faktwerk

Why faktwerk

The more AI a team uses, the more results pile up: drafts, summaries, explanations of code, broken-down tickets, risk estimates. They grow faster than anyone can read them. Keeping the overview means knowing where every statement comes from, whether it still holds and what a change further up means.

It hits hardest those who are not experts in the field. Marketing wants to know what three dozen pull requests mean for the next campaign – without reading the code. Management needs one page, engineering every detail. In the end there are three documents, and nobody knows whether their numbers still match.

faktwerk turns this around: every statement exists exactly once, as a fact. Everything else is just a view of it.

One fact, many views

A fact is the smallest unit of knowledge – a description, a metric, a feature from a pull request. It knows which system it comes from and how to find the way back. Facts come from the tools a team already uses: GitHub, GitLab, Jira or plain documents.

A view puts facts together for one role or workflow, like a table of contents. The same facts appear in many views, each sorted, grouped and detailed differently. As a report to read, as a table of metrics and later as a spatial model for a headset.

When a fact changes, every view shows the new state right away. Nothing is copied.

Zoom in instead of reading everything

Facts sit on levels. At the bottom is the source, such as a pull request or a ticket. Above it are facts derived from it: the single features, their functional and technical description. At the top a summary that explains the levels below in a few sentences.

Every view starts at the top and stays understandable. Whoever wants more zooms in – but doesn’t have to. And every fact knows its parents: from any summary there is a way back to the line in the source system.

When something changes at the bottom

A pull request gets reworked, and suddenly the release report is wrong. Today nobody notices. faktwerk overwrites nothing in that case – it puts a proposal beside every affected fact. A human compares, then promotes or drops it. Only a promoted proposal moves one level further up.

That keeps the summary honest without anyone hunting for what else needs to change.

How it feels

1. Facts emerge. A new pull request comes in. faktwerk breaks it into single features and describes each functionally and technically, with metrics such as changed lines or estimated risk.
2. A human reviews. The draft is read and corrected in a form of the matching view. Nothing unreviewed moves on.
3. The views are ready. Every role sees the same facts in its own compilation.
4. Decisions flow back. An approval or a comment in the view lands back in the source system, for example on the pull request.

AI drafts, humans decide

In faktwerk, AI only helps with drafting. Every draft is a proposal until someone approves it. After that everything is fixed: views and feedback to the source come out of the same data the same way every time, traceable and verifiable. Nothing is written back to another system unreviewed.

What faktwerk is made of

A small set of clearly separated building blocks around one single data source:

• A core holds all facts in one fixed data model. A ticket from GitHub looks just like one from Jira there and can be compared directly.
• Importers, one per source system, translate the data into this model. A new source only needs a new importer.
• Workflows describe steps: what is shown, what happens automatically and where a human decides. A workflow engine runs them and waits until someone has acted.
• One interface shows every view and every form, without its own code per workflow.

The fixed data model is on purpose: nobody has to build a schema for their project first, faktwerk is meant to start without configuration.

Who it is for

For everyone who works with AI in complex systems and has to keep track of many AI results. First software teams struggling with the review load of AI pull requests. Then product and project management, who need one shared truth for all roles. And industries where every statement needs a traceable origin and an approval.

faktwerk, monsterbrain and deciwa

monsterbrain organises the knowledge of one person or team in folders and Markdown files. faktwerk is a database for very many objects from many source systems. deciwa makes single AI changes visible and ratable.

Status

A concept. The architecture is planned through, nothing is built yet. Before it starts I want to know where the pain is greatest: in talks with teams that have to review many AI pull requests. “faktwerk” is a working title.

Interested in this tool? Let's talk.