diff --git a/.openclaw/agents/deployer.md b/.openclaw/agents/deployer.md
index e668931..53611cb 100644
--- a/.openclaw/agents/deployer.md
+++ b/.openclaw/agents/deployer.md
@@ -21,10 +21,20 @@ On stage `deploy-staging` your job is to run the staging test suite and write a
 
 ### Steps:
 
-1. Run the staging test suite against the live staging environment:
+1. Run the staging test suite against the live staging environment.
+   **CANONICAL: run INSIDE the `orchestrator-staging` container via `docker exec`**
+   (ORCH-048, ADR-001) — NOT from the host:
    ```bash
-   python3 scripts/staging_check.py --base-url http://localhost:8501 --mode stub
+   docker exec orchestrator-staging \
+     python3 /repos/orchestrator/scripts/staging_check.py \
+     --base-url http://localhost:8501 --mode stub
    ```
+   Why: the B6 registry-isolation check reads the registry from the running
+   instance's own process-env (`.env.staging`). Running from the host leaves
+   `ORCH_PROJECTS_JSON` unset → B6 falls back to the default (ET+ORCH) registry
+   → false FAIL → spurious rollback. The script path is `/repos/orchestrator/scripts/…`
+   (bind-mount); `scripts/` is NOT copied into the image, so `/app/scripts` does
+   not exist. Details: `docs/operations/STAGING_CHECK.md`.
 
 2. Check the exit code:
    - Exit code **0** = all tests PASS → `staging_status: SUCCESS`
diff --git a/CHANGELOG.md b/CHANGELOG.md
index c008b1f..2e3cbfb 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -23,6 +23,7 @@
 - Цепочка стадий: `... testing → deploy-staging → deploy → done` (была без `deploy-staging`).
 
 ### Fixed
+- **Staging-чек B6 читает реестр из окружения работающего staging-инстанса** (ORCH-048): блок B6 «Registry: sandbox present, prod ET/ORCH absent» в `scripts/staging_check.py` давал **ложный FAIL** (`prod-ET=YES(BAD!)`, `prod-ORCH=YES(BAD!)`) при фактически исправной изоляции — единственный чек suite, который не ходил к инстансу по HTTP, а импортировал `src.projects` локально через host-path хак `sys.path.insert(0, "/repos/orchestrator")` + `importlib.reload`, строя реестр из `ORCH_PROJECTS_JSON` **process-env запускающего процесса**. При фактическом запуске деплоером с хоста переменная не задана → дефолт `_DEFAULT_PROJECTS` (ET+ORCH) → ложный FAIL → лишний откат `deploy-staging → development`. Решение (вариант «в», ADR-001): host-path хак удалён; suite канонически запускается ВНУТРИ контейнера `orchestrator-staging` через `docker exec … python3 /repos/orchestrator/scripts/staging_check.py` (`scripts/` доступен только через bind-mount, `import src.projects` резолвится через `PYTHONPATH=/app` из кода контейнера, env — `.env.staging`) → B6 читает реестр именно работающего инстанса, без HTTP-bootstrap и «курицы-яйца». Логика вердикта вынесена в чистую `_evaluate_b6(known) -> (passed, detail)` (инвариант `passed ⟺ SANDBOX ∈ known ∧ PROD_ET ∉ known ∧ PROD_ORCH ∉ known`, формат detail сохранён) + `_known_project_ids_from_registry()` / `_run_b6()` с детерминированным FAIL при недоступности источника (не ложный PASS, не необработанное исключение). Синхронно обновлены `.openclaw/agents/deployer.md` (команда стадии через `docker exec`) и `docs/operations/STAGING_CHECK.md`. `src/projects.py`, `.env*` и прочие чеки A/B4/B5/C не тронуты; реестр `QG_CHECKS` и `check_staging_status` (ADR-0003) не менялись. ADR `docs/work-items/ORCH-048/06-adr/ADR-001-b6-registry-via-in-container-run.md`. Тесты: `tests/test_staging_check_b6.py`.
 - **Testing-гейт `check_tests_passed` читает `result:` наравне с `verdict:`/`status:`** (ORCH-047): парсер `_parse_tests_verdict` (`src/qg/checks.py`) теперь принимает три равноправных машиночитаемых поля frontmatter `13-test-report.md` — `result:` (канон промпта тестера `.openclaw/agents/tester.md`, `result: PASS|FAIL`), плюс легаси `verdict:` и `status:` (enduro-trails ET-001..ET-014); достаточно любого одного непустого. Устраняет рассинхрон контракта: тестер честно эмитил `result: PASS` без `verdict:`/`status:`, парсер попадал в ветку «нет машинного вердикта» → откат `testing → development` в петлю до исчерпания `MAX_DEVELOPER_RETRIES` (наблюдалось на ORCH-17; ORCH-016 прошёл лишь из-за избыточного дублирования полей). Семантика приоритетов сохранена и распространена на все три поля через объединённую строку: negative-токен в любом поле авторитетен (перебивает positive), наборы токенов заморожены (обратная совместимость). Сигнатура гейта, имя и реестр `QG_CHECKS` не менялись. ADR `docs/work-items/ORCH-047/06-adr/ADR-001-result-field-in-tests-gate.md`. Тесты: `tests/test_qg.py::TestCheckTestsPassed`.
 - БАГ-8: провал deploy/deploy-staging → корректный откат на `development`.
 - Изоляция тестов от живого Plane API (PR #27): autouse-фикстура сброса settings.
diff --git a/scripts/staging_check.py b/scripts/staging_check.py
index 87edf59..75ba892 100644
--- a/scripts/staging_check.py
+++ b/scripts/staging_check.py
@@ -8,8 +8,14 @@ Checks:
   Block C — E2E   (create task in SANDBOX → trigger pipeline via /webhook/plane
                    → verify branch + job enqueued → CLEANUP in finally)
 
-Usage (inside the container or with correct env set):
-    python3 scripts/staging_check.py [--base-url http://localhost:8501] [--mode stub|full-real]
+Usage — CANONICAL: run INSIDE the orchestrator-staging container (ORCH-048, ADR-001)
+so B6 reads the registry from the running instance's own env (.env.staging):
+    docker exec orchestrator-staging \
+      python3 /repos/orchestrator/scripts/staging_check.py \
+      --base-url http://localhost:8501 [--mode stub|full-real]
+
+Running from the host leaves ORCH_PROJECTS_JSON unset → B6 falls back to the
+default (ET+ORCH) registry → false FAIL. See docs/operations/STAGING_CHECK.md.
 
 Exit code: 0 = all PASS, non-zero = at least one FAIL.
 
@@ -214,6 +220,59 @@ SANDBOX_PROJECT_ID = "8c5a3025-4f9d-4190-b79f-fa06276bb27e"
 PROD_ET_PROJECT_ID = "7a79f0a9-5278-49cd-9007-9a338f238f9c"
 PROD_ORCH_PROJECT_ID = "8da6aa25-a60e-44d6-a1e2-d8ae59aa7d6a"
 
+B6_LABEL = "B6 Registry: sandbox present, prod ET/ORCH absent"
+
+
+def _evaluate_b6(known: set[str]) -> tuple[bool, str]:
+    """Pure verdict logic for the B6 registry-isolation check (ORCH-048).
+
+    PASS ⟺ SANDBOX ∈ known ∧ PROD_ET ∉ known ∧ PROD_ORCH ∉ known (TR-2).
+    ``detail`` keeps the human-readable ``sandbox=…, prod-ET=…, prod-ORCH=…``
+    format (TR-3). Isolated from any I/O so both outcomes are unit-testable
+    without a live staging instance or docker (02-trz §9, ADR-001).
+    """
+    sandbox_present = SANDBOX_PROJECT_ID in known
+    et_absent = PROD_ET_PROJECT_ID not in known
+    orch_absent = PROD_ORCH_PROJECT_ID not in known
+    passed = sandbox_present and et_absent and orch_absent
+    detail = (
+        f"sandbox={'YES' if sandbox_present else 'NO'}, "
+        f"prod-ET={'NO(good)' if et_absent else 'YES(BAD!)'}, "
+        f"prod-ORCH={'NO(good)' if orch_absent else 'YES(BAD!)'}"
+    )
+    return passed, detail
+
+
+def _known_project_ids_from_registry() -> set[str]:
+    """Registry of the *running staging instance* — its own process-env (ORCH-048).
+
+    The suite is canonically run INSIDE ``orchestrator-staging`` via
+    ``docker exec`` (ADR-001), so ``src.projects`` resolves through the
+    container's ``PYTHONPATH=/app`` to ``/app/src/projects.py`` and reads
+    ``ORCH_PROJECTS_JSON`` from ``.env.staging``. This reflects exactly the
+    registry the live instance serves webhooks with — no host-path hack, no HTTP
+    bootstrap dependency.
+    """
+    from src.projects import known_plane_project_ids
+    return known_plane_project_ids()
+
+
+def _run_b6(results: Results) -> None:
+    """Run the B6 registry-isolation check and record its verdict.
+
+    Builds the known-id set from the running instance's registry and applies
+    ``_evaluate_b6``. Any failure to obtain the registry yields a deterministic
+    FAIL with a clear detail (TR-4) — never an unhandled exception and never a
+    false PASS.
+    """
+    try:
+        known = _known_project_ids_from_registry()
+    except Exception as e:
+        results.add(B6_LABEL, False, f"registry source unavailable: {e}")
+        return
+    passed, detail = _evaluate_b6(known)
+    results.add(B6_LABEL, passed, detail)
+
 
 def block_b(results: Results):
     print(f"\n{_BOLD}[Block B] ACCESS{_RESET}")
@@ -260,28 +319,11 @@ def block_b(results: Results):
     except Exception as e:
         results.add("B5 Gitea: orchestrator-sandbox accessible, push=true", False, str(e))
 
-    # B6 — Registry: sandbox in known IDs, prod ET/ORCH NOT in known IDs
-    try:
-        # Import from inside the container (script runs in /repos/orchestrator context)
-        sys.path.insert(0, "/repos/orchestrator")
-        # Force reload to pick up container env
-        import importlib
-        if "src.projects" in sys.modules:
-            importlib.reload(sys.modules["src.projects"])
-        from src.projects import known_plane_project_ids
-        known = known_plane_project_ids()
-        sandbox_present = SANDBOX_PROJECT_ID in known
-        et_absent = PROD_ET_PROJECT_ID not in known
-        orch_absent = PROD_ORCH_PROJECT_ID not in known
-        ok = sandbox_present and et_absent and orch_absent
-        detail = (
-            f"sandbox={'YES' if sandbox_present else 'NO'}, "
-            f"prod-ET={'NO(good)' if et_absent else 'YES(BAD!)'}, "
-            f"prod-ORCH={'NO(good)' if orch_absent else 'YES(BAD!)'}"
-        )
-        results.add("B6 Registry: sandbox present, prod ET/ORCH absent", ok, detail)
-    except Exception as e:
-        results.add("B6 Registry: sandbox present, prod ET/ORCH absent", False, str(e))
+    # B6 — Registry: sandbox in known IDs, prod ET/ORCH NOT in known IDs (ORCH-048).
+    # Reads the registry of the running staging instance from its own process-env
+    # (canonical: docker exec inside orchestrator-staging — ADR-001). No host-path
+    # hack; deterministic FAIL if the registry source is unavailable (TR-4).
+    _run_b6(results)
 
 
 # ---------------------------------------------------------------------------
diff --git a/tests/test_staging_check_b6.py b/tests/test_staging_check_b6.py
new file mode 100644
index 0000000..0eb8940
--- /dev/null
+++ b/tests/test_staging_check_b6.py
@@ -0,0 +1,151 @@
+"""ORCH-048: unit tests for the B6 registry-isolation verdict in staging_check.py.
+
+B6 «Registry: sandbox present, prod ET/ORCH absent» is the staging-isolation
+safety check. Its verdict logic is isolated into the pure function
+``_evaluate_b6(known) -> (passed, detail)`` so both outcomes (clean staging
+registry → PASS, polluted registry → FAIL) can be tested without standing up a
+live staging instance or docker (02-trz §9, ADR-001).
+
+These tests target that pure function plus the deterministic-degradation path
+(``_run_b6``) and statically assert the host-path hack is gone (TR-6 / TC-06).
+"""
+
+import importlib.util
+import pathlib
+
+import pytest
+
+# ---------------------------------------------------------------------------
+# Load scripts/staging_check.py by path (scripts/ is not an importable package).
+# ---------------------------------------------------------------------------
+_SCRIPT_PATH = (
+    pathlib.Path(__file__).resolve().parent.parent / "scripts" / "staging_check.py"
+)
+
+
+def _load_module():
+    spec = importlib.util.spec_from_file_location("staging_check", _SCRIPT_PATH)
+    module = importlib.util.module_from_spec(spec)
+    spec.loader.exec_module(module)
+    return module
+
+
+sc = _load_module()
+
+SANDBOX = sc.SANDBOX_PROJECT_ID
+PROD_ET = sc.PROD_ET_PROJECT_ID
+PROD_ORCH = sc.PROD_ORCH_PROJECT_ID
+
+
+# ---------------------------------------------------------------------------
+# TC-01 — clean staging registry → PASS
+# ---------------------------------------------------------------------------
+def test_tc01_clean_registry_passes():
+    passed, detail = sc._evaluate_b6({SANDBOX})
+    assert passed is True
+    assert "sandbox=YES" in detail
+    assert "prod-ET=NO(good)" in detail
+    assert "prod-ORCH=NO(good)" in detail
+
+
+# ---------------------------------------------------------------------------
+# TC-02 — prod-ET leaked into registry → FAIL
+# ---------------------------------------------------------------------------
+def test_tc02_prod_et_present_fails():
+    passed, detail = sc._evaluate_b6({SANDBOX, PROD_ET})
+    assert passed is False
+    assert "sandbox=YES" in detail
+    assert "prod-ET=YES(BAD!)" in detail
+    assert "prod-ORCH=NO(good)" in detail
+
+
+# ---------------------------------------------------------------------------
+# TC-03 — prod-ORCH leaked into registry → FAIL
+# ---------------------------------------------------------------------------
+def test_tc03_prod_orch_present_fails():
+    passed, detail = sc._evaluate_b6({SANDBOX, PROD_ORCH})
+    assert passed is False
+    assert "sandbox=YES" in detail
+    assert "prod-ET=NO(good)" in detail
+    assert "prod-ORCH=YES(BAD!)" in detail
+
+
+# ---------------------------------------------------------------------------
+# TC-04 — sandbox absent (empty registry) → deterministic FAIL, no exception
+# ---------------------------------------------------------------------------
+def test_tc04_empty_registry_fails_without_sandbox():
+    passed, detail = sc._evaluate_b6(set())
+    assert passed is False
+    assert "sandbox=NO" in detail
+
+
+# ---------------------------------------------------------------------------
+# TC-05 — both prod projects leaked → FAIL
+# ---------------------------------------------------------------------------
+def test_tc05_both_prod_present_fails():
+    passed, detail = sc._evaluate_b6({SANDBOX, PROD_ET, PROD_ORCH})
+    assert passed is False
+    assert "prod-ET=YES(BAD!)" in detail
+    assert "prod-ORCH=YES(BAD!)" in detail
+
+
+# ---------------------------------------------------------------------------
+# TC-06 — registry source no longer depends on the host-path hack (TR-6)
+# ---------------------------------------------------------------------------
+def test_tc06_no_host_path_hack_in_source():
+    source = _SCRIPT_PATH.read_text(encoding="utf-8")
+    # The host-worktree path injection and the env-of-the-launcher reload that
+    # caused the false FAIL must be gone from the B6 mechanics.
+    assert 'sys.path.insert(0, "/repos/orchestrator")' not in source
+    assert "importlib.reload" not in source
+
+
+def test_tc06_registry_loader_uses_src_projects():
+    # The verdict input is built from src.projects.known_plane_project_ids()
+    # resolved via the running instance's own PYTHONPATH/env — not from a
+    # host-path-injected import. We verify the loader delegates to that function.
+    import src.projects as projects_mod
+
+    sentinel = {"sentinel-id"}
+    original = projects_mod.known_plane_project_ids
+    projects_mod.known_plane_project_ids = lambda: sentinel
+    try:
+        known = sc._known_project_ids_from_registry()
+    finally:
+        projects_mod.known_plane_project_ids = original
+    assert known == sentinel
+
+
+# ---------------------------------------------------------------------------
+# TC-07 — degraded registry source → deterministic FAIL (not false PASS, not raise)
+# ---------------------------------------------------------------------------
+def test_tc07_source_failure_is_deterministic_fail(monkeypatch):
+    def _boom():
+        raise RuntimeError("registry import blew up")
+
+    monkeypatch.setattr(sc, "_known_project_ids_from_registry", _boom)
+
+    results = sc.Results()
+    # Must not raise.
+    sc._run_b6(results)
+
+    assert len(results._items) == 1
+    label, passed, detail = results._items[0]
+    assert passed is False
+    assert "registry source unavailable" in detail
+    assert "registry import blew up" in detail
+
+
+# ---------------------------------------------------------------------------
+# _run_b6 happy path wiring (clean registry → PASS result recorded)
+# ---------------------------------------------------------------------------
+def test_run_b6_records_pass_for_clean_registry(monkeypatch):
+    monkeypatch.setattr(
+        sc, "_known_project_ids_from_registry", lambda: {SANDBOX}
+    )
+    results = sc.Results()
+    sc._run_b6(results)
+    assert len(results._items) == 1
+    _label, passed, detail = results._items[0]
+    assert passed is True
+    assert "sandbox=YES" in detail