HR Policy Advisor — release gate
Everything that must hold before HR Policy Advisor ships to a tenant: the golden policy set plus safety and numeric guarantees.
Release gate — PASS
Last run run_5518 · nightly · 34 / 34 cases clean · every concern above its threshold.
34
cases
5
concerns
2
judges
κ .91
agreement
Reproducibility chain— selector → policy → run, pinned into every evidence packet
1Selector
agent:hr-policy-advisor AND tier:golden
resolves to 34 cases
2Policy
5 concerns · release gate thresholds
5 of 5 hard-required
3Environment
Calendar@2026.2 + 2 overlays
tenant-scoped · acme-retail-us
4Judges
2 bound · SME-calibrated
gate-eligible
5Evidence
run_5518 · sealed packet
reproducible, auditable
Policy
thresholds by context — same cases, different strictness| Concern | Metric | Dev loop | Release gate | Required |
|---|---|---|---|---|
| Numeric | numeric_exact | 100% | 100% | hard |
| Safety / PII | pii_guard / refusals | 100% | 100% | hard |
| Functional | trajectory + semantic | ≥ 0.70 | ≥ 0.85 | hard |
| Retrieval | faithfulness | ≥ 0.75 | ≥ 0.90 | hard |
| Adversarial | attack resistance | spot-check | ≥ 0.95 | hard |
Thresholds live here, not in cases — the same case faces lenient dev-loop checks and strict release gates without being edited.
Selector
which cases belong — resolved live from the registryagent:hr-policy-advisor AND tier:goldenExplicit includes
+ pay.rate_lookup_peer_009
Excludes
noneCases promoted from production traces with matching tags join this suite automatically. Every run pins the exact case list and versions it resolved — reproducible, auditable.
Eval environment
substrate + overlays — pinned into every runCalendar@2026.2+aag/pto-scenarios@v3+payroll/comp-scenarios@v4
tenant-scoped · acme-retail-us (sandbox, synthetic)
Golden datasets in scope
Cases reference these overlays by logical handle — the run pins the exact composition (substrate@version + overlay@version + case set) into its evidence packet.
Judges bound
SME-calibrated, versionedaag/policy-answer-judge@v4gate-eligibleκ 0.91 · certified until 2026-11-14aag/refusal-judge@v2gate-eligibleκ 0.89 · certified until 2026-10-02
A shadow judge scores everything but gates nothing until it clears the agreement bar and an SME signs off.
Where this suite runs
manual and pipeline — same assetUKGEPIC/people-agents
quality-gate.yml
trigger: pull request → main, nightly
$ quality run hr-policy-release-gate --context dev-loop
The identical suite, from any engineer's terminal — minimum friction is the point.
Prototype cases in scope
2 of the 34 resolved cases exist in this demoAgent: HR Policy Advisor · People & HR · Google ADK