has a TUI. In an interactive terminal it opens a
  scrollable, colour-coded view of the head-to-head — winning cells and deltas
  in green, regressions in red, the reference column dimmed, plus the warnings,
  the SUSTAINED LOAD block and the verdict. --plain forces the old text
  table (still the default when stdout is not a terminal); --json is
  unchanged. ↑/↓/PgUp/PgDn/Home/End (or j/k/space/g/G)
  scroll; q / Esc / Enter exits.

Full Changelog: v0.6.0...v0.6.1

loadbearer v0.6.0

• New loadbearer score command. Re-score an existing result file against a
  different baseline, profile or curve without re-running the benchmarks — the
  file keeps every raw measurement, so the grade is just one view of it that can
  be recomputed in a second.
  • loadbearer score result.json --baseline our-fleet.toml scores against a
    baseline you built with loadbearer baseline; --profile and --curve-k
    override the values the file was scored with; --output writes a new
    result file; --json emits it to stdout.
  • A short banner shows what changed (baseline / profile / curve / tool
    version) and the old → new overall.
  • Subtests the baseline has no entry for are left out of the score with a
    note, rather than being a hard error — a fleet baseline that omits the
    OS-dependent network component is a normal thing to re-score against. The
    file's full raw is preserved in the output.
• Baseline::load reads a baseline TOML from a path (used by score --baseline).

Full Changelog: v0.5.1...v0.6.0

loadbearer v0.5.1

• The --soak phase now runs inside the TUI. On an interactive loadbearer run --soak, the sustained-load phase gets its own live screen — a progress
  gauge, current throughput, current clock, retained-vs-peak-so-far, and a
  throughput sparkline that grows a mark per sample — instead of running as a
  plain stderr line after the TUI closed. q skips the soak and keeps the
  graded result; a soak that produced at least a few samples is still embedded
  in the result file. The results screen gains a one-line SOAK summary.

Full Changelog: v0.5.0...v0.5.1

loadbearer v0.5.0

• New sustained-load / thermal soak test. The graded benchmarks are all
  short bursts and measure a machine near its boost clocks; the soak test holds
  every logical CPU under a blended integer + floating-point load for a fixed
  stretch (default 90 s), samples aggregate throughput and CPU frequency once a
  second, and reports the unthrottled peak, the steady-state rate, the
  percentage retained, the throttle onset time, and steady-window
  stability. It is the signal that tells two thin-and-lights with identical
  burst numbers apart. Not scored — measured and shown, like the
  --net-target link probe.
  • loadbearer soak [--duration SECS] [--threads N] [--output FILE] [--json]
    runs it on its own.
  • loadbearer run --soak [--soak-duration SECS] appends it to a full
    assessment; the result is embedded in the result JSON under soak.
  • loadbearer compare shows a SUSTAINED LOAD block — absolute steady
    throughput and retained-vs-own-peak — when every result file carries soak
    data.
• Refinement to memory/bw_read_mt (0.4.0): the per-thread buffers now start
  their timed read together at a barrier (removing staggered-start skew, most
  visible at --duration short) and the per-thread floor rose 16 MiB → 32 MiB
  (insurance against a large shared L3 on server parts). The metric's meaning is
  unchanged; the baseline anchor stays at 28 GiB/s.

Full Changelog: v0.4.0...v0.5.0

loadbearer v0.4.0

• Two new CPU subtests: AES-256-GCM and SHA-256 throughput. These pick up
  the CPU's AES-NI / CLMUL / SHA hardware where present (detected at runtime),
  so a chip that lacks crypto acceleration — common before ~2019 — now shows up
  in the grade. Adds the aes-gcm and sha2 crates.
• New memory subtest: Sequential read, all cores. The other bandwidth
  subtests are single-threaded; this one runs the read kernel on every logical
  CPU and sums the rates, capturing whether the memory subsystem scales past
  one or two cores (many dual-channel laptops don't).
• reference-v1 gains cpu/aes_gcm, cpu/sha256 and memory/bw_read_mt
  anchors (provisional, like the rest).
• Because these join the CPU and memory geometric means, scores shift for
  machines that lack crypto acceleration or don't scale memory bandwidth.

Full Changelog: v0.3.1...v0.4.0

loadbearer v0.3.1

• Single-threaded subtests are pinned to one core while they run, so the OS
  scheduler can't bounce the measurement between core types (P/E cores on Intel
  hybrid CPUs, big.LITTLE on ARM) part-way through. On a 13th-gen Intel laptop
  this cut single-threaded memory-bandwidth run-to-run variance from ±40–60%
  to a few percent. Each such iteration runs on a throwaway thread pinned to the
  fastest core (highest rated frequency on Linux; the first reported core
  elsewhere). All-core subtests are unaffected. Adds the core_affinity crate.

Full Changelog: v0.3.0...v0.3.1

loadbearer v0.3.0

• The network component is no longer folded into the overall grade. It is
  still a first-class component — measured, scored, shown, and used by
  compare — but OVERALL is now the geometric mean of CPU, memory and disk
  only. The loopback network figures depend heavily on the host OS and any
  security tooling (an EDR's packet inspection can add tens of microseconds per
  syscall), so folding them into a hardware grade produced misleading results
  — e.g. a fast Windows laptop grading F on network and dragging its overall
  down two letters. The result JSON gains a graded flag per component.
• compare now warns when result files are from different operating systems,
  since the network component in particular is not comparable across OSes.
• Scoring profiles no longer carry a network weight (it had no effect on the
  overall); server's description no longer claims to favour it.

Full Changelog: v0.2.1...v0.3.0

loadbearer v0.2.1

• loadbearer --version / -V now report a full build version: the crate
  version plus the git commit, build date, target triple and profile, e.g.
  loadbearer 0.2.1 (a1b2c3d4e 2026-08-28, x86_64-pc-windows-msvc, release).
  A build.rs captures this at compile time; it honours SOURCE_DATE_EPOCH
  for reproducible builds and falls back to unknown for the commit when built
  outside a git checkout.

Full Changelog: v0.2.0...v0.2.1

loadbearer v0.2.0

loadbearer 0.2.0 adds Network as a fourth first-class graded component, plus
an optional real-link test between two machines.

• Network is now a first-class graded component. Four loopback subtests over
  127.0.0.1 measuring the machine's network stack, not any physical link:
  single-stream and all-core TCP throughput, TCP request/response round-trip
  latency, and UDP small-packet send rate. No packets leave the machine; it
  needs no network access and no target.
• loadbearer run --net-target HOST:PORT optionally probes a real link (TCP
  upload, round-trip, UDP send rate) to a loadbearer net-server running on
  another machine. Reported in its own block and in the result JSON's link
  field — it measures the path, not either host, so it is deliberately not
  graded.
• loadbearer net-server [--bind ADDR] runs the server side of that link test.
• Scoring profiles gained a network weight: server now favours it (1.4),
  content-creation de-emphasises it (0.5).
• reference-v1 baseline gained [components.network] anchors.

Downloads below: a self-contained loadbearer.exe for Windows (x86-64) and a
Linux (x86-64) tarball. See the
wiki
for how the network component works.

Full Changelog: v0.1.0...v0.2.0

loadbearer v0.1.0

First release.

• loadbearer run benchmarks CPU (integer and floating-point throughput single-
  and all-core, BLAKE3 hashing, DEFLATE compression), memory (sequential
  read/write/copy bandwidth, random-access latency) and disk (sequential
  read/write, random 4K read/write IOPS at queue depth 1), scores each metric
  against an embedded reference baseline, and grades every component and the
  machine as a whole on an S-to-F scale.
• Interactive terminals get a TUI with live per-subtest progress, an overall
  gauge with ETA, and a scrollable graded results screen; q cancels a run
  cleanly. Piped output, --plain and --json use text / JSON instead.
• loadbearer compare produces a head-to-head verdict between two or more result
  files, computed from the raw metrics so it is independent of the baseline and
  curve each file was scored with. It warns on mismatched presets/baselines and
  skips metrics not present in every file.
• loadbearer info prints the machine inventory; loadbearer list prints the
  benchmarks, baseline and profiles; loadbearer baseline prints the built-in
  baseline or regenerates one from result files.
• run settings resolve command-line switch, then --config TOML file, then
  built-in default. Scoring profiles: general, dev-workstation,
  content-creation, server.
• Disk reads and random I/O use unbuffered I/O (O_DIRECT on Linux,
  FILE_FLAG_NO_BUFFERING on Windows) to bypass the page cache, with a buffered
  fallback and a RAM-disk guard.
• Result files are versioned (loadbearer.result/1) and keep the full unscored
  raw metrics alongside the scored output.

Full Changelog: https://github.com/issinoho/loadbearer/commits/v0.1.0