Limitations
An honest page, because a manual that only lists strengths is a sales page. These are the places where Dlubal CALC is weaker than it looks, and what to do about each.
You are the engineer
Dlubal CALC produces results you sign. It shows the path to every number so you can check it, which is a different promise from being right. The reviewer is you.
The practical version: read the assumptions before the result, and check the governing case yourself when the answer is close to a limit.
Where answers are weaker
Example: you ask for the partial factors of a national annex that has never been uploaded. The answer says it was not found rather than producing a set that looks right. That sentence is the most valuable line in the answer, and it is easy to read past.
Anything the model was not shown. If a rule is not in the knowledge base and you did not attach it, the answer is reasoning without a source. It will say so, and that sentence is the important part of the answer.
Defaults and dialogs in the software. Mia will not tell you where a checkbox sits or what a setting defaults to unless a source said so. When you get "check it in the program" instead of an answer, that is the rule working, not a gap.
Very long conversations. Context is finite. A chat that has run through several subjects gives worse answers than a fresh chat with a clean description, which is why Chats recommends one chat per task.
Anything time-sensitive without a search. The model's own knowledge has a cutoff. For a recent release or a changed regulation, force a web search or supply the document.
What it cannot do at all
- Calculate inside an uploaded model file. It reads what the file contains. To calculate a model, connect RFEM: Connecting RFEM.
- See your other chats. Context stops at the conversation.
- Reach your machine on its own. Only through the Dlubal-Connector, and only while it is running and paired.
Reports do not render every script yet
| Script | In a PDF today |
|---|---|
| Latin, including diacritics | Renders |
| Chinese | Renders through an embedded fallback font |
| Korean, Russian, Ukrainian, Hindi | Not covered, characters come out as empty boxes |
| Arabic | Deferred, the engine cannot shape right-to-left text |
If your language is in the last two rows, read the results in the chat and hold off on the PDF. Reports has the full table.
Where to go next
- How Mia works for what happens between question and answer.
- Citations for how much a source reference claims.