dndwright

A domain-neutral D&D 5e (2024) rules & character-sheet computation engine — formulas as data, not code.

A character sheet is modelled as a directed acyclic computation graph — nodes are values, edges are dependencies, and formulas are data (a JSON-serialisable DSL), not code. Pure Python (pydantic + stdlib), no application or framework coupling: map your own character data in, read computed stats out.

⚠️ Early development (alpha). The API is still moving and may change between minor versions while at 0.x. Usable today — pin a version if you depend on it.

Install

pip install git+https://github.com/sligara7/dndwright.git
# or, for local development:
pip install -e ".[dev]"

Quickstart

from dndwright import evaluate_character

sheet = evaluate_character({
    "ability_scores": {"strength": 8, "dexterity": 14, "constitution": 14,
                       "intelligence": 18, "wisdom": 12, "charisma": 10},
    "class_data": {"class_name": "wizard"},
    "species_data": {"name": "Human", "speed": 30},
    "level": 5,
})

sheet["proficiency_bonus"]    # 3
sheet["ability_modifiers"]    # {"intelligence": 4, "dexterity": 2, ...}
sheet["spellcasting_type"]    # "full_caster"
# ...plus armor_class, hit_points, hit_dice, initiative, saves, features, ...

Lower level — assemble typed inputs and evaluate against the ruleset:

from dndwright import DND_5E_2024_RULESET, assemble_character_inputs, evaluate, apply_modifiers
from dndwright.rules.components import ClassMechanics

inputs   = assemble_character_inputs(class_mechanics=..., ability_scores={...}, level=5)
computed = apply_modifiers(evaluate(DND_5E_2024_RULESET, inputs), inputs)

Command line

Installing the package also installs a dndwright command (no Python required):

dndwright eval character.json          # character JSON → computed sheet (or '-' for stdin)
dndwright graph --format mermaid        # export the computation DAG (mermaid|dot)
dndwright content magic_items           # dump bundled content (omit category to list)
dndwright validate ruleset.json         # check a ruleset (built-in if omitted)

Rolling dice

A self-contained, typed dice engine (dndwright.dice) — deterministic by default:

from dndwright.dice import DiceEngine

eng = DiceEngine(seed=42)               # reproducible (stdlib RNG)
eng.roll("4d6kh3").total                # keep highest 3 of 4
eng.roll("1d20", advantage=True)        # -> ExpressionResult
eng.roll_attack(modifier=5, target_ac=15).is_hit
eng.roll_damage("2d8", is_critical=True)  # crit doubles the dice

# unpredictable production rolls (no NumPy dependency):
import secrets
DiceEngine(rng=secrets.SystemRandom())

Combat rules

Pure, persistence-free 5e combat (dndwright.combat) — state is a frozen value object, every op is (state, input) → (new_state, explanation):

from dndwright.combat import CombatantState, apply_damage, roll_death_save
from dndwright.dice import DiceEngine

s = CombatantState(current_hp=8, max_hp=20, temp_hp=3)
s, applied = apply_damage(s, 10)            # temp HP absorbs first, overkill tracked
s, save = roll_death_save(s, DiceEngine(seed=1))   # nat 20 → 1 HP; 3 fails → dead
s.is_stable, s.is_dead, s.hp_percentage

Your app owns persistence: load a row → call these → write the new state back. The rules never see a database.

Why a computation graph?

Derived character values form a dependency DAG: ability scores → modifiers → proficiency → save DCs / spell slots / AC / HP. dndwright represents that DAG explicitly and stores the formulas as data (FormulaSpec: an op + args), so the rules are inspectable, testable, and serialisable — not buried in imperative code. DND_5E_2024_RULESET is a 138-node graph (incl. damage-defence channels).

Composable — snap mini-graphs onto the ruleset

Items, feats and species traits are themselves tiny graphs. compose() merges a Component's nodes and contributions onto a base ruleset and returns a new, larger Ruleset — the base is never mutated. Because each contribution keeps its target node's id, every existing edge downstream re-derives for free: a set/add/union on one node ripples out to every modifier, save, skill and attack that depends on it.

from dndwright import DND_5E_2024_RULESET, compose, modifier
ring = modifier("ring_of_protection", target="armor_class", amount=1)
rs = compose(DND_5E_2024_RULESET, ring)   # base untouched; AC now aggregates the +1

Re-skin for any setting — theme scaling

The same engine runs sci-fi, modern-warfare, steampunk or cosmic-horror. A ThemeScalingLayer folds three kinds of override onto a ruleset via apply_theme_scaling() (pure, like compose): input_overrides re-baseline a node's default value, lookup_overrides deep-merge into the lookup tables (so plate armour can read AC 19 instead of 18), and flavor_renames relabel terms for display without ever changing a computed value. The graph's shape never changes — only its numbers and names.

from dndwright import DND_5E_2024_RULESET, apply_theme_scaling, get_theme_scaling
rs = apply_theme_scaling(DND_5E_2024_RULESET, get_theme_scaling("sci_fi"))
rs.lookup_tables["armor_base_ac"]["plate"]   # 19 (base is still 18, untouched)

What's inside

API stability

The public API is exactly dndwright.__all__, pinned by tests/test_api_contract.py. Versioning follows SemVer; at 0.x minor versions may break, with every change recorded in CHANGELOG.md. Maintainers: the release process is documented in RELEASING.md.

Credits & license

MIT licensed (see LICENSE). The bundled content and rules tables encode game mechanics derived from the D&D System Reference Document 5.2.1 (English, published May 1, 2025; © Wizards of the Coast, CC-BY-4.0) — source PDF: SRD_CC_v5.2.1.pdf. See NOTICE. Not affiliated with or endorsed by Wizards of the Coast. Contains no PHB/DMG/MM content.