Post-alpha backlog
Rendered from docs/backlog-post-alpha.md at commit b9d69f7299ed. The repository file is authoritative.
Items deliberately left out of the public alpha (2026-10-12). None of these is claimed fixed or verified; each names the gate that would close it.
| item | status in the alpha | closing evidence |
|---|---|---|
macOS code-execution sandbox (sandbox-exec deny-by-default profile) |
disabled pending verification — code_exec tasks are skipped on macOS unless MODELSAAR_MACOS_SANDBOX=verify |
run_mac_tests.command → mac-tests-PASSED (sandbox tests executed under sandbox-exec, no skips); then re-enable by default |
| real-client walkthrough (Open WebUI, OpenClaw, OpenCode, Hermes) | NOT RUN; owner steps in site/VALIDATION.md |
one recorded workflow per client (modelsaar decisions showing the request) |
live evaluation under scorer 2026.10.12-3 |
NOT RUN; the quoted result keeps scorer 2026.10.11-2 |
a separately labelled GX10 run (run_live_verify.command) |
real-model embeddings; vLLM / llama.cpp backends; /tokenize |
NOT RUN (fixture-only) | live run with a declared embeddings: contract / a generic server |
| assumed residency for runtimes that cannot report it | conservative indefinite assumption unless assumed_residency_ttl_s is set |
a residency signal for those runtimes, or operator guidance per runtime |
| CI run at the published revision | definition only; no run exists until the branches are pushed | green run on the PR revision |
| evaluation with a reasoning-matched budget and all eight GX10 models | not run | separately labelled run |
| website copy refinement, SEO, launch optimisation | out of scope for the alpha | after real-user feedback (site/launch/feedback-log.md) |