# Security Guardrails (Read-only)

Date: 2026-03-23  
Status: active constraints for current environment

## 1. Core principle

Integration is analytics-only.
Source 1C remains system of record and is never mutated by this stack.

## 2. Hard restrictions

1. Read-only access to 1C in operational path.
2. `execute_code` is forbidden in operational contour.
3. No document posting/editing/deletion from bridge layer.
4. No config modification in client production 1C for analytics needs.
5. Proxy endpoint should remain local/internal (`127.0.0.1` / trusted network).

## 3. Runtime guardrails

- `ALLOW_DANGEROUS_WITH_APPROVAL=false`
- bounded query windows (`limit_per_set`)
- explicit refresh runs with audit (`refresh_runs`)
- feature/risk runs fully auditable (`feature_runs`, `risk_runs`)

## 4. Access segregation

Use separate technical 1C account with minimal read rights:

- metadata read
- approved query execution
- object read by reference

No write role grants.

## 5. Data handling

- Canonical/feature/risk stores are derivative analytical caches.
- They do not become write-back channels to 1C.
- Logs should avoid unnecessary sensitive payload duplication.

## 6. Operational checks

Before daily operation:

1. verify `/health` for bridge/proxy
2. verify no dangerous flags are enabled
3. verify latest runs are successful and recent

## 7. Incident policy

If abnormal or risky behavior is detected:

1. stop proxy access
2. preserve logs and run records
3. require manual review before restart