docs: release notes for 0.1.3

docs/release-notes/0.1.3.md

@@ -0,0 +1,72 @@
+# ComputeArena CLI 0.1.3
+
+This release fixes benchmarking with BaseRT 0.2.5. On 0.1.2, every default
+run with the released BaseRT 0.2.5 ended, after the benchmark had completed,
+with `BaseRT returned incompatible headline measurement metadata`, and no
+report was saved. Nothing else changes.
+
+## Install or upgrade
+
+On macOS or Linux:
+
+```sh
+curl -LsSf https://computearena.ai/install.sh | sh
+```
+
+The same command upgrades an existing installation. Saved reports, the
+installation signing key, login sessions, model receipts, and installed
+runtimes are kept. Release archives and checksums are also available below for
+manual installation.
+
+BaseRT 0.2.5 itself needs no change. If BaseRT is not installed yet, or is
+older than 0.2.5, `computearena basert install` fetches the current release.
+
+## What changed
+
+- **Runs with BaseRT 0.2.5 complete and submit.** BaseRT 0.2.5 runs each
+  prefill size at its own context capacity and records the largest one the
+  sweep used as the run's context size: 16384 for the default sweep, which
+  reaches PP16384. ComputeArena 0.1.2 expected the headline's 4096 there and
+  refused the run once it had finished. The client now expects the value
+  BaseRT 0.2.5 records. The measurement is unchanged: PP512 and TG128 still
+  run first in a freshly loaded model with a 4K reservation, every workload's
+  capacity, order, and repetitions are still checked before signing, and the
+  reports still normalize to `computearena-throughput/3`. computearena.ai
+  applies the same rule.
+
+## Reports and compatibility
+
+0.1.2 could not save a BaseRT 0.2.5 report, so there is nothing to resubmit
+from it: run the benchmark again with 0.1.3. Reports saved by 0.1.0 to 0.1.2
+with BaseRT 0.2.4 or llama.cpp are unaffected and remain submittable when
+they contain the full default sweep. Benchmarks already published on
+computearena.ai are not changed.
+
+A BaseRT harness built from the development branch before the 0.2.5 release
+reports version 0.2.5 but records the headline's capacity instead. 0.1.3 does
+not sign its runs, and computearena.ai does not accept reports from it;
+replace it with `computearena basert install`. BaseRT 0.2.4 keeps working as
+before: its runs use the older protocol, the report records that, and the
+client says so before the run.
+
+## Supported systems
+
+Prebuilt ComputeArena binaries are available for:
+
+- macOS on Apple Silicon;
+- Linux x86_64 with glibc 2.31 or newer; and
+- Linux arm64 with glibc 2.31 or newer.
+
+Windows and Intel Macs are not supported in this release. ComputeArena does not
+bundle an inference runtime or model; install BaseRT or llama.cpp and obtain a
+compatible model before benchmarking. Prebuilt BaseRT runtimes exist for macOS
+on Apple Silicon and for Linux arm64 with CUDA; on Linux x86-64, use llama.cpp
+or a BaseRT benchmark harness you built yourself.
+
+The macOS binary is not signed with an Apple Developer ID. The release archive
+is checksum-verified and carries a Sigstore keyless signature from the release
+workflow. Browser-downloaded copies may need their quarantine attribute removed
+as described in the README.
+
+Accuracy benchmarks, MLX, vLLM, Windows, and Intel macOS support are not part
+of v0.1.3.