> ## Documentation Index
> Fetch the complete documentation index at: https://docs.reflections.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# ADR-0024: Loti content-monitoring feed integration

> Ingest daily Loti content-monitoring results into Postgres for downstream talent protection workflows.

<Info>**Status:** Accepted **Date:** 2026-02-23 **Deciders:** Reflections Maintainers</Info>

## Context

Loti provides daily JSONL feeds of content-monitoring results (face-match detections across platforms). Their Lambda endpoint is IP-whitelisted -- only pre-registered static IPs can fetch data.

Railway (the hosting platform) assigns ephemeral IPs to containers. This means the API cannot call Loti directly; a relay with a static IP is needed.

Requirements:

* Fetch daily JSONL files from Loti's IP-whitelisted endpoint.
* Forward results to the API in idempotent chunks.
* Minimize cost and operational surface area.

## Decision

Use an EC2 t3.micro relay with a daily cron job that fetches JSONL from Loti and POSTs it in 500-line chunks to `POST /internal/loti-feed`.

Data flow:

```text theme={null}
Loti Lambda --(IP-whitelisted)--> EC2 relay (cron, static EIP)
    --(chunked POST, x-internal-secret)--> Railway API /internal/loti-feed
    --(upsert ON CONFLICT DO NOTHING)--> Postgres loti_results table
```

The API endpoint validates JSONL, parses each line against a Zod schema, and upserts into `loti_results` deduped by `result_id`.

## Alternatives considered

| Alternative                    | Why rejected                                                                                            |
| ------------------------------ | ------------------------------------------------------------------------------------------------------- |
| Direct S3 cross-account access | Loti uses IP whitelisting, not IAM cross-account roles. Would require Loti to change their auth model.  |
| Lambda-to-Lambda push          | Requires Loti to implement a push model; they only support pull.                                        |
| Serverless NAT Gateway         | \~\$32/mo for a NAT Gateway + Lambda just to get a static IP. Overkill for one daily cron.              |
| Railway static IP add-on       | Not available on the current plan. Even if available, would couple relay concern to app infrastructure. |

## Consequences

**Benefits:**

* Low cost (\~\$7.50/mo for t3.micro + EIP).
* Minimal surface: single bash script, single cron entry, single API route.
* Fully idempotent: `ON CONFLICT (result_id) DO NOTHING` means re-running is safe.
* Chunked delivery (500 lines/request) keeps request sizes manageable.

**Costs:**

* EC2 instance to maintain (patching, monitoring).
* Additional infrastructure outside Railway.

**Risks:**

* Silent cron failure if EC2 goes down or Loti endpoint changes.

**Mitigations:**

* Monitor daily row count in `loti_results` -- alert if zero new rows for 48 hours.
* EC2 instance tagged and documented; relay script is simple enough to recreate in minutes.

## Implementation notes

* The migration creates the `loti_results` table with appropriate indexes.
* DB query module provides `upsertLotiResults` re-exported via the write entrypoint.
* API route validates and upserts JSONL chunks, authenticated via internal secret header.
* Body limit override is configured to accommodate larger JSONL payloads (10 MB).

## Related ADRs

* [ADR-0009: API architecture and authorization enforcement](/decisions/adr-0009)
* [ADR-0010: Ingestion orchestration, idempotency, and recovery](/decisions/adr-0010)