isanna heimdall

What it is

A finished study usually dies as a document. Here it leaves watchers behind: standing questions bound to signals — a figure in a filing, a price level, a rate derived from both — that are re-read on a schedule, so a conclusion is revisited when its evidence moves rather than when someone remembers to look.

Figures come from the primary source. A filing signal is read out of the filed document itself and carries the span it was read from, the accession it was filed under and the period it covers — so any number on the desk can be walked back to the bytes it came from.

How it works

Arming is deliberate

A study proposes watchers; putting one into service is an operator’s decision rather than a side effect of running the study.

Verbatim or nothing

A claim read out of a document resolves only against a span that appears verbatim in the stored bytes. There is no paraphrase path.

Named uncertainty

When a signal cannot be read, the sweep says which kind of unreadable it is rather than returning one generic failure — the cases call for different work, and a single message would send someone looking in the wrong place.

The historical study engine underneath began as a fork of bda, by Bruno Dantas, used with his knowledge. The standing watch described here is this project’s own work.