results(linux-x64): epyc7b13 8v 10x run — isolation ratios reproduce vs Mac baseline

[EinarMontalvo]

📊 Summary

Linux-x64 (AMD EPYC 7B13, 8 vCPU / QEMU) 10× run of the isolation/concurrency harness — the diff vs main is the linux numbers side-by-side with the Mac baseline (and alongside results/linux-x64-c7i-20260709).

Runtime 3.2.1.30 (114509c6) x86_64 · langchain-core 0.3.86. All four groups 10/10 reps clean, and the isolation ratios reproduce cross-platform (absolute timings differ from Mac/c7i, as expected):

group	stock LangChain	stock RocketRide
concurrent-processing	.batch (shared conn) crashes 0/64	✅ safe, M=16 ~1.01s (0.894–1.150), 0 err
fault-isolation	in-process .abatch loses all 0/4	✅ survives (only crasher dies)
data-isolation	shared dict, 32 workers silently loses 2–95%	✅ 0 lost / 0 leaked
authoring-effort	14–17 lines + up to 5 hidden decisions	✅ 0 concurrency lines

Ran natively via the real C++ ./engine ai/eaas.py (not the bundled python — that would void the isolation measurement).

🛠️ Provisioning notes (Ubuntu 24.04) — not fixed in this PR

On a fresh Ubuntu 24.04 host, the documented one-click path still blocked before the engine could start. These host/setup gaps are not fixed in this PR (results-only):

1. python missing — Ubuntu ships python3 only; some docs/commands expect python. Workaround: apt install python-is-python3 (or always use python3 / the venv binary).
2. libc++.so.1 missing — Linux engine links against LLVM libc++. Without libc++1 / libc++1-18, startup fails with:
   error while loading shared libraries: libc++.so.1: cannot open shared object file
scripts/setup-ubuntu.sh does not currently install it.

After installing those manually, make start + full run_isolated.sh completed cleanly.

⚠️ Heads-up — readiness false-healthy (same bug called out in PR #3; not fixed in this PR)

scripts/start_engine.sh readiness gate can false-pass — same issue already mentioned in PR #3; still not fixed here.

On connection-refused, curl -w '%{http_code}' prints 000 and || echo 000 appends another → "000000", which != "000" → script declares the engine healthy before it is listening. We hit this on Ubuntu 24.04 when the binary failed to load libc++:

engine healthy on :5565 (HTTP 000000, pid …)
./engine: error while loading shared libraries: libc++.so.1: ...

✅ Test plan

• Manual Ubuntu 24.04 prereqs (python-is-python3, libc++1)
• make provision / make provision-competitors
• make start → real healthy status (HTTP 401)
• REPS=10 bash run_isolated.sh → all 4 groups 10/10 clean
• python aggregate.py → scorecard refreshed from runs/