CEX HFT: define latency measurement and evidence guideline

[proerror77]

Parent

• PRD: CEX 做市执行安全与热路径修正 #350

Change contract

Publish one code-audited guideline that defines canonical CEX HFT latency timestamps, benchmarks, percentile evidence, and outcome validation. This issue changes documentation only; it does not alter runtime behavior.

Acceptance criteria

• Audit the current market-data and latency instrumentation paths on the declared base SHA.
• Define distinct semantics for exchange event time E, trade/match time T, userspace WS frame delivery, parse completion, book commit, strategy intent, local write, venue ACK, and private order report.
• State that public Trade local-minus-T is not a benchmark for order-book transport latency.
• Require same stream, endpoint, host, clock discipline, sample count, and capture-boundary disclosure for before/after comparisons.
• Require p50, p95, p99, and p99.9 with sample count and clock error; outcome checks include stressed cancel success and stratified post-fill markout.
• Separate repository evidence, exchange-wire evidence, deployment evidence, and trading-outcome evidence.
• Record current code gaps and route each runtime change to a separate follow-up issue.

Out of scope

• Runtime code changes
• Live trading enablement
• Exchange deployment or latency claims
• Fixed microsecond SLOs without measured venue-specific evidence

Dependency / merge order

• Parent: PRD: CEX 做市执行安全与热路径修正 #350
• No code dependency; documentation can merge independently.

Focused validation

• Documentation links resolve.
• git diff --check passes.
• Every code-path claim is anchored to a file/symbol on the declared base SHA.

Rollout / rollback impact

None. Documentation-only; rollback is reverting the guideline.