What changed

• add a base-owned JMH profile for TableRegionFingerprintService
• measure the complete started-table region map and a representative per-tile batch (four private/public hands, discards, melds, and all 136 wall slots)
• record both average time and gc.alloc.rate.norm for both workloads
• add performance-region-fingerprint workflow/label routing and document the profile
• add the missing normalized-allocation guardrail to the existing infrastructure fingerprint profile

Behavior protection

The fixture uses the real TableRenderSnapshotFactory and TableRenderLayout. Before every JMH iteration it compares production results with an independent reference implementation of the current Objects.toString(value, ""), colon-delimited, per-character FNV contract. It also fixes the exact DelimitedFingerprintBuilder output. These sentinels run outside the timed section, so a later candidate cannot pass by changing fingerprint semantics.

The profile requires both primary time metrics to pass and treats allocation as a regression guardrail. TableRegionFingerprintService.class is required during profile preflight. All benchmark/config/workflow paths remain protected by the existing base-owned gate.

Scope

This is measurement infrastructure only. It does not modify src/main production behavior.

Validation

• git diff --cached --check
• parsed perf/ab/gate-config.json with PowerShell ConvertFrom-Json
• no local Gradle, JMH, test, or server execution, per project validation policy
• GitHub Build/Test/Performance infrastructure checks are the authoritative validation