Repository navigation
Headless balance simulation: seeded trials over a scenario file in enzan-sim #11
Description
Goal
Measure balance without a screen: run many seeded trials of the core's mechanics natively and report the distributions, so a designer can see what the numbers do before anyone plays.
Scope
enzan-sim, a binary over enzan-core:
- Input: a scenario file (JSON) describing the data to simulate — rarities, an affix pool, a drop table, a progression rule — plus the number of trials and the base seed. Reading and parsing the file live in the binary; the core receives checked types only.
- Trials: trial
iusesRng::stream(base_seed, i), so any single trial can be replayed on its own and the run does not depend on how trials are ordered or split. - Measurements (first set):
- loot: how often each rarity and each affix appears, the distribution of the affix count, and of each affix's value (min / mean / percentiles);
- progression: the level reached after a given amount of experience, and the experience needed to reach given levels.
- Output: a summary on stdout, as JSON (
--json) or a readable table.
Rules
| Rule | Owner |
|---|---|
| the same scenario and seed give the same report, natively and wherever the binary runs | the binary's run loop (uses only the core's deterministic draws) |
trial i is reproducible on its own from (base_seed, i) |
the trial |
| the binary holds no mechanics: every number comes from a core function | the binary's entry point |
Done when
- A sample scenario under
examples/runs and prints a report. - Tests cover: the report is the same for the same seed, a trial replays alone, and a scenario with an error (e.g. a duplicate affix id) is reported with the core's error.
just checkis green.
Out of scope
The balance lab in the browser, the TypeScript binding, a shared data crate for other front ends, plotting.
Activity