Documentation · How answers work
How answers work
Some questions have an exact number. Others need an explanation. ForgeOrbis reads your question, decides which kind of answer it deserves, and always shows its work.
Two kinds of answers
“How many” and “what is the average” are counting questions; they have one correct figure. “Why did this happen” is an explanatory question; it needs a few sentences drawn from the right log lines. ForgeOrbis routes each question to the answer it deserves, so you never get a vague paragraph where a number belongs, or a bare number where you needed the story.
Exact
computed from your logs“how many labels did STATION_A1 print on 2026-07-10?”
The figure is computed directly from your data. It is exact and repeatable, never written by the model.
Explained
cited evidence“why did LabelRun stop printing around 14:20 on 2026-07-09?”
The print service lost its print queue connection at 14:22, retried three times, and recovered when the watchdog restarted it at 14:24. No labels were lost.
Explained answers are built from evidence
For a “what happened” question, ForgeOrbis gathers the log lines that actually relate to it and writes the answer only from those. Every factual sentence ends with a citation to the exact file, line, and timestamp, each with a match score so you can weigh how strong the evidence is. Open any citation to read the original line yourself.
Nothing in the answer comes from the model’s memory. If a sentence is not supported by a line in your logs, it does not appear.
Honest refusal
When nothing in your logs clears the bar, the answer is NOT FOUND IN LOGS. ForgeOrbis is tuned to refuse rather than guess. In an audit, a system that will say “I don’t have evidence for that” is worth more than one that always has something to say.
Why you can trust it
- Exact answers are computed from your data, so numbers are never invented.
- Explained answers quote your logs and cite every claim.
- When the evidence is not there, the system refuses instead of guessing.
See the measured accuracy behind these behaviours in Accuracy & benchmarks.