---
title: Troubleshooting and FAQ
description: Common problems when running Prologue, and what usually causes them.
---


## No captures are appearing

- **HTTP captures missing.** Confirm your traffic is actually going through the Extractor's proxy port (`8080` by default), not straight at your system — anything that bypasses the proxy is invisible to Prologue. Also check the request method: only `POST`, `PUT`, and `DELETE` are captured, so a `GET`-only smoke test produces nothing to see.
- **SQL Server captures missing.** Confirm the connection string's account can enable Change Data Capture (or that CDC is already enabled, if you set `enableChangeDataCapture: false`), and that a `tables` allowlist, if you set one, actually includes the table you changed.
- **PostgreSQL captures missing.** Confirm the server has `wal_level = logical` — the Extractor can create its own replication slot and publication, but it can't change that server setting for you, and logical replication won't work without it.
- **OpenTelemetry captures missing.** Confirm `openTelemetry.enabled` is `true`, and that your service's name matches `serviceNames` (or that list is empty, to capture every service).

## The Interpreter says it found no capture files

Batch mode reads `.jsonl` files from the folder in `--captures`/`PROLOGUE_CAPTURES` (`/captures` by default in the image). This usually means the Extractor is configured for `output.kind: "Api"` (posting to a Receiver) rather than `"Json"`, or the mounted folder doesn't match where the Extractor is actually writing — see [Configuration](/prologue/reference/configuration/#prologueoutput).

## Captures from two different Prologues are mixed together

Every capture is stamped with a `prologueId`. If a folder or a Receiver's MongoDB collection holds captures from more than one Prologue, pass `--prologue-id`/`PROLOGUE_ID` to the Interpreter to interpret just the one you want — see [Running the Interpreter](/prologue/guides/running-the-interpreter/).

## The event model has mechanical, generic names

That's the heuristic result — structurally correct, but naming things after what the evidence literally showed rather than domain language. Configure the `llm` section in `cratis-prologue.json` to have the Interpreter refine names and derive a system name; see [Configure LLM refinement](/prologue/guides/running-the-interpreter/#configure-llm-refinement). Without a configured model, or in a non-interactive run, the Interpreter never guesses beyond the heuristics — it finalizes with its best mechanical effort rather than inventing domain language it can't verify.

## A service-mode Interpreter container keeps exiting

That's by design, not a crash. A service-mode session waiting on your answer past `PROLOGUE_GRACE_PERIOD`, or sitting idle past `PROLOGUE_IDLE_TIMEOUT`, exits cleanly on purpose — session state is already persisted in MongoDB, so an orchestrator restarting the container resumes the same session rather than losing it. If you're driving batch mode instead (the common case outside Studio), this doesn't apply.

## Where to ask

The [Cratis Discord](https://discord.gg/kt4AMpV8WV) is the fastest way to get unstuck if this page doesn't cover it.