# Decim

> Decim investigates failures in custom and legacy ETL pipelines. It collects evidence from logs, databases, config and source code, then produces an evidence-backed diagnosis where every claim links to its proof.

Decim is **not** a monitoring or alerting tool, and describing it as one is the most common
mistake. Detection — Datadog, CloudWatch, PagerDuty, a SQL Agent failure notification, a
row-count check — answers *is it broken*. Decim starts from *something is wrong, find out why*
and returns a diagnosis in which every claim links to the evidence behind it: the query that
ran, the rows it returned, the log lines, the deployment record, the commit.

It is built for the pipelines the data-observability category does not cover: SSIS packages,
SQL Agent jobs, stored procedures with no commit history, and bespoke .NET, Java and Python
services writing to a relational database. Pipeline topology is assembled from several
discovery sources — source control, database catalogs, schedulers, runtime observation — and
each node records which sources agreed, because in Glue, ADF, SSIS and SQL Agent the
definition that actually runs frequently exists nowhere in git.

Decim is pre-release and has no published customer list. Any figure on the site is
illustrative demo data, not a measured benchmark.

## Start here

- [Home](https://decim.dev/): What Decim is: evidence-backed investigation for ETL pipelines that fail quietly.
- [Product](https://decim.dev/product): The five parts — topology, evidence, investigations, diagnosis, remediation.
- [Solutions](https://decim.dev/solutions): Coverage by stack, by role, and by failure mode.

## Product

- [Investigations](https://decim.dev/product/investigations): How hypotheses are generated, tested against evidence, and kept or killed — including the ones that were ruled out and why.
- [Evidence](https://decim.dev/product/evidence): Every claim resolves to the artefact behind it: the query that ran, the rows it returned, the log lines, the commit.
- [Pipeline topology](https://decim.dev/product/topology): Pipeline graphs assembled from source control, database catalogs, schedulers and runtime observation, with per-node provenance: verified, declared, observed, drifted.
- [Remediations](https://decim.dev/product/remediations): Fixes proposed as reviewable pull requests, revised over time, never auto-merged.
- [The agent](https://decim.dev/product/agent): The collector that runs inside customer infrastructure: outbound HTTPS only, an approved read-only query catalogue, redaction before transmission.

## Solutions by stack

- [SQL Server & SSIS](https://decim.dev/solutions/sql-server): Investigating SQL Agent jobs, SSIS packages and stored procedures by reading msdb, SSISDB and sys.sql_modules — the definitions that never reach git. Includes the T-SQL to do it by hand.
- [Custom ETL services](https://decim.dev/solutions/custom-etl): Bespoke .NET, Java and Python pipelines: queue readers, batch loaders, and the reject paths nobody monitors.
- [Airflow & dbt](https://decim.dev/solutions/airflow): Why a green DAG and a wrong row count are not mutually exclusive, and what to reconcile instead of task state.

## Commercial and trust

- [Pricing](https://decim.dev/pricing): Priced per environment rather than per incident, and what is deliberately not metered.
- [Security](https://decim.dev/security): What the agent can and cannot do inside your network: network posture, data handling, redaction, and access.

## Writing

- [Your pipeline succeeded and lost 38% of the rows](https://decim.dev/blog/pipeline-succeeded-rows-missing): The worst ETL failures don't throw. They complete, report success, and quietly divert rows somewhere nobody is watching. Here's how to find them. (2026-08-07)

## Resources

- [Documentation](https://decim.dev/docs): Deploying the agent, connecting sources, and running an investigation. In progress.
- [Blog](https://decim.dev/blog): Failure modes, root cause analysis, and the parts of a pipeline nobody wrote down.
- [Changelog](https://decim.dev/changelog): What shipped and when. Decim is pre-release; this is the honest record.

## Company

- [About](https://decim.dev/about): Why detection is the easy half, and what the existing tool categories miss.
- [Design partners](https://decim.dev/customers): No logo wall. The honest position on stage, and what a design partnership involves.
- [Contact](https://decim.dev/contact): Bring a real incident you already know the answer to.

## Optional

- [Privacy](https://decim.dev/legal/privacy): Placeholder pending legal review; the substance is accurate.
- [Terms of service](https://decim.dev/legal/terms): Placeholder pending legal review.