Add Monte Carlo simulator, game metrics and sim CLI (Phase 5)

`dojo sim` plays seeded games over a scenario matrix (variant, players,
Kiai bonus, start position, bot lineup) in parallel, prints a summary per
scenario and optionally writes one CSV row per game. Metrics are replayed
from the event log. Plan updated with the Phase 5 baseline.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

README.md

@@ -32,6 +32,23 @@ uv run dojo board validate --quadrant other.txt --legend other.toml   # check an
 `validate` exits with `0` when the board is fine (warnings allowed), `1` on rule violations,
 `2` when the file is malformed or missing.
 
+## Simulation
+
+Plays many seeded games with bots and prints metrics per scenario. Every combination of the
+listed values is a scenario; game `i` of each scenario uses seed `--seed + i`, and the lineup
+rotates one seat per game. Bots: `random`, `balanced`, `farmer`, `aggressive`.
+
+```sh
+uv run dojo sim                                   # variants A and B, 4 players, 200 games each
+uv run dojo sim --games 1000 --csv games.csv      # more games; one CSV row per game
+uv run dojo sim --variant B --players 3 4 6 8 --start 0 3   # player counts x start position
+uv run dojo sim --kiai 2 1                        # Kiai +2 vs +1
+uv run dojo sim --lineup balanced,random balanced # several lineups (comma-separated seats)
+uv run dojo sim --seed 42 --jobs 1                # other dice; single process
+```
+
+Minutes are estimated as 45 s per turn plus 30 s per Kumite.
+
 ## Development
 
 ```sh

agents/plans/2026-09-25-dojo-board-and-sim.md

@@ -101,16 +101,29 @@ Deliverables:
 
 ### Phase 5 — Simulator and metrics
 
-- [ ] 5.1 Monte Carlo runner over a config matrix; seeded and reproducible.
+- [x] 5.1 Monte Carlo runner over a config matrix; seeded and reproducible. `src/dojo/sim.py`: `Scenario` (variant × players × Kiai bonus × start position × lineup), `matrix`, `run(files, scenarios, games, seed, jobs)`. Game `i` uses seed `base + i` in every scenario (same dice across scenarios); the lineup rotates one seat per game; process pool, results independent of `jobs`.
-- [ ] 5.2 Metrics:
+- [x] 5.2 Metrics (`src/dojo/metrics.py`, computed from the event log; Kumite counts exclude the Exam):
-  - [ ] game length in rounds and estimated minutes (turn and fight duration as parameters);
+  - [x] game length in rounds and estimated minutes (45 s per turn + 30 s per Kumite);
-  - [ ] Kumite per round, upset rate, chain length;
+  - [x] Kumite per round, upset rate, higher-beats-lower rate, longest chain per game (Kumite attacked by the active player in one turn);
-  - [ ] win rate by seat and by bot profile;
+  - [x] win rate by seat and by bot profile (per seat occupied);
-  - [ ] exam attempts per game, exam success rate;
+  - [x] exam attempts per game, exam success rate;
-  - [ ] rounds spent stuck on 14 (variant B deadlock check);
+  - [x] turns started on 14, longest streak (variant B deadlock check);
-  - [ ] stripe income split: squares vs cards vs Kumite;
+  - [x] net stripe income per turn: squares vs cards vs Kumite vs Exam; square income in the player's own quadrant vs elsewhere (home farming);
-  - [ ] lead changes per game, final spread.
+  - [x] lead changes per game (strict leader), final spread.
-- [ ] 5.3 Report output (plain text / CSV).
+- [x] 5.3 Report output: `uv run dojo sim` prints a text summary per scenario; `--csv` writes one row per game (`src/dojo/report.py`). Commands in `README.md`.
+- Baseline (1,000 games per variant, 4 players, lineup balanced/farmer/aggressive, seed 0), for Phase 6:
+
+  | Metric | A | B |
+  |---|---|---|
+  | Rounds mean (p10–p90) | 18.4 (14–23) | 13.1 (9–17) |
+  | Minutes mean (p10–p90) | 66 (50–85) | 46 (33–62) |
+  | Kumite per round | 1.25 | 1.16 |
+  | Upsets / higher beats lower | 29% / 55% | 27% / 52% |
+  | Exam attempts per game, pass rate | 1.26, 79% | 1.22, 82% |
+  | Turns on 14 per game, longest streak | 1.26, 4 | 1.73, 7 |
+  | Income per turn: square (home / away), card, Kumite | +0.52 (0.38 / 0.14), 0.00, +0.04 | +0.54 (0.39 / 0.14), 0.00, +0.22 |
+  | Lead changes, final spread | 5.6, 9.3 | 3.5, 11.5 |
+  | Wins by seat | 23–27% | 23–27% |
 
 ### Phase 6 — Balance experiments
 
@@ -147,14 +160,14 @@ Deliverables:
 ### Open
 
 - Q2 Board asymmetry: keep the quadrant as source of truth, or later switch to the full board as source of truth with a symmetry *report* (warnings) instead of generation? *(Deferred by the developer.)*
-- Q4 Target net income per turn — derived from 1–2 h and ~15–20 turns per player; confirm with first simulations.
+- Q4 Target net income per turn — derived from 1–2 h and ~15–20 turns per player; confirm with first simulations. Baseline measured net income ≈ 0.56 (A) / 0.75 (B) stripes per turn.
 
 ### Risks
 
 - Movement enumeration: self-avoiding walks of length up to 12 can explode in open rooms. Keep rooms small; prune by outcome (destination + ordered opponents crossed + Dojo entry).
 - Engine speed ~1,400 turns/s (random and heuristic bots, real board). If Phase 5 needs more: precompute neighbour lists in `Board`.
 - Simulation conclusions depend on bot quality; compare several profiles before trusting a result.
-- Board draft 1 puts 4 Allenamento in each 3x3 room next to the Safe Room: possible "home farming" (see the Phase 2 review). Measure stripe income by region in the simulator.
+- Board draft 1 puts 4 Allenamento in each 3x3 room next to the Safe Room: possible "home farming" (see the Phase 2 review). Measured (Phase 5 baseline): square income is +0.38/turn in the own quadrant vs +0.14/turn elsewhere; 6.4 decides whether that is a problem.
 
 ## External documentation
 

src/dojo/cli.py

@@ -1,9 +1,14 @@
 import argparse
+import os
 import sys
+from itertools import groupby
 from pathlib import Path
 
 from dojo.board import QUADRANTS, Board, BoardError, load_board
 from dojo.grid import GridFormatError, render_grid
+from dojo.metrics import summarize
+from dojo.report import format_summary, write_games_csv
+from dojo.sim import BOTS, VARIANTS, BoardFiles, matrix, run
 from dojo.validate import Report, validate
 
 DEFAULT_QUADRANT = Path("board/quadrant.txt")
@@ -13,6 +18,8 @@ EXIT_OK = 0
 EXIT_DESIGN_ERRORS = 1
 EXIT_FORMAT_ERRORS = 2
 
+DEFAULT_LINEUP = "balanced,farmer,aggressive"
+
 
 def main(argv: list[str] | None = None) -> int:
     parser = argparse.ArgumentParser(prog="dojo")
@@ -32,12 +39,36 @@ def main(argv: list[str] | None = None) -> int:
         "-o", "--output", type=Path, help="write to this file instead of stdout"
     )
 
+    sim_parser = commands.add_parser(
+        "sim", help="simulate many games and print metrics per scenario"
+    )
+    _add_board_arguments(sim_parser)
+    sim_parser.add_argument("--games", type=int, default=200, help="games per scenario")
+    sim_parser.add_argument("--seed", type=int, default=0, help="seed of the first game")
+    sim_parser.add_argument("--variant", nargs="+", choices=list(VARIANTS), default=["A", "B"])
+    sim_parser.add_argument("--players", nargs="+", type=int, default=[4])
+    sim_parser.add_argument("--kiai", nargs="+", type=int, default=[2], help="Kiai bonus")
+    sim_parser.add_argument("--start", nargs="+", type=int, default=[0], help="start position")
+    sim_parser.add_argument(
+        "--lineup",
+        nargs="+",
+        default=[DEFAULT_LINEUP],
+        help=f"comma-separated bots, repeated to fill the seats; bots: {', '.join(BOTS)}",
+    )
+    sim_parser.add_argument(
+        "--jobs", type=int, default=os.cpu_count() or 1, help="worker processes"
+    )
+    sim_parser.add_argument("--csv", type=Path, help="write one row per game to this file")
+
     args = parser.parse_args(argv)
     legend = _resolve_legend(args.quadrant, args.legend)
     board = _load(args.quadrant, legend)
     if board is None:
         return EXIT_FORMAT_ERRORS
 
+    if args.command == "sim":
+        return _simulate(args, BoardFiles(args.quadrant, legend), parser)
+
     if args.board_command == "validate":
         report = validate(board)
         print(_format_report(args.quadrant, legend, report))
@@ -51,6 +82,27 @@ def main(argv: list[str] | None = None) -> int:
     return EXIT_OK
 
 
+def _simulate(args: argparse.Namespace, files: BoardFiles, parser: argparse.ArgumentParser) -> int:
+    if args.games < 1:
+        parser.error("--games must be at least 1")
+    try:
+        scenarios = matrix(
+            args.variant,
+            args.players,
+            args.kiai,
+            args.start,
+            [tuple(lineup.split(",")) for lineup in args.lineup],
+        )
+    except ValueError as error:
+        parser.error(str(error))
+    games = run(files, scenarios, args.games, seed=args.seed, jobs=max(1, args.jobs))
+    for scenario, rows in groupby(games, key=lambda game: game.scenario):
+        print(format_summary(summarize(scenario, rows)), end="\n\n")
+    if args.csv is not None:
+        write_games_csv(args.csv, games)
+    return EXIT_OK
+
+
 def _add_board_arguments(parser: argparse.ArgumentParser) -> None:
     parser.add_argument("--quadrant", type=Path, default=DEFAULT_QUADRANT)
     parser.add_argument(

src/dojo/metrics.py

@@ -0,0 +1,253 @@
+"""Per-game measurements taken from the event log, and their aggregation over many games."""
+
+import statistics
+from collections import Counter
+from collections.abc import Iterable, Sequence
+from dataclasses import dataclass
+
+from dojo.board import Pos, quadrant_of
+from dojo.events import (
+    Event,
+    ExamTaken,
+    FightKind,
+    GameOver,
+    KumiteFought,
+    Moved,
+    SentToStart,
+    Source,
+    Teleported,
+    TrackChanged,
+    TurnStarted,
+)
+from dojo.rules import RulesConfig
+
+# Table-time estimate (plan, resolved questions).
+TURN_SECONDS = 45
+KUMITE_SECONDS = 30
+
+
+@dataclass(frozen=True)
+class GameMetrics:
+    """One game, flat so it can be written as a CSV row. Kumite counts exclude the Exam."""
+
+    scenario: str
+    seed: int
+    lineup: str
+    winner_seat: int | None
+    winner_bot: str | None
+    rounds: int
+    turns: int
+    minutes: float
+    kumite: int
+    upsets: int
+    expected_wins: int
+    max_chain: int
+    exam_attempts: int
+    exam_passes: int
+    turns_on_14: int
+    longest_14_streak: int
+    income_square: int
+    income_card: int
+    income_kumite: int
+    income_exam: int
+    square_income_home: int
+    square_income_away: int
+    lead_changes: int
+    final_spread: int
+
+
+def measure(
+    events: Sequence[Event],
+    config: RulesConfig,
+    corners: Sequence[int],
+    bots: Sequence[str],
+    *,
+    scenario: str,
+    seed: int,
+) -> GameMetrics:
+    """Replay ``events`` of one game. ``corners[seat]`` and ``bots[seat]`` describe each seat."""
+    players = len(corners)
+    track = [config.start_position] * players
+    position: list[Pos | None] = [None] * players
+    blue_top = config.brown_position - 1
+    streak = [0] * players
+    longest_streak = 0
+    turns_on_14 = 0
+    income = Counter[Source]()
+    home = away = 0
+    kumite = upsets = expected = 0
+    chain = max_chain = 0
+    exam_attempts = exam_passes = 0
+    turns = 0
+    rounds = 0
+    winner: int | None = None
+    leader: int | None = None
+    lead_changes = 0
+    active = 0
+
+    for event in events:
+        match event:
+            case TurnStarted(player=player):
+                turns += 1
+                active = player
+                chain = 0
+                if track[player] == blue_top:
+                    turns_on_14 += 1
+                    streak[player] += 1
+                    longest_streak = max(longest_streak, streak[player])
+                else:
+                    streak[player] = 0
+            case Moved(player=player, path=path) if path:
+                position[player] = path[-1]
+            case Teleported(player=player, destination=destination):
+                position[player] = destination
+            case SentToStart(player=player, square=square):
+                position[player] = square
+            case KumiteFought(exam=False, attacker=attacker, kind=kind):
+                kumite += 1
+                upsets += kind is FightKind.UPSET
+                expected += kind is FightKind.EXPECTED
+                if attacker == active:
+                    chain += 1
+                    max_chain = max(max_chain, chain)
+            case ExamTaken(passed=passed):
+                exam_attempts += 1
+                exam_passes += passed
+            case TrackChanged(player=player, before=before, after=after, source=source):
+                delta = after - before
+                income[source] += delta
+                square = position[player]
+                if source is Source.SQUARE and square is not None:
+                    if quadrant_of(square) == corners[player]:
+                        home += delta
+                    else:
+                        away += delta
+                track[player] = after
+                top = max(track)
+                leaders = [p for p in range(players) if track[p] == top]
+                if len(leaders) == 1 and leaders[0] != leader:
+                    lead_changes += leader is not None
+                    leader = leaders[0]
+            case GameOver(winner=game_winner, rounds=game_rounds):
+                winner, rounds = game_winner, game_rounds
+            case _:
+                pass
+
+    fights = kumite + exam_attempts
+    return GameMetrics(
+        scenario=scenario,
+        seed=seed,
+        lineup=",".join(bots),
+        winner_seat=winner,
+        winner_bot=None if winner is None else bots[winner],
+        rounds=rounds,
+        turns=turns,
+        minutes=round((turns * TURN_SECONDS + fights * KUMITE_SECONDS) / 60, 1),
+        kumite=kumite,
+        upsets=upsets,
+        expected_wins=expected,
+        max_chain=max_chain,
+        exam_attempts=exam_attempts,
+        exam_passes=exam_passes,
+        turns_on_14=turns_on_14,
+        longest_14_streak=longest_streak,
+        income_square=income[Source.SQUARE],
+        income_card=income[Source.CARD],
+        income_kumite=income[Source.KUMITE],
+        income_exam=income[Source.EXAM],
+        square_income_home=home,
+        square_income_away=away,
+        lead_changes=lead_changes,
+        final_spread=max(track) - min(track),
+    )
+
+
+@dataclass(frozen=True)
+class Spread:
+    mean: float
+    p10: float
+    median: float
+    p90: float
+
+    @classmethod
+    def of(cls, values: Sequence[float]) -> Spread:
+        if len(values) < 2:
+            value = float(values[0]) if values else 0.0
+            return cls(value, value, value, value)
+        deciles = statistics.quantiles(values, n=10, method="inclusive")
+        return cls(statistics.fmean(values), deciles[0], statistics.median(values), deciles[-1])
+
+
+@dataclass(frozen=True)
+class Summary:
+    """Aggregate of one scenario. Rates are per turn unless named otherwise."""
+
+    scenario: str
+    games: int
+    finished: float
+    rounds: Spread
+    minutes: Spread
+    kumite_per_round: float
+    upset_rate: float
+    expected_rate: float
+    max_chain: Spread
+    exam_attempts_per_game: float
+    exam_success: float
+    turns_on_14_per_game: float
+    longest_14_streak: int
+    income_per_turn: dict[Source, float]
+    square_income_home_per_turn: float
+    square_income_away_per_turn: float
+    lead_changes: Spread
+    final_spread: Spread
+    win_rate_by_seat: tuple[float, ...]
+    win_rate_by_bot: dict[str, float]
+
+
+def summarize(scenario: str, games: Iterable[GameMetrics]) -> Summary:
+    rows = list(games)
+    if not rows:
+        raise ValueError(f"{scenario}: no games to summarize")
+    turns = sum(r.turns for r in rows) or 1
+    rounds = sum(r.rounds for r in rows) or 1
+    kumite = sum(r.kumite for r in rows)
+    attempts = sum(r.exam_attempts for r in rows)
+    seats = len(rows[0].lineup.split(","))
+
+    appearances = Counter[str]()
+    wins = Counter[str]()
+    for row in rows:
+        appearances.update(row.lineup.split(","))
+        if row.winner_bot is not None:
+            wins[row.winner_bot] += 1
+    seat_wins = Counter(r.winner_seat for r in rows if r.winner_seat is not None)
+    incomes = {
+        Source.SQUARE: sum(r.income_square for r in rows),
+        Source.CARD: sum(r.income_card for r in rows),
+        Source.KUMITE: sum(r.income_kumite for r in rows),
+        Source.EXAM: sum(r.income_exam for r in rows),
+    }
+
+    return Summary(
+        scenario=scenario,
+        games=len(rows),
+        finished=sum(r.winner_seat is not None for r in rows) / len(rows),
+        rounds=Spread.of([r.rounds for r in rows]),
+        minutes=Spread.of([r.minutes for r in rows]),
+        kumite_per_round=kumite / rounds,
+        upset_rate=sum(r.upsets for r in rows) / (kumite or 1),
+        expected_rate=sum(r.expected_wins for r in rows) / (kumite or 1),
+        max_chain=Spread.of([r.max_chain for r in rows]),
+        exam_attempts_per_game=attempts / len(rows),
+        exam_success=sum(r.exam_passes for r in rows) / (attempts or 1),
+        turns_on_14_per_game=sum(r.turns_on_14 for r in rows) / len(rows),
+        longest_14_streak=max(r.longest_14_streak for r in rows),
+        income_per_turn={source: total / turns for source, total in incomes.items()},
+        square_income_home_per_turn=sum(r.square_income_home for r in rows) / turns,
+        square_income_away_per_turn=sum(r.square_income_away for r in rows) / turns,
+        lead_changes=Spread.of([r.lead_changes for r in rows]),
+        final_spread=Spread.of([r.final_spread for r in rows]),
+        win_rate_by_seat=tuple(seat_wins[s] / len(rows) for s in range(seats)),
+        # Wins per seat occupied: an average bot scores 1 / players.
+        win_rate_by_bot={bot: wins[bot] / count for bot, count in sorted(appearances.items())},
+    )

src/dojo/report.py

@@ -0,0 +1,53 @@
+"""Simulation output: a plain-text summary per scenario and one CSV row per game."""
+
+import csv
+import dataclasses
+from collections.abc import Iterable
+from pathlib import Path
+
+from dojo.events import Source
+from dojo.metrics import GameMetrics, Spread, Summary
+
+
+def format_summary(summary: Summary) -> str:
+    income = summary.income_per_turn
+    seats = ", ".join(f"{seat} {rate:.0%}" for seat, rate in enumerate(summary.win_rate_by_seat))
+    bots = ", ".join(f"{bot} {rate:.0%}" for bot, rate in summary.win_rate_by_bot.items())
+    lines = [
+        f"== {summary.scenario}  ({summary.games} games, {summary.finished:.0%} finished)",
+        f"Rounds        {_spread(summary.rounds)}",
+        f"Minutes       {_spread(summary.minutes)}",
+        f"Kumite        {summary.kumite_per_round:.2f}/round; upsets {summary.upset_rate:.0%}, "
+        f"higher beats lower {summary.expected_rate:.0%}",
+        f"Longest chain {_spread(summary.max_chain)}  (Kumite in one turn, per game)",
+        f"Exam          {summary.exam_attempts_per_game:.2f} attempts/game, "
+        f"{summary.exam_success:.0%} passed",
+        f"Stuck on 14   {summary.turns_on_14_per_game:.2f} turns/game, "
+        f"longest streak {summary.longest_14_streak} turns",
+        f"Income/turn   square {income[Source.SQUARE]:+.2f} "
+        f"(home {summary.square_income_home_per_turn:+.2f}, "
+        f"away {summary.square_income_away_per_turn:+.2f}), "
+        f"card {income[Source.CARD]:+.2f}, kumite {income[Source.KUMITE]:+.2f}, "
+        f"exam {income[Source.EXAM]:+.2f}",
+        f"Lead changes  {_spread(summary.lead_changes)}",
+        f"Final spread  {_spread(summary.final_spread)}",
+        f"Wins by seat  {seats}",
+        f"Wins by bot   {bots}",
+    ]
+    return "\n".join(lines)
+
+
+def write_games_csv(path: Path, games: Iterable[GameMetrics]) -> None:
+    fields = [field.name for field in dataclasses.fields(GameMetrics)]
+    with path.open("w", encoding="utf-8", newline="") as file:
+        writer = csv.DictWriter(file, fieldnames=fields)
+        writer.writeheader()
+        for game in games:
+            writer.writerow(dataclasses.asdict(game))
+
+
+def _spread(spread: Spread) -> str:
+    return (
+        f"mean {spread.mean:.1f}  p10 {spread.p10:.1f}  median {spread.median:.1f}  "
+        f"p90 {spread.p90:.1f}"
+    )

src/dojo/sim.py

@@ -0,0 +1,128 @@
+"""Monte Carlo runner: many seeded games over a matrix of rule configurations.
+
+Game ``i`` of every scenario uses seed ``base + i``, so scenarios are compared on the same
+dice. The lineup rotates one seat per game, separating seat effects from bot effects.
+Results do not depend on the number of worker processes.
+"""
+
+import itertools
+import random
+from collections.abc import Sequence
+from concurrent.futures import ProcessPoolExecutor
+from dataclasses import dataclass, replace
+from functools import cache
+from pathlib import Path
+
+from dojo.board import Board, load_board
+from dojo.bots import PROFILES, HeuristicAgent, RandomAgent
+from dojo.engine import Agent, Game, RandomSource, seat_corners
+from dojo.metrics import GameMetrics, measure
+from dojo.rules import VARIANT_A, VARIANT_B, RulesConfig
+
+VARIANTS = {"A": VARIANT_A, "B": VARIANT_B}
+RANDOM_BOT = "random"
+BOTS = (RANDOM_BOT, *PROFILES)
+
+
+@dataclass(frozen=True)
+class Scenario:
+    variant: str
+    players: int
+    kiai_bonus: int
+    start_position: int
+    # Bot names, repeated in order to fill the seats.
+    lineup: tuple[str, ...]
+
+    def __post_init__(self) -> None:
+        if self.variant not in VARIANTS:
+            raise ValueError(f"unknown variant {self.variant!r}: expected one of {list(VARIANTS)}")
+        unknown = [bot for bot in self.lineup if bot not in BOTS]
+        if unknown or not self.lineup:
+            raise ValueError(f"unknown bots {unknown}: expected some of {list(BOTS)}")
+        seat_corners(self.players)
+
+    @property
+    def name(self) -> str:
+        return (
+            f"{self.variant} p{self.players} kiai+{self.kiai_bonus} start{self.start_position} "
+            + ",".join(self.lineup)
+        )
+
+    @property
+    def config(self) -> RulesConfig:
+        return replace(
+            VARIANTS[self.variant],
+            kiai_bonus=self.kiai_bonus,
+            start_position=self.start_position,
+        )
+
+    def seating(self, game: int) -> tuple[str, ...]:
+        seats = [self.lineup[seat % len(self.lineup)] for seat in range(self.players)]
+        shift = game % self.players
+        return tuple(seats[shift:] + seats[:shift])
+
+
+def matrix(
+    variants: Sequence[str],
+    players: Sequence[int],
+    kiai_bonuses: Sequence[int],
+    start_positions: Sequence[int],
+    lineups: Sequence[tuple[str, ...]],
+) -> list[Scenario]:
+    return [
+        Scenario(*axes)
+        for axes in itertools.product(variants, players, kiai_bonuses, start_positions, lineups)
+    ]
+
+
+@dataclass(frozen=True)
+class BoardFiles:
+    quadrant: Path
+    legend: Path | None
+
+
+def make_agent(bot: str, seed: int, seat: int) -> Agent:
+    if bot == RANDOM_BOT:
+        return RandomAgent(random.Random(f"bot:{seed}:{seat}"))
+    return HeuristicAgent(PROFILES[bot])
+
+
+def play(board: Board, scenario: Scenario, index: int, seed: int) -> GameMetrics:
+    bots = scenario.seating(index)
+    agents = [make_agent(bot, seed, seat) for seat, bot in enumerate(bots)]
+    game = Game(board, scenario.config, agents, RandomSource(seed))
+    game.run()
+    return measure(
+        game.events,
+        scenario.config,
+        seat_corners(scenario.players),
+        bots,
+        scenario=scenario.name,
+        seed=seed,
+    )
+
+
+def run(
+    files: BoardFiles,
+    scenarios: Sequence[Scenario],
+    games: int,
+    seed: int = 0,
+    jobs: int = 1,
+) -> list[GameMetrics]:
+    """Play ``games`` games of each scenario; results are ordered by scenario, then game."""
+    tasks = [(s, i, seed + i) for s in scenarios for i in range(games)]
+    if jobs == 1:
+        board = _board(files)
+        return [play(board, *task) for task in tasks]
+    with ProcessPoolExecutor(jobs) as pool:
+        chunk = max(1, len(tasks) // (jobs * 8))
+        return list(pool.map(_play_task, itertools.repeat(files), tasks, chunksize=chunk))
+
+
+def _play_task(files: BoardFiles, task: tuple[Scenario, int, int]) -> GameMetrics:
+    return play(_board(files), *task)
+
+
+@cache
+def _board(files: BoardFiles) -> Board:
+    return load_board(files.quadrant, files.legend)

tests/test_cli.py

@@ -56,3 +56,28 @@ def test_legend_next_to_the_quadrant_is_used_by_default(
 
     assert main(["board", "validate", "--quadrant", str(quadrant)]) == EXIT_FORMAT_ERRORS
     assert "legend links tunnel '9'" in capsys.readouterr().err
+
+
+def test_sim_prints_a_summary_per_scenario_and_writes_games(
+    tmp_path: Path, capsys: pytest.CaptureFixture[str]
+) -> None:
+    games = tmp_path / "games.csv"
+    argv = ["sim", "--games", "2", "--variant", "A", "B", "--players", "3", "--jobs", "1"]
+
+    assert main([*argv, "--csv", str(games)]) == EXIT_OK
+
+    output = capsys.readouterr().out
+    assert "== A p3 kiai+2 start0 balanced,farmer,aggressive  (2 games" in output
+    assert "== B p3 kiai+2 start0" in output
+    rows = games.read_text(encoding="utf-8").splitlines()
+    assert rows[0].startswith("scenario,seed,lineup,winner_seat")
+    assert len(rows) == 1 + 4
+
+
+@pytest.mark.parametrize("bad", [["--lineup", "balanced,cheater"], ["--games", "0"]])
+def test_sim_rejects_bad_arguments(bad: list[str], capsys: pytest.CaptureFixture[str]) -> None:
+    with pytest.raises(SystemExit) as exit_info:
+        main(["sim", *bad])
+
+    assert exit_info.value.code == 2
+    assert "error:" in capsys.readouterr().err

tests/test_metrics.py

@@ -0,0 +1,108 @@
+from dataclasses import replace
+
+import pytest
+
+from dojo.board import Pos
+from dojo.events import (
+    Event,
+    ExamTaken,
+    FightKind,
+    GameOver,
+    KumiteFought,
+    Moved,
+    Source,
+    TrackChanged,
+    TurnStarted,
+)
+from dojo.metrics import KUMITE_SECONDS, TURN_SECONDS, GameMetrics, measure, summarize
+from dojo.rules import VARIANT_A
+
+CORNERS = (0, 1)
+BOTS = ("balanced", "random")
+HOME_0 = Pos(0, 1)  # quadrant 0
+HOME_1 = Pos(0, 14)  # quadrant 1
+
+
+def fought(attacker: int, defender: int, kind: FightKind, *, exam: bool = False) -> KumiteFought:
+    return KumiteFought(attacker, defender, 0, 0, 4, 3, True, kind, exam, cards=())
+
+
+def run(events: list[Event]) -> GameMetrics:
+    return measure(events, VARIANT_A, CORNERS, BOTS, scenario="test", seed=3)
+
+
+def test_measure_counts_fights_income_and_leads() -> None:
+    metrics = run(
+        [
+            TurnStarted(1, 0, 7),
+            Moved(0, (HOME_0,)),
+            TrackChanged(0, 0, 1, Source.SQUARE),
+            TurnStarted(1, 1, 5),
+            Moved(1, (HOME_0,)),
+            TrackChanged(1, 0, 1, Source.SQUARE),
+            fought(1, 0, FightKind.EQUAL),
+            TrackChanged(1, 1, 4, Source.KUMITE),
+            TrackChanged(0, 1, 0, Source.KUMITE),
+            fought(1, 0, FightKind.UPSET),
+            TurnStarted(2, 0, 6),
+            fought(1, 0, FightKind.EXPECTED),  # Player 1 defends: not part of a chain.
+            TrackChanged(0, 0, 5, Source.CARD),
+            fought(0, 1, FightKind.EQUAL, exam=True),
+            ExamTaken(0, 1, passed=True),
+            GameOver(0, 2),
+        ]
+    )
+
+    assert (metrics.kumite, metrics.upsets, metrics.expected_wins) == (3, 1, 1)
+    assert metrics.max_chain == 2
+    assert (metrics.exam_attempts, metrics.exam_passes) == (1, 1)
+    assert (metrics.income_square, metrics.income_kumite, metrics.income_card) == (2, 2, 5)
+    assert (metrics.square_income_home, metrics.square_income_away) == (1, 1)
+    # Leader: 0 (1-0), tie (1-1) keeps it, 1 (0-4), 0 (5-4).
+    assert metrics.lead_changes == 2
+    assert metrics.final_spread == 1
+    assert (metrics.winner_seat, metrics.winner_bot, metrics.rounds) == (0, "balanced", 2)
+    assert metrics.minutes == pytest.approx(round((3 * TURN_SECONDS + 4 * KUMITE_SECONDS) / 60, 1))
+
+
+def test_measure_tracks_turns_started_on_14() -> None:
+    metrics = run(
+        [
+            TrackChanged(0, 0, 14, Source.CARD),
+            TurnStarted(1, 0, 7),
+            TurnStarted(1, 1, 7),
+            TurnStarted(2, 0, 7),
+            TrackChanged(0, 14, 15, Source.KUMITE),
+            TurnStarted(2, 1, 7),
+            TurnStarted(3, 0, 7),
+            TrackChanged(0, 15, 14, Source.SQUARE),
+            TurnStarted(4, 0, 7),
+            GameOver(None, 4),
+        ]
+    )
+
+    assert metrics.turns_on_14 == 3
+    assert metrics.longest_14_streak == 2
+    assert metrics.winner_seat is None
+
+
+def test_summarize_splits_wins_by_seat_and_bot() -> None:
+    base = run([GameOver(None, 10)])
+    rows = [
+        replace(base, winner_seat=0, winner_bot="balanced"),
+        replace(base, lineup="random,balanced", winner_seat=1, winner_bot="balanced"),
+        replace(base, lineup="random,balanced", winner_seat=0, winner_bot="random"),
+        base,
+    ]
+
+    summary = summarize("test", rows)
+
+    assert summary.finished == 0.75
+    assert summary.win_rate_by_seat == (0.5, 0.25)
+    assert summary.win_rate_by_bot == {"balanced": 0.5, "random": 0.25}
+    assert summary.rounds.median == 10
+
+
+def test_summarize_needs_games() -> None:
+    with pytest.raises(ValueError, match="no games"):
+        summarize("test", [])

tests/test_sim.py

@@ -0,0 +1,58 @@
+from pathlib import Path
+
+import pytest
+
+from dojo.sim import BoardFiles, Scenario, matrix, run
+
+BOARD = Path(__file__).parent.parent / "board"
+FILES = BoardFiles(BOARD / "quadrant.txt", BOARD / "legend.toml")
+LINEUP = ("balanced", "random")
+
+
+def test_matrix_crosses_every_axis() -> None:
+    scenarios = matrix(["A", "B"], [3, 4], [2, 1], [0], [LINEUP])
+
+    assert len(scenarios) == 8
+    assert scenarios[0].name == "A p3 kiai+2 start0 balanced,random"
+    assert scenarios[-1].config.kiai_bonus == 1
+    assert scenarios[-1].config.brown_requires_kumite
+
+
+def test_lineup_fills_the_seats_and_rotates_each_game() -> None:
+    scenario = Scenario("A", 3, 2, 0, LINEUP)
+
+    assert scenario.seating(0) == ("balanced", "random", "balanced")
+    assert scenario.seating(1) == ("random", "balanced", "balanced")
+    assert scenario.seating(3) == scenario.seating(0)
+
+
+@pytest.mark.parametrize(
+    ("variant", "players", "lineup", "message"),
+    [
+        ("C", 4, LINEUP, "unknown variant"),
+        ("A", 4, ("balanced", "cheater"), "unknown bots"),
+        ("A", 4, (), "unknown bots"),
+        ("A", 9, LINEUP, "supported range"),
+    ],
+)
+def test_invalid_scenarios_are_rejected(
+    variant: str, players: int, lineup: tuple[str, ...], message: str
+) -> None:
+    with pytest.raises(ValueError, match=message):
+        Scenario(variant, players, 2, 0, lineup)
+
+
+def test_runs_are_reproducible_and_independent_of_worker_count() -> None:
+    scenarios = matrix(["A", "B"], [3], [2], [0], [LINEUP])
+
+    serial = run(FILES, scenarios, games=2, seed=5, jobs=1)
+    parallel = run(FILES, scenarios, games=2, seed=5, jobs=2)
+
+    assert serial == parallel
+    assert [(g.scenario, g.seed) for g in serial] == [
+        (scenarios[0].name, 5),
+        (scenarios[0].name, 6),
+        (scenarios[1].name, 5),
+        (scenarios[1].name, 6),
+    ]
+    assert serial[1].lineup == "random,balanced,balanced"