A grimdark clan-survival game for the terminal. Start with a handful of villagers and a fog-black map. Send people out; the world reveals itself tile by tile through scarcity, story, and violence. Grow a hearth into a war-band into a people, or bury them.
Think A Dark Room that grows a spine into King of Dragon Pass, rendered in glyphs.
Status: v0.26.0. SP 8 (the long game) is the front: a hearth that walks out comes back as a rival band, the clan is asked once what it is (the answer shapes the ending), and surplus hands can now raise the works - a palisade, a smokehouse, a shrine - permanent and modest, bought with hands the forage line wanted. Playable start to finish, and deliberately small: one map, four jobs, twenty seasons. Combat arrived with SP 6 as raid violence; the deeper war tiers of the spine below are designed but not built. See
roadmap.mdfor what each slice has to prove before the next one starts, and for the standing rule that no slice ships until a year of it has been played.
Read the unknown at cost. Then commit resources you cannot take back.
That is the whole game, at every scale it reaches. Early it is this tile is forest, paid for with hands you needed elsewhere, committed as a season of foraging you cannot undo. Later it is Stonefold fields forty spears and no bows, and the commitment is a war-stack.
The mechanical form is that fog is a property of a fact, not of a tile. Everything knowable has a value, an age, and a price to look again. Terrain never goes stale. A neighbour's grain store goes stale in a season, and their intent goes stale faster. You are always acting on a picture that is partly out of date, and the game is choosing which part you can afford to be wrong about.
It already pays. Measured over 200 seeds, a clan that never scouts ends with 1.4 people; one that scouts ends with 6.5; one that also reads the seasons ends with 7.8.
Time is seasonal, four turns to a year, and winter never negotiates. The tension is always the same shape: not enough hands, not enough food, and the dark is full of things you have not scouted yet. Each season you split a finite clan four ways:
- Forage brings in food, scaled by the season and by the ground you have walked. Each known tile supports only so many foragers; hands beyond that come back with nothing. Winter yields nothing anywhere, because there is nothing out there to find.
- Scout costs you the hands that went, and how many went decides what they bring home. Two can cover ground: they walk into the dark and learn what is out there, and the clan can work it thinly. A third means the party can stop and survey somewhere, which is what lifts a tile to the whole crew its terrain will carry. Scouting is the only thing that raises the ceiling on how much labour can be spent on food at all. The party comes home with an account of where it went, what is there, and whether it was worth the walk. Ground the clan leans on thins out and ground it rests comes back, and a survey only records what was true the season it was made, so a clan that stops looking is soon planning its seasons around a wood that is not what it remembers.
- Tend slows the rot in the store, and can never quite stop it.
- Militia meets a raid when one comes. Combat is single-stack with stated odds, and the margin grades the cost: a lost raid buries the clan's dead, a rout scatters the band far enough that its camp shows on the map. The untyped count is a spear line, and other lines can stand beside it: what the wall is made of changes what it holds.
Then the season resolves, the world asks you something with no clean answer, and you live with it. Children eat and cannot work. Everyone eats regardless.
When the store cannot cover the winter, you decide who goes short. Equal shares, the workers first, or the children first. There is no right answer: an even split is measurably worse for survival, because spreading a shortfall pushes every hearth into the starvation threshold at once. What it buys is that nobody was wronged. The clan is kin groups, not a number, and the hearth you feed last remembers being fed last.
Events are TOML, and they have a memory. A choice can write to a counter that persists for the whole run, and a later event can require not merely that something happened but that you answered it a particular way.
Overrule the elder in your first year and it runs, over years, toward a man who stops arguing and starts arranging. Defer to him, at a cost you pay that evening, and it runs somewhere else. Nothing announces this and no meter is shown. The powerful moments are rare because they are hard to reach, never because a die came up short.
Eighty-seven entries so far, keyed on the season, the ground, the hearths, how much the clan knows and how long since it checked. A clan that fed strangers in a year it could not afford to is remembered by somebody three days' walk away; a clan that has buried enough people answers questions differently.
The engine is a pure Python library with no I/O, no rendering, and no dependencies. The TUI is a thin, shed-able skin over it.
src/hearthfall/engine/ pure logic, stdlib only, fully tested
src/hearthfall/tui/ Textual skin; the engine does not know it exists
src/hearthfall/data/ TOML content: events, tallies, terrain
tests/ the engine is tested; the skin is not
You can drive a full game from a Python REPL with zero terminal. Every random draw goes through one seeded, injectable RNG, so a seed reproduces a run exactly. Both properties are enforced by tests, not by good intentions, and so is the engine's freedom from dependencies.
The language choice was re-opened against Rust in August 2026 and confirmed; spec.md §8 has
the reasoning and the measurements. What a stricter type system would have bought is bought
directly instead: pyright runs in strict mode over the engine and gates CI at zero errors.
uv venv && uv pip install -e .
.venv/bin/hearthfall
.venv/bin/hearthfall --seed 42 # replay a run exactlyHit Ctrl+P to open the command palette, where you can set standing orders, advance the season, change your glyph tier (ascii/unicode/nerd), and save or load the game.
- Python 3.14+
- Textual (frontend only; the engine needs nothing)
./run_tests.shspec.md: the contract. Read it before changing semantics.roadmap.md: the sub-projects, their kill gates, and what playing each one actually revealed.patchnotes.md: release notes.
MIT. See LICENSE.
if any of this is useful to you and you'd like to chip in:
- liberapay · liberapay.com/bdkl
- bitcoin
bc1qkge6zr45tzqfwfmvma2ylumt6mg7wlwmhr05yv