About
Detecting a failure is the easy half
Every data team already knows when something broke. What costs them the evening is working out why: which pipeline, which step, which change, and how to prove it.
Monitoring stops where the work starts
An alert fires, or finance writes in to say the EUR totals have been missing since Tuesday. Something already detected the problem. What happens next is manual: an engineer works out which pipeline produces the number, opens its code, finds the logs for the last few runs, compares a failing night with a good one, checks what was deployed, writes a count query, and eventually finds the line that changed.
That work is skilled and slow, and it is the same shape every time. It is also almost entirely reconstructable from evidence that already exists: the repository, the logs, the ticket, the commit history. Decim does that reconstruction, and shows its working.
What Decim actually does
You open an incident with a plain description of what is wrong, and optionally the affected period, a Jira ticket and any files you have. Decim has already read your GitHub repository and found the pipelines in it: what triggers each one, its steps, the code behind each step, and where its logs live.
- Locate. It matches the incident to a pipeline and a step, and cites the code that made the match.
- Read the logs. It walks the steps, most suspicious first, pulling logs from Google Cloud Logging or through the optional Windows agent, and compares the failing run with successful runs up to 30 days back. Every finding cites the exact log lines.
- Weigh explanations. It keeps up to five competing hypotheses, each with the evidence for and against, and revises them as evidence arrives.
- Settle it. When the deciding evidence is somewhere it cannot read, usually a database, it writes a read-only query or a file request for a person to run, and takes the result back in.
- Conclude. A cause is confirmed only with corroborating evidence. Otherwise the answer is "likely", "unclear", or "needs log access", and it says which.
If there is a fix, it proposes one as an SQL repair for you to run, a code patch, or both, and opens it as a draft pull request. Nothing is merged automatically. The product overview goes through each stage in detail.
Built for the pipelines you actually run
Decim reads code rather than relying on a parser for one framework, so it is not tied to a particular stack. Teams running custom ETL services and scheduled jobs in Python, .NET, Java or SQL, cloud data teams on services such as AWS Glue, and warehouse teams all hit the same problem: a pipeline that reports success while the numbers are wrong, and an investigation that spans code, logs and tickets in several systems. That investigation is the job Decim is built for.
What we believe
An explanation you cannot check is worthless. Every finding Decim reports cites the evidence behind it: the file and line range, the log excerpt with its offsets and hash, the commit. If the evidence does not support a conclusion, saying so is the correct answer.
Absence is evidence. The most expensive failures do not throw. They complete, report success, and quietly drop rows. A log line that appeared every night and stopped is often the whole story.
Rejected explanations are part of the record. A hypothesis that was ruled out stays visible with the evidence that ruled it out, so the reader can see what was considered, not just what won.
Read-only by default, limits published. Decim reads your code, logs and tickets. It writes only a draft pull request when someone asks, never runs SQL against your systems, and lists what it cannot do yet on the security page.
Who builds it
Decim is a Ryware product, built by engineers who have spent a long time on the receiving end of a pipeline that broke at an unreasonable hour. It is pre-release and working with design partners; there is no customer list on this site, and we will not invent one.
Get started
Bring us a pipeline that broke last week
The fastest way to evaluate this is a real incident you already know the answer to. If Decim gets it wrong, that is a far more useful demo than one where it doesn't.