Security posture

Security posture for early OpsRail evaluations.

OpsRail Command Center is designed to support low-risk pilots using synthetic, masked, or de-identified operational files before any production integration or PHI-bearing workflow.

01 · Pilot posture

Designed for low-risk evaluation.

The defaults below apply to every pilot. Production use cases that involve PHI follow a separate security review path.

No PHI required for pilot

Evaluations run on synthetic, masked, or de-identified operational files.

Manual CSV upload only

No SFTP, no API ingest, no agent installs. Files are uploaded one at a time.

No production integrations

No connections to plan, vendor, retailer, or PBM systems during evaluation.

No write-back

Read-only relative to your systems. The product does not push anything back.

Data can be deleted

Uploaded files, parsed rows, incidents, and briefs can be removed on request after evaluation.

OpenAI used only for briefs

AI is invoked only after deterministic detection, on structured incident evidence — never on raw CSVs.

02 · Current MVP architecture

Single data path. Deterministic at the core.

The product is a single web app, a single backend, and a single database. AI is downstream of the deterministic engine.

Upload

CSV files

Frontend

Next.js

Backend

FastAPI

Database

PostgreSQL

Analysis

Deterministic engine

Brief

OpenAI API

Frontend
Next.js · Vercel-ready deployment
Backend
FastAPI · Railway-ready deployment
Database
PostgreSQL
AI
OpenAI API — only for brief generation, post-detection
Source of truth
Deterministic incident engine

03 · What is sent to OpenAI

AI is downstream of detection.

The deterministic engine produces an incident first. Only the resulting structured evidence is used to generate a brief — never the raw CSV files.

  • Only structured incident evidence is sent to the model for brief generation.

  • Raw uploaded CSVs are not required to be sent to OpenAI.

  • The model is instructed not to invent unsupported facts and to stay within the provided evidence.

  • AI output is a communication aid. The deterministic engine remains the source of truth.

Brief output is reviewable before sharing. Reviewers should validate against the linked incident evidence before sending to executives, vendors, or retailers.

04 · Security controls for pilot

Baseline controls during evaluation.

  • HTTPS in transit in hosted environments

  • Secrets stored as environment variables, not in code

  • Database-backed auditability of uploads, incidents, and briefs

  • No-PHI pilot mode is the default evaluation path

  • Manual data deletion available on request

  • Limited user access during pilot evaluations

  • De-identified or synthetic data recommended for all evaluations

05 · Important limitations

What this page is — and is not.

We’d rather be explicit than aspirational. Procurement and security reviewers should treat this page as a pilot-readiness overview, not a compliance attestation.

Future production workflow may require…
  • The MVP is not yet SOC 2 certified.
  • The MVP is not yet HITRUST certified.
  • Future production PHI use may require a BAA, a vendor security review, formal access controls, audit logs, retention policies, and data processing terms.
  • SSO and role-based access control are not included in the no-PHI MVP.
  • This page is a pilot-readiness overview — not a compliance attestation.

Procurement-light evaluation

Run an evaluation with synthetic, masked, or de-identified data — before any security review.