feat: rapid-mlx doctor — 4-tier regression harness#125

Merged

raullenchai merged 28 commits intomainfrom

feat/doctor-harness

Apr 16, 2026

Commits on Apr 16, 2026

chore: clean up ruff lint debt (43 errors → 0)

Your Name
and
claude
committed
feat(doctor): scaffold + smoke tier — regression harness foundation

Your Name
and
claude
committed
fix(doctor): codex review fixes — imports, source-checkout guard, unique run dirs

Your Name
and
claude
committed
fix(doctor): atomically reserve run dir to survive concurrent invocations

Your Name
and
claude
committed
feat(doctor): check tier — server + API + perf + baseline diff

Your Name
and
claude
committed
fix(doctor): codex review — regression_suite parsing, per-model baselines, threshold names

Your Name
and
claude
committed
chore(doctor): record initial Qwen3.5-4B baseline for check tier

Your Name
and
claude
committed
fix(doctor): codex round 2 — tc_latency_ms sign + injective slug

Your Name
and
claude
committed
fix(doctor): codex round 3 — legacy slug fallback for baselines

Your Name
and
claude
committed
fix(doctor): codex round 4 — validate legacy baseline model identity

Your Name
and
claude
committed
feat(doctor): full tier — multi-model loop + agent profile sweep

Your Name
and
claude
committed
fix(doctor): codex round 1 — per-model diff + check name disambiguation

Your Name
and
claude
committed
fix(doctor): codex round 2 — restore diff.md on no-baseline + injective slug

Your Name
and
claude
committed
feat(doctor): regression-detection tests + harness/README + Makefile

Your Name
and
claude
committed
fix(doctor): Makefile auto-detects Python, no hard 3.12 dep (codex P2)

Your Name
and
claude
committed
fix(doctor): Makefile prefers active venv, skips system 'python' (codex round 2)

Your Name
and
claude
committed
fix(doctor): Makefile validates interpreter version, handles pyenv (codex round 3)

Your Name
and
claude
committed
feat(doctor): benchmark tier — cross-model scorecard sweep

Your Name
and
claude
committed
fix(doctor): codex round 1 — explicit-models check + LM Studio layout

Your Name
and
claude
committed
fix(doctor): pass discovered local path to server (codex round 2 P1)

Your Name
and
claude
committed
fix(doctor): discovery requires actual weight files, not just config.json

Your Name
and
claude
committed
fix(doctor): validate every shard against the safetensors index (codex round 2 P1)

Your Name
and
claude
committed
fix(doctor): treat unreadable safetensors index as incomplete (codex round 3)

Your Name
and
claude
committed
fix(doctor): self-review fixes — alias resolution, docs accuracy, escape hygiene

Your Name
and
claude
committed
fix(doctor): scope 600s boot timeout to full tier only (codex P2)

Your Name
and
claude
committed
fix(doctor): per-model boot-timeout heuristic (codex round 3 P2)

Your Name
and
claude
committed
fix(doctor): single generous boot timeout (codex round 4 P2)

Your Name
and
claude
committed
style: ruff format across codebase (CI lint fix)

Your Name
and
claude
committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat: rapid-mlx doctor — 4-tier regression harness#125

feat: rapid-mlx doctor — 4-tier regression harness#125
raullenchai merged 28 commits intomainfrom
feat/doctor-harness

Commits on Apr 16, 2026

chore: clean up ruff lint debt (43 errors → 0)

feat(doctor): scaffold + smoke tier — regression harness foundation

fix(doctor): codex review fixes — imports, source-checkout guard, unique run dirs

fix(doctor): atomically reserve run dir to survive concurrent invocations

feat(doctor): check tier — server + API + perf + baseline diff

fix(doctor): codex review — regression_suite parsing, per-model baselines, threshold names

chore(doctor): record initial Qwen3.5-4B baseline for check tier

fix(doctor): codex round 2 — tc_latency_ms sign + injective slug

fix(doctor): codex round 3 — legacy slug fallback for baselines

fix(doctor): codex round 4 — validate legacy baseline model identity

feat(doctor): full tier — multi-model loop + agent profile sweep

fix(doctor): codex round 1 — per-model diff + check name disambiguation

fix(doctor): codex round 2 — restore diff.md on no-baseline + injective slug

feat(doctor): regression-detection tests + harness/README + Makefile

fix(doctor): Makefile auto-detects Python, no hard 3.12 dep (codex P2)

fix(doctor): Makefile prefers active venv, skips system 'python' (codex round 2)

fix(doctor): Makefile validates interpreter version, handles pyenv (codex round 3)

feat(doctor): benchmark tier — cross-model scorecard sweep

fix(doctor): codex round 1 — explicit-models check + LM Studio layout

fix(doctor): pass discovered local path to server (codex round 2 P1)

fix(doctor): discovery requires actual weight files, not just config.json

fix(doctor): validate every shard against the safetensors index (codex round 2 P1)

fix(doctor): treat unreadable safetensors index as incomplete (codex round 3)

fix(doctor): self-review fixes — alias resolution, docs accuracy, escape hygiene

fix(doctor): scope 600s boot timeout to full tier only (codex P2)

fix(doctor): per-model boot-timeout heuristic (codex round 3 P2)

fix(doctor): single generous boot timeout (codex round 4 P2)

style: ruff format across codebase (CI lint fix)