research: characterize governance latency across platforms

[SignalLayerLabs]

Question

How stable is MARGINAL's local governance overhead across operating systems and supported agent integrations?

Contribution wanted

Run a small, preregistered measurement matrix across at least two environments and contribute raw aggregate timing evidence.

Minimum evidence

• Engine + exact version, OS, Python/runtime version and MARGINAL commit.
• Warmup policy and number of repeated measurements.
• Median, p95 and sample count for governance latency.
• Separate Codex / Claude Code / OpenCode observations rather than pooling incompatible surfaces.
• No performance or savings claim beyond the measured setup.

Guardrail

Do not upload prompts, source code, commands, credentials or private repository names.