dark insight factory

Built on a dark artifact factory (which is just a dark software factory that makes "artifacts", like markdown, for example) instead of software. Same underlying architecture, though.

Sources in, insights out.

Essentially three phases: Sources, Synthesis, and Sharing

Sources

Published meadows as a source

source conversion for assimilation of externally-generated material (blog posts, etc)

longer-form AI unit generated material that how someone might reasonably keep externally-generated notes in their notes

Synthesis

At this point, whether something was externally created or internally doesn't matter too much, and part of synthesis can involve more AI unit generated material

Synthesis of individual blocks

synthesis of Markdown blocks into proposed improvements (targeted... in contrast to the "bigger changes" described below). The type of DeepSearch I described in example - using DeepSearch in Claude Code in a terminal in Obsidian to find a page about a software concept is exceptionally good and could be run in parallel to get all the things linked in one block.

Generative synthesis

Bigger changes (multi-file) that impact existing files can be kicked off as parallel agents with the same kind of agent task -- change review apps as for code, but optimized for documentation change. The type of organization I did for things like Glossary would fit in here.

Background synthesis

These could be agent task -- trigger type -- cron and just continuously making suggestions in parallel agents that can be reviewed in agent task -- change review apps and easily accepted or denied. (Those "change review packets" could just be stuff in the UI... the user doesn't have to be aware of the workings)

Continuous background checks for:

Sharing

For technical people with access to the whole graph, could all be shared in git. For normal people (or a broader. subsetted audience), could be a Meadow