A suggestion for Legends mode (long)
I’ve been thinking about a possible addition/tool for Dwarf Fortress Legends mode, and I’m curious whether anything similar already exists or whether anyone would be interested in pursuing it.
The core idea is NOT “AI-generated lore” or “AI storytelling.”
Dwarf Fortress already generates enormous amounts of lore, history, causality, relationships, wars, artifacts, dynasties, etc. The problem is not lack of content. The problem is that the content is often difficult for humans to parse emotionally or contextually.
What I’m imagining is more like a dynamic contextual wiki layered over exported Legends data.
The user selects a focal point:
- a fortress,
- a historical figure,
- a battle,
- an artifact,
- a dynasty,
- a ruin,
- a monster,
- etc.
The system then builds a local contextual view around that point:
- upstream causes,
- downstream consequences,
- adjacent important people/places/events,
- recurring references,
- emotionally significant associations,
- and major historical impacts.
Importantly, this would NOT attempt to narrativize the entire world simultaneously. The user drives inquiry recursively.
For example:
You click on Blackglass Tower.
The system summarizes:
- who built it,
- why it became important,
- major wars or deaths associated with it,
- important residents,
- why it was abandoned,
- and what later events connect back to it.
Then maybe the user clicks on the war associated with the tower.
Now the lens recenters around that war:
- causes,
- battles,
- famous deaths,
- refugees,
- political fallout,
- legends,
- survivor stories.
The important part is that the lore itself remains the source of truth. The system is not inventing canon. It is organizing and contextualizing existing simulation history into something more emotionally readable.
I think this could largely be accomplished through weighted graph traversal and significance ranking:
- wars rank highly,
- dynastic collapses rank highly,
- famous deaths rank highly,
- repeated references rank highly,
- random low-impact daily events get deprioritized unless later connected to something meaningful.
Different filters/lenses could also alter weighting:
- military,
- political,
- emotional,
- mythic,
- economic,
- dynastic,
- cultural.
So “Tell me about the Siege of the Purple Pony” would produce a different contextualization than “Tell me the war stories from the Siege of the Purple Pony.”
I also suspect this is more achievable than true procedural storytelling because the system does not need to generate meaning from scratch. It only needs to surface and organize potentially meaningful relationships already present in the simulation.
A lightweight version might not even require an LLM at all. A more advanced version could use an LLM ONLY as a summarization/presentation layer over bounded source facts, not as a lore generator.
The closest comparison I can think of is:
- a semantic wiki,
- a contextual historian,
- or a dynamic mythology browser for DF worlds.
Has anyone attempted something similar already?