dsh-eval-guard 0.2.3 → 0.3.1

package/eval/arena/bomberman/workspace/engine/engine.py

@@ -0,0 +1,256 @@
+"""Bomb Arena v0.1. Pure deterministic public rules engine; Python 3.10+."""
+from __future__ import annotations
+from collections import deque
+from dataclasses import dataclass, asdict
+import copy
+import random
+import math
+from fractions import Fraction
+
+VERSION = 'bomb-arena-0.5-public-state'
+DIRS = {'U': (0,-1), 'D': (0,1), 'L': (-1,0), 'R': (1,0), 'S': (0,0)}
+
+@dataclass(frozen=True)
+class Config:
+    size: int = 13
+    players: int = 2
+    max_turns: int = 300
+    shrink_start: int = 150
+    shrink_every: int = 20
+    fuse: int = 4
+    radius: int = 3
+    capacity: int = 2
+    flame_turns: int = 2
+
+def config_for(n):
+    if not 2 <= n <= 10: raise ValueError('players must be between 2 and 10')
+    if n == 2: return Config()
+    if n == 4: return Config(17,4,450,250,25)
+    size = 2*math.ceil(math.sqrt(14*n))+1
+    start = 150+25*(n-2)
+    every = 20 if n <= 3 else 25
+    return Config(size,n,start+every*(size//2)+25,start,every)
+
+def make_map(seed, cfg):
+    n = cfg.size
+    board = [['.' for _ in range(n)] for _ in range(n)]
+    for y in range(n):
+        for x in range(n):
+            if x in (0,n-1) or y in (0,n-1) or (x%2 == 0 and y%2 == 0):
+                board[y][x] = '#'
+    rng = random.Random(seed)
+    if cfg.players == 2: spawns = [(1,1),(n-2,n-2)]
+    elif cfg.players == 4: spawns = [(1,1),(n-2,1),(n-2,n-2),(1,n-2)]
+    else:
+        # Equally spaced along the perimeter's odd-coordinate cells. Corner/edge
+        # positions need seat rotation across matches; do not claim exact symmetry.
+        ring = ([(x,1) for x in range(1,n-2,2)] + [(n-2,y) for y in range(1,n-2,2)]
+                + [(x,n-2) for x in range(n-2,1,-2)] + [(1,y) for y in range(n-2,1,-2)])
+        offset = rng.randrange(len(ring))
+        spawns = [ring[(offset+(i*len(ring)+cfg.players//2)//cfg.players)%len(ring)] for i in range(cfg.players)]
+    safe = {(x,y) for sx,sy in spawns for y in range(n) for x in range(n)
+            if abs(x-sx)+abs(y-sy) <= 2}
+    used = set()
+    for y in range(1,n-1):
+        for x in range(1,n-1):
+            if (x,y) in used or board[y][x] == '#': continue
+            orbit = {(x,y)}
+            if cfg.players in (2,4): orbit.add((n-1-x,n-1-y))
+            if cfg.players == 4: orbit |= {(n-1-y,x),(y,n-1-x)}
+            used |= orbit
+            crate = rng.random() < .42
+            if crate and not (orbit & safe):
+                for xx,yy in orbit: board[yy][xx] = '+'
+    return board, spawns
+
+
+def resolve_moves(origins, targets, allow_swap=True):
+    """Alive players only, terrain already applied. No player-order priority."""
+    if len(set(origins))!=len(origins):raise ValueError('exclusive mode requires unique starting cells')
+    result=list(targets)
+    counts={p:targets.count(p) for p in targets}
+    for i,p in enumerate(targets):
+        if counts[p]>1:result[i]=origins[i]
+    if not allow_swap:
+        for i in range(len(origins)):
+            for j in range(i):
+                if result[i]==origins[j] and result[j]==origins[i]:result[i],result[j]=origins[i],origins[j]
+    while True:
+        stopped={origins[i] for i in range(len(origins)) if result[i]==origins[i]}
+        updated=[origins[i] if result[i] in stopped else result[i] for i in range(len(origins))]
+        if updated==result:break
+        result=updated
+    assert len(set(result))==len(result)
+    return result
+
+class Game:
+    move_mode="swap"
+    def __init__(self, seed=0, cfg=None):
+        self.cfg = cfg or Config()
+        self.seed, self.turn, self.next_bomb = seed, 0, 0
+        self.board, spawns = make_map(seed, self.cfg)
+        self.players = [dict(x=x,y=y,alive=True,death_turn=None,reason='',bombs=0,crates=0)
+                        for x,y in spawns]
+        self.bombs = [] # id, owner, x, y, explode_at (absolute zero-based action turn)
+        self.flames = {} # (x,y) -> last lethal action turn, inclusive
+        self.events = []
+        self.flame_sources = {}
+        self.kill_credit = [Fraction(0) for _ in self.players]
+
+    def clone(self): return copy.deepcopy(self)
+    def over(self):
+        return self.turn >= self.cfg.max_turns or sum(p['alive'] for p in self.players) <= 1
+
+    def _kill(self, i, reason):
+        p = self.players[i]
+        if p['alive']:
+            p.update(alive=False, death_turn=self.turn, reason=reason)
+            self.events.append(dict(type='death',player=i,reason=reason,turn=self.turn))
+            if reason=='blast':
+                owners=sorted(owner for owner,expiry in self.flame_sources.get((p['x'],p['y']),{}).items() if expiry>=self.turn and owner!=i)
+                if owners:
+                    share=Fraction(1,len(owners))
+                    for owner in owners:self.kill_credit[owner]+=share
+                    self.events.append(dict(type='kill_credit',victim=i,owners=owners,share=float(share)))
+
+    def shrink_layer(self, t=None):
+        t = self.turn if t is None else t
+        return 0 if t < self.cfg.shrink_start else 1+(t-self.cfg.shrink_start)//self.cfg.shrink_every
+
+    def step(self, actions, forfeits=None):
+        if self.over(): raise RuntimeError('game already finished')
+        if len(actions) != self.cfg.players: raise ValueError('one action per player')
+        self.events = []
+        t, n = self.turn, self.cfg.size
+        # 1. Expire flames, simultaneous forfeits, then deterministic shrink.
+        self.flames = {k:v for k,v in self.flames.items() if v >= t}
+        for i in forfeits or []: self._kill(i, 'forfeit')
+        layer = self.shrink_layer()
+        if t >= self.cfg.shrink_start and (t-self.cfg.shrink_start)%self.cfg.shrink_every == 0:
+            for y in range(n):
+                for x in range(n):
+                    if min(x,y,n-1-x,n-1-y) <= layer:
+                        self.board[y][x] = '#'
+                        self.flames.pop((x,y), None)
+            self.bombs = [b for b in self.bombs if self.board[b['y']][b['x']] != '#']
+            for i,p in enumerate(self.players):
+                if self.board[p['y']][p['x']] == '#': self._kill(i, 'shrink')
+            self.events.append(dict(type='shrink',layer=layer))
+        # 2. Place at the pre-move position. Co-located simultaneous requests:
+        # exactly one requester succeeds; >=2 requesters all fail (no seat priority).
+        occupied = {(b['x'],b['y']) for b in self.bombs}
+        requests = {}
+        for i,(p,a) in enumerate(zip(self.players, actions)):
+            if not p['alive']: continue
+            d,place = a
+            if d not in DIRS or place not in (0,1): raise ValueError('invalid action')
+            cell = (p['x'],p['y'])
+            count = sum(b['owner']==i for b in self.bombs)
+            if place and cell not in occupied and count < self.cfg.capacity:
+                requests.setdefault(cell,[]).append(i)
+        for (x,y), ids in sorted(requests.items()):
+            if len(ids) != 1: continue
+            i = ids[0]
+            self.bombs.append(dict(id=self.next_bomb,owner=i,x=x,y=y,explode_at=t+self.cfg.fuse))
+            self.next_bomb += 1
+            self.players[i]['bombs'] += 1
+            self.events.append(dict(type='place',player=i,x=x,y=y))
+            occupied.add((x,y))
+        # Terrain and all destinations are computed before resolving player conflicts.
+        ids=[i for i,p in enumerate(self.players) if p['alive']]
+        origins=[(self.players[i]['x'],self.players[i]['y']) for i in ids];targets=[]
+        for i,old in zip(ids,origins):
+            dx,dy=DIRS[actions[i][0]];x,y=old[0]+dx,old[1]+dy
+            targets.append((x,y) if 0<=x<n and 0<=y<n and self.board[y][x]=='.' and ((x,y)==old or (x,y) not in occupied) else old)
+        destinations=targets if self.move_mode=='overlap' else resolve_moves(origins,targets,self.move_mode=='swap')
+        for i,old,want,dest in zip(ids,origins,targets,destinations):
+            self.players[i]['x'],self.players[i]['y']=dest
+            if want!=old and dest==old:self.events.append(dict(type='blocked',player=i,target=list(want)))
+        # 4. All explosions use ONE pre-explosion crate/bomb snapshot. A hit bomb
+        # triggers immediately AND blocks that ray. A hit crate also blocks it.
+        at = {(b['x'],b['y']):b for b in self.bombs}
+        pending = deque(b for b in self.bombs if b['explode_at']<=t or (b['x'],b['y']) in self.flames)
+        exploded, hit, destroyed = set(), set(), {}
+        hit_sources = {}
+        while pending:
+            b = pending.popleft()
+            if b['id'] in exploded: continue
+            exploded.add(b['id']); hit.add((b['x'],b['y']))
+            hit_sources.setdefault((b['x'],b['y']),set()).add(b['owner'])
+            self.events.append(dict(type='explode',player=b['owner'],x=b['x'],y=b['y']))
+            for dx,dy in list(DIRS.values())[:4]:
+                for r in range(1,self.cfg.radius+1):
+                    x,y = b['x']+dx*r,b['y']+dy*r
+                    if not (0<=x<n and 0<=y<n) or self.board[y][x]=='#': break
+                    hit.add((x,y))
+                    hit_sources.setdefault((x,y),set()).add(b['owner'])
+                    if self.board[y][x]=='+':
+                        destroyed.setdefault((x,y),set()).add(b['owner']); break
+                    if (x,y) in at:
+                        pending.append(at[x,y]); break
+        self.bombs = [b for b in self.bombs if b['id'] not in exploded]
+        for (x,y), owners in destroyed.items():
+            self.board[y][x]='.'
+            # A shared crate is credited to every hitting player, display only.
+            for i in owners: self.players[i]['crates'] += 1
+        for cell in hit: self.flames[cell] = max(self.flames.get(cell,-1),t+self.cfg.flame_turns-1)
+        for cell,owners in hit_sources.items():
+            provenance=self.flame_sources.setdefault(cell,{})
+            for owner in owners:provenance[owner]=max(provenance.get(owner,-1),t+self.cfg.flame_turns-1)
+        # 5. Kill on old or newly created flames. A dead owner's bombs remain.
+        for i,p in enumerate(self.players):
+            if (p['x'],p['y']) in self.flames: self._kill(i,'blast')
+        self.turn += 1
+        return self.snapshot()
+
+    def survival_ranks(self):
+        stamps = [self.cfg.max_turns+1 if p['alive'] else p['death_turn'] for p in self.players]
+        return [1+sum(v>s for v in stamps) for s in stamps]
+
+    def score_parts(self):
+        ranks = self.survival_ranks()
+        survival = [Fraction(self.cfg.players-r)-Fraction(ranks.count(r)-1,2) for r in ranks]
+        kills = [k/2 for k in self.kill_credit]
+        return survival, kills, [s+k for s,k in zip(survival,kills)]
+
+    def ranks(self):
+        points = self.score_parts()[2]
+        return [1+sum(other>p for other in points) for p in points]
+
+    def snapshot(self):
+        return dict(turn=self.turn,board=[''.join(r) for r in self.board],
+                    players=copy.deepcopy(self.players),bombs=copy.deepcopy(self.bombs),
+                    flames=[[x,y,e] for (x,y),e in sorted(self.flames.items())],events=copy.deepcopy(self.events))
+
+    def header(self, me):
+        c = self.cfg
+        return f'INIT {c.size} {c.players} {me} {c.max_turns} {c.fuse} {c.radius} {c.capacity} {c.flame_turns} {c.shrink_start} {c.shrink_every}\n'
+
+    def state_text(self):
+        lines = [f'TURN {self.turn}','BOARD']+[''.join(r) for r in self.board]
+        for i,p in enumerate(self.players):
+            lines.append(f"P {i} {p['x']} {p['y']} {int(p['alive'])}")
+        lines.append(f'BOMBS {len(self.bombs)}')
+        for b in self.bombs: lines.append(f"B {b['owner']} {b['x']} {b['y']} {b['explode_at']}")
+        lines.append(f'FLAMES {len(self.flames)}')
+        for (x,y),expiry in sorted(self.flames.items()): lines.append(f'F {x} {y} {expiry}')
+        lines.append(f'STATUS {len(self.players)}')
+        for i,p in enumerate(self.players):
+            k=self.kill_credit[i]
+            death=-1 if p['alive'] else p['death_turn']
+            lines.append(f'S {i} {death} {k.numerator} {k.denominator}')
+        sources=[(x,y,owner,expiry) for (x,y),owners in sorted(self.flame_sources.items())
+                 if (x,y) in self.flames and self.flames[x,y]>=self.turn
+                 for owner,expiry in sorted(owners.items()) if expiry>=self.turn]
+        lines.append(f'SOURCES {len(sources)}')
+        for x,y,owner,expiry in sources:lines.append(f'O {x} {y} {owner} {expiry}')
+        return '\n'.join(lines+['END',''])
+
+    def result(self):
+        survival, kills, points = self.score_parts()
+        return dict(kill_credit=list(map(float,self.kill_credit)),
+                    survival_ranks=self.survival_ranks(),survival_points=list(map(float,survival)),
+                    kill_points=list(map(float,kills)),points=list(map(float,points)),
+                    points_exact=list(map(str,points)),kill_credit_exact=list(map(str,self.kill_credit)),
+                    seed=self.seed,config=asdict(self.cfg),turns=self.turn,ranks=self.ranks(),
+                    survivors=[i for i,p in enumerate(self.players) if p['alive']],players=copy.deepcopy(self.players))

package/eval/arena/bomberman/workspace/engine/verify_replay.py

@@ -0,0 +1,24 @@
+"""Recompute a JSON replay from its seed/actions and compare every frame."""
+import argparse,json
+from pathlib import Path
+from engine import Game,Config,VERSION
+
+def verify(path):
+    data=json.loads(Path(path).read_text(encoding='utf-8'))
+    if data['version']!=VERSION:raise ValueError('Replay requires its original engine version')
+    g=Game(data['result']['seed'],Config(**data['result']['config']))
+    assert g.snapshot()==data['frames'][0],'initial frame differs'
+    for expected in data['frames'][1:]:
+        assert expected['input_text']==g.state_text(),f'turn {g.turn}, input differs'
+        observed=g.step(expected['actions'],expected.get('forfeits',[]))
+        for key,value in observed.items():assert value==expected[key],f"turn {g.turn}, {key} differs"
+    assert g.over(),'replay ends before the game is complete'
+    assert g.turn==data['result']['turns'],'turn count differs'
+    assert g.ranks()==data['result']['ranks'],'ranks differ'
+    assert g.result()['kill_credit']==data['result']['kill_credit'],'kill credit differs'
+    for field in ('survival_ranks','survival_points','kill_points','points','points_exact','kill_credit_exact'):
+        assert g.result()[field]==data['result'][field],f'{field} differs'
+    return dict(path=str(path),frames=len(data['frames']),verified=True)
+if __name__=='__main__':
+    ap=argparse.ArgumentParser();ap.add_argument('files',nargs='+');args=ap.parse_args()
+    for file in args.files:print(json.dumps(verify(file),ensure_ascii=False))

package/eval/arena/bomberman/workspace/run.cmd

@@ -0,0 +1,28 @@
+@echo off
+setlocal
+set "BOMB_RUNTIME="
+if defined BOMB_PYTHON if exist "%BOMB_PYTHON%" set "BOMB_RUNTIME=%BOMB_PYTHON%"
+if not defined BOMB_RUNTIME if exist "%~dp0.runtime\python.exe" set "BOMB_RUNTIME=%~dp0.runtime\python.exe"
+if not defined BOMB_RUNTIME if exist "%USERPROFILE%\.cache\codex-runtimes\codex-primary-runtime\dependencies\python\python.exe" set "BOMB_RUNTIME=%USERPROFILE%\.cache\codex-runtimes\codex-primary-runtime\dependencies\python\python.exe"
+if defined BOMB_RUNTIME goto found
+python3 -c "import sys; sys.exit(0 if sys.version_info >= (3,10) else 1)" >nul 2>nul
+if not errorlevel 1 set "BOMB_RUNTIME=python3"
+if defined BOMB_RUNTIME goto found
+python -c "import sys; sys.exit(0 if sys.version_info >= (3,10) else 1)" >nul 2>nul
+if not errorlevel 1 set "BOMB_RUNTIME=python"
+if not defined BOMB_RUNTIME (
+  echo Python 3.10+ not found. Set BOMB_PYTHON to the full executable path.
+  exit /b 1
+)
+:found
+call "%BOMB_RUNTIME%" -c "import sys; sys.exit(0 if sys.version_info >= (3,10) else 1)"
+if errorlevel 1 exit /b 1
+pushd "%~dp0"
+if "%~1"=="" (
+  call "%BOMB_RUNTIME%" -X utf8 clock.py
+) else (
+  call "%BOMB_RUNTIME%" -X utf8 %*
+)
+set "BOMB_EXIT=%errorlevel%"
+popd
+exit /b %BOMB_EXIT%

package/eval/arena/bomberman/workspace/run.sh

@@ -0,0 +1,16 @@
+#!/bin/sh
+set -eu
+cd "$(CDPATH= cd -- "$(dirname -- "$0")" && pwd)"
+if [ -n "${BOMB_PYTHON:-}" ]; then
+    runtime=$BOMB_PYTHON
+elif [ -x .runtime/bin/python3 ]; then
+    runtime=.runtime/bin/python3
+elif command -v python3 >/dev/null 2>&1; then
+    runtime=$(command -v python3)
+else
+    printf '%s\n' 'Python 3.10+ not found. Report the missing runtime to the organizer, or set BOMB_PYTHON to an existing Python path.' >&2
+    exit 1
+fi
+"$runtime" -c 'import sys; sys.exit(0 if sys.version_info >= (3,10) else 1)'
+if [ "$#" -eq 0 ]; then set -- clock.py; fi
+exec "$runtime" -X utf8 "$@"

package/eval/arena/bomberman/workspace/viewer/index.html

@@ -0,0 +1,42 @@
+<!doctype html>
+<html lang="zh-CN"><head><meta charset="utf-8"><meta name="viewport" content="width=device-width,initial-scale=1"><title>炸弹人 · BOMB ARENA</title>
+<style>
+:root{color-scheme:dark;--muted:#93a3b9;--line:#293347;--accent:#c6ff61}*{box-sizing:border-box}body{margin:0;background:#0a101b;color:#eef2fb;font:15px 'Segoe UI','Microsoft YaHei',sans-serif}header{display:flex;justify-content:space-between;align-items:center;padding:22px 32px;border-bottom:1px solid var(--line)}h1{font-size:24px;margin:0;letter-spacing:.04em}h1 small{font-size:12px;color:var(--accent);margin-left:14px;letter-spacing:.16em}.meta{color:var(--muted);font-size:13px}.wrap{max-width:1320px;margin:auto;padding:24px 32px;display:grid;grid-template-columns:minmax(400px,1fr) 290px;gap:24px}.arena{background:#101927;border:1px solid var(--line);border-radius:16px;overflow:hidden}.bar{display:flex;justify-content:space-between;align-items:center;padding:17px 20px;border-bottom:1px solid var(--line);font-size:13px;color:var(--muted)}#counter{color:white;font-variant-numeric:tabular-nums}canvas{display:block;width:min(100%,740px);height:auto;margin:10px auto}aside{display:flex;flex-direction:column;gap:14px}.card{background:#121c2b;border:1px solid var(--line);border-radius:12px;padding:16px}.label{font-size:11px;text-transform:uppercase;letter-spacing:.13em;color:var(--muted);margin-bottom:12px}.player{display:grid;grid-template-columns:32px 1fr;gap:11px;align-items:center;padding:13px 0;border-bottom:1px solid #263144}.player:last-child{border-bottom:0;padding-bottom:0}.avatar{width:31px;height:31px;border-radius:9px;display:grid;place-items:center;font-weight:800;color:#0a101b}.name{font-size:14px;overflow:hidden;text-overflow:ellipsis;white-space:nowrap;max-width:190px}.stats{font-size:11px;color:var(--muted);margin-top:4px}.dead{opacity:.42}#status{font-size:23px;font-weight:700;line-height:1.5}#statusSub{color:var(--muted);font-size:12px;line-height:1.8;margin-top:7px}.events{height:140px;overflow:auto;font-size:12px;line-height:1.85;color:#c1cada}.controls{display:flex;flex-wrap:wrap;align-items:center;gap:9px;padding:15px 20px;border-top:1px solid var(--line)}button,select,.file{font:inherit;font-size:12px;border:1px solid #3b485d;border-radius:7px;background:#1c293c;color:#edf3fc;padding:8px 12px;cursor:pointer}button:hover,.file:hover{border-color:var(--accent)}#play{background:var(--accent);border-color:var(--accent);color:#18210a;font-weight:700;min-width:68px}#seek{width:100%;accent-color:var(--accent);margin:0 0 5px}.file input{display:none}.hint{color:var(--muted);font-size:11px;line-height:1.9}.legend{display:flex;gap:14px;flex-wrap:wrap;font-size:11px;color:var(--muted);padding:0 20px 17px}.swatch{display:inline-block;width:9px;height:9px;border-radius:2px;margin-right:5px}footer{max-width:1320px;margin:auto;padding:0 32px 22px;color:#62758f;font-size:11px}.film header,.film .controls,.film footer,.film .hint{display:none}.film .wrap{padding:15px;max-width:1500px;grid-template-columns:minmax(400px,1fr) 270px}.film canvas{width:min(100%,850px)}@media(max-width:820px){.wrap{grid-template-columns:1fr;padding:14px}.film .wrap{grid-template-columns:1fr}aside{display:grid;grid-template-columns:1fr 1fr}header{padding:16px}.events{height:85px}}
+.many #players{display:grid;grid-template-columns:1fr 1fr;gap:8px}.many .player{grid-template-columns:24px minmax(0,1fr);gap:6px;padding:8px 0}.many .avatar{width:24px;height:24px;font-size:12px;border-radius:6px}.many .name{font-size:11px;max-width:100%}.many .stats{font-size:9px;line-height:1.5}.many .wrap{grid-template-columns:minmax(400px,1fr) 340px}.many .events{height:110px}@media(max-width:820px){.many .wrap{grid-template-columns:1fr}.many aside{grid-template-columns:1fr}}
+</style></head><body>
+<header><h1>炸弹人 <small>BOMB ARENA / 01</small></h1><span class="meta" id="meta">公开规则 · 确定性回放</span></header>
+<main class="wrap"><section class="arena"><div class="bar"><span id="mode">加载回放</span><span id="counter">TURN 000 / 000</span></div><canvas id="board" width="780" height="780" aria-label="炸弹人对战棋盘"></canvas><div class="legend"><span><i class="swatch" style="background:#b78a57"></i>可破坏箱子</span><span><i class="swatch" style="background:#ffcb58"></i>火焰</span><span>炸弹数字 = 距引爆行动轮数</span></div><div class="controls"><input id="seek" aria-label="回合进度" type="range" min="0" max="1" value="0"><button id="play">播放</button><button id="prev">上一步</button><button id="next">下一步</button><button id="reset">重播</button><select id="speed" aria-label="播放速度"><option value="2">2 回合/秒</option><option value="5" selected>5 回合/秒</option><option value="10">10 回合/秒</option><option value="20">20 回合/秒</option></select><button id="film">录制模式 F</button><button id="full">全屏</button><label class="file">导入 JSON<input id="file" type="file" accept=".json"></label></div></section>
+<aside><div class="card"><div class="label">PLAYERS / 选手</div><div id="players"></div></div><div class="card"><div class="label">MATCH / 战况</div><div id="status">等待开局</div><div id="statusSub"></div></div><div class="card"><div class="label">EVENTS / 回合事件</div><div id="events" class="events"></div></div><div class="card"><div class="label">ACTIONS / 当前步动作</div><div id="actions" class="events"></div></div><div class="hint">空格：播放 / 暂停　← →：逐步<br>F：切换录制模式　R：重播<br>单局分＝生存分＋0.5×击杀份额；同分并列。</div></aside></main><footer>无外部依赖 · 回放中不运行选手代码</footer>
+<script>
+let data=/*REPLAY_DATA*/null,index=0,playing=false,last=0,acc=0;
+const $=s=>document.getElementById(s),ctx=$('board').getContext('2d'),colors=['#62d5ff','#ff7b8d','#c7a2ff','#c6ef66','#ffb65c','#69efd1','#f88dda','#eee2a8','#81a3ff','#f6f7fb'];
+function rounded(x,y,w,h,r,fill){ctx.fillStyle=fill;ctx.beginPath();ctx.roundRect(x,y,w,h,r);ctx.fill()}
+function load(d){if(!d||!Array.isArray(d.frames)||!d.frames.length||!d.result)throw Error('不是有效的回放');data=d;document.body.classList.toggle('many',d.result.config.players>4);index=0;playing=false;acc=0;$('seek').max=d.frames.length-1;$('meta').textContent=`${d.version} · SEED ${d.result.seed}`;const requested=Number(new URLSearchParams(location.hash.slice(1)).get('turn'));if(Number.isFinite(requested)&&requested>0){const found=d.frames.findIndex(f=>f.turn>=requested);index=found<0?d.frames.length-1:found;}draw();}
+function draw(){ctx.clearRect(0,0,780,780);if(!data){ctx.fillStyle='#9aa9be';ctx.font='20px sans-serif';ctx.textAlign='center';ctx.fillText('导入回放 JSON 开始观看',390,390);return;}
+const f=data.frames[index],cfg=data.result.config,n=cfg.size,c=720/n,ox=30,oy=30,flame=new Set(f.flames.map(p=>p[0]+','+p[1]));
+for(let y=0;y<n;y++)for(let x=0;x<n;x++){let xx=ox+x*c,yy=oy+y*c,ch=f.board[y][x];rounded(xx+1,yy+1,c-2,c-2,3,ch==='#'?'#344157':(x+y)%2?'#172436':'#1a293d');if(ch==='#'){ctx.fillStyle='#425169';ctx.fillRect(xx+4,yy+4,c-8,3)}if(ch==='+'){rounded(xx+4,yy+4,c-8,c-8,3,'#987347');ctx.strokeStyle='#c39b66';ctx.lineWidth=2;ctx.strokeRect(xx+8,yy+8,c-16,c-16);ctx.beginPath();ctx.moveTo(xx+8,yy+8);ctx.lineTo(xx+c-8,yy+c-8);ctx.stroke();}if(flame.has(x+','+y)){rounded(xx+2,yy+2,c-4,c-4,5,'#d98432');rounded(xx+c*.21,yy+c*.12,c*.58,c*.76,6,'#ffcd61');rounded(xx+c*.36,yy+c*.29,c*.28,c*.42,4,'#fff5b5')}}
+for(const b of f.bombs){let xx=ox+(b.x+.5)*c,yy=oy+(b.y+.5)*c;ctx.beginPath();ctx.arc(xx,yy,c*.34,0,Math.PI*2);ctx.fillStyle='#080d16';ctx.fill();ctx.strokeStyle=colors[b.owner];ctx.lineWidth=3;ctx.stroke();ctx.fillStyle='#fff';ctx.textAlign='center';ctx.textBaseline='middle';ctx.font=`bold ${c*.34}px sans-serif`;ctx.fillText(Math.max(0,b.explode_at-f.turn),xx,yy)}
+f.players.forEach((p,i)=>{if(!p.alive)return;let xx=ox+(p.x+.5)*c,yy=oy+(p.y+.5)*c;let group=f.players.map((q,j)=>q.alive&&q.x===p.x&&q.y===p.y?j:-1).filter(j=>j>=0),shared=group.length>1;if(shared){let angle=2*Math.PI*group.indexOf(i)/group.length;xx+=Math.cos(angle)*c*.24;yy+=Math.sin(angle)*c*.24}let r=shared?c*.18:c*.32;ctx.shadowBlur=10;ctx.shadowColor=colors[i];rounded(xx-r,yy-r,r*2,r*2,Math.max(4,r*.4),colors[i]);ctx.shadowBlur=0;ctx.fillStyle='#142135';ctx.font=`bold ${c*(shared?.23:.34)}px sans-serif`;ctx.textAlign='center';ctx.textBaseline='middle';ctx.fillText(String.fromCharCode(65+i),xx,yy)});
+$('actions').replaceChildren();
+if(index===0){$('actions').textContent='初始状态，尚未执行动作';}
+else {
+ const prev=data.frames[index-1], req=f.requested_actions||f.actions||[];
+ prev.players.forEach((p,i)=>{if(!p.alive)return;let line=document.createElement('div'),a=req[i],after=f.players[i];
+ line.textContent=`${String.fromCharCode(65+i)}：${a?({U:'上',D:'下',L:'左',R:'右',S:'等待'}[a[0]]+(a[1]?'＋放弹':'')):'无有效响应'}；(${p.x},${p.y}) → (${after.x},${after.y})`;
+ $('actions').append(line);});
+ for(const d of f.diagnostics||[]){let line=document.createElement('div');line.textContent=`${String.fromCharCode(65+d.player)}：${d.error}`;$('actions').append(line);}
+}
+$('counter').textContent=`TURN ${String(f.turn).padStart(3,'0')} / ${cfg.max_turns}`;$('mode').textContent=`${cfg.players===2?'双人对决':cfg.players+' 人混战'} · ${n} × ${n}`;$('seek').value=index;$('play').textContent=playing?'暂停':'播放';
+$('players').replaceChildren();f.players.forEach((p,i)=>{let div=document.createElement('div');div.className='player'+(p.alive?'':' dead');let av=document.createElement('div');av.className='avatar';av.style.background=colors[i];av.textContent=String.fromCharCode(65+i);let info=document.createElement('div'),name=document.createElement('div'),stat=document.createElement('div');name.className='name';name.textContent=data.result.names[i].split(/[\\/]/).pop();stat.className='stats';let active=f.bombs.filter(b=>b.owner===i).length;stat.textContent=p.alive?`存活 · 炸弹 ${active}/${cfg.capacity} · 破箱 ${p.crates}`:`第 ${p.death_turn} 回合淘汰 · ${p.reason}`;if(index===data.frames.length-1&&data.result.kill_credit){stat.textContent+=` · 击杀份额 ${Number(data.result.kill_credit[i]).toFixed(2)}`;if(data.result.points_exact)stat.textContent+=` · 生存分 ${data.result.survival_points[i]}＋击杀分 ${Number(data.result.kill_points[i]).toFixed(2)}＝${data.result.points_exact[i]} 分 · 第 ${data.result.ranks[i]} 名`;}info.append(name,stat);div.append(av,info);$('players').append(div)});
+let alive=f.players.filter(p=>p.alive).length,finished=index===data.frames.length-1;
+$('status').textContent=finished?(data.result.ranks.filter(r=>r===1).length===1?'胜负已定':'并列结束'):`${alive} 人存活`;
+if(finished){$('statusSub').textContent=data.result.ranks.map((r,i)=>`${String.fromCharCode(65+i)}：第 ${r} 名${data.result.points_exact?'（'+data.result.points_exact[i]+'分）':''}`).join('　')}else{let next=f.turn<cfg.shrink_start?cfg.shrink_start:cfg.shrink_start+Math.ceil((f.turn-cfg.shrink_start)/cfg.shrink_every)*cfg.shrink_every;$('statusSub').textContent=`下一次缩圈：第 ${next} 回合\n剩余 ${Math.max(0,next-f.turn)} 回合`}
+if(data.result.valid_for_scoring===false){$('status').textContent='无效对局 · 不计成绩';$('statusSub').textContent='存在启动故障，名次仅供诊断。';}
+$('events').replaceChildren();let events=[];for(let k=Math.max(0,index-5);k<=index;k++)for(let e of data.frames[k].events||[])events.push([Math.max(0,data.frames[k].turn-1),e]);for(let [t,e] of events.slice(-10).reverse()){let div=document.createElement('div'),who=e.player===undefined?'':String.fromCharCode(65+e.player);div.textContent=`${t} · `+({place:`${who} 放置炸弹`,explode:`${who} 炸弹引爆`,death:`${who} 淘汰（${e.reason}）`,shrink:`边界收缩至第 ${e.layer} 层`,blocked:`${who} 移动受阻，目标 (${(e.target||[]).join(',')})`,kill_credit:`${String.fromCharCode(65+(e.victim||0))} 被击杀；${(e.owners||[]).map(i=>String.fromCharCode(65+i)).join('、')} 各记 ${Number(e.share||0).toFixed(2)}`}[e.type]||e.type);$('events').append(div)}
+}
+function toggle(){if(!data)return;if(index>=data.frames.length-1)index=0;playing=!playing;acc=0;draw()}
+$('play').onclick=toggle;$('prev').onclick=()=>{playing=false;index=Math.max(0,index-1);draw()};$('next').onclick=()=>{playing=false;if(data)index=Math.min(data.frames.length-1,index+1);draw()};$('reset').onclick=()=>{index=0;acc=0;playing=true;draw()};$('seek').oninput=()=>{playing=false;index=+$('seek').value;draw()};$('film').onclick=()=>document.body.classList.toggle('film');$('full').onclick=()=>{if(document.fullscreenElement)document.exitFullscreen();else document.documentElement.requestFullscreen()};$('file').onchange=async e=>{try{load(JSON.parse(await e.target.files[0].text()))}catch(err){$('status').textContent='回放读取失败';$('statusSub').textContent=String(err)}};
+document.addEventListener('keydown',e=>{if(e.target.tagName==='INPUT'||e.target.tagName==='SELECT')return;if(e.code==='Space'){e.preventDefault();toggle()}if(e.key==='ArrowLeft')$('prev').click();if(e.key==='ArrowRight')$('next').click();if(e.key.toLowerCase()==='f')$('film').click();if(e.key.toLowerCase()==='r')$('reset').click()});
+function tick(now){if(playing&&data){acc+=Math.min(now-last,250);let step=1000/+$('speed').value;if(acc>=step){let delta=Math.floor(acc/step);index=Math.min(data.frames.length-1,index+delta);acc%=step;if(index===data.frames.length-1)playing=false;draw()}}last=now;requestAnimationFrame(tick)}
+if(data)load(data);else draw();requestAnimationFrame(tick);
+</script></body></html>

package/eval/arena/ctf/bots/README.md

@@ -0,0 +1,18 @@
+# 各模型的交卷代码
+
+都是模型在比赛中独立写出的原样代码，主办方没有改动。基准在 `../workspace/baseline.cpp`。
+
+| 文件 | 模型 | 工具 | 正式赛名次 | 实力值 |
+|---|---|---|---:|---:|
+| `gpt-6-astra.cpp` | GPT-6 Astra | Codex | 1 | 254 |
+| `gpt-6.1-sol.cpp` | GPT-6.1 Sol | Codex | 2 | 252 |
+| `fable-5.1.cpp` | Fable 5.1 | Claude Code | 3 | 213 |
+| `sonnet-5.5.cpp` | Sonnet 5.5 | Claude Code | 4 | 203 |
+| `opus-5.5.cpp` | Opus 5.5 | Claude Code | 5 | 196 |
+| `deepseek-v4.1-flash.cpp` | DeepSeek V4.1 Flash | dsh | 6 | 129 |
+| — | 基准 | — | — | 0 |
+| `qwen-3.8-max.cpp` | Qwen 3.8 Max | Qwen Code | 7 | −21 |
+| `mimo-v2.6-pro.cpp` | MiMo V2.6 Pro | MiMo Code | 8 | −26 |
+| `kimi-k3.cpp` | Kimi K3 | Kimi Code | 9 | −35 |
+| `glm-5.3.cpp` | GLM 5.3 | ZCode | 10 | −92 |
+| `doubao-seed-2.1-turbo.cpp` | 豆包 Seed 2.1 Turbo | trae-agent | 11 | −307 |