File size: 11,852 Bytes
c10fe82
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
"""Local real-model play: drive the actual runtime against the configured
nemotron (default) + qwen (secondary) endpoints, with a small population so a
run is cheap and easy to read.

Unlike selfplay_sim.py (which injects a deterministic fake completer), this
calls the real OpenAI-compatible endpoints in the config, so it exercises the
true prompt -> model -> validator loop. Use it to watch whether NPCs actually
talk to each other and coordinate.

Usage:
    .venv\\Scripts\\python.exe scripts\\localplay_sim.py --ticks 20 --count 2
"""

from __future__ import annotations

import argparse
import json
from collections import Counter
from dataclasses import replace
from pathlib import Path
import sys

REPO_ROOT = Path(__file__).resolve().parents[1]
SRC_ROOT = REPO_ROOT / "src"
if str(SRC_ROOT) not in sys.path:
    sys.path.insert(0, str(SRC_ROOT))

from world_simulator.api.runtime import GameRuntime  # noqa: E402
from world_simulator.config import NpcConfig, load_game_config  # noqa: E402
from world_simulator.domain import Beast, Vec3  # noqa: E402
from world_simulator.simulation.connectors.deterministic import (  # noqa: E402
    DeterministicWorldSimulator,
)
from world_simulator.simulation.connectors.openai_compatible import (  # noqa: E402
    OpenAICompatibleWorldSimulator,
)
from world_simulator.simulation.connectors.routing import RoutingWorldSimulator  # noqa: E402
from world_simulator.simulation.spawning import create_world  # noqa: E402


def build_runtime(config_path: Path, count: int, *, beast: bool = False) -> GameRuntime:
    config = load_game_config(config_path)
    # count is per-faction: count=2 -> 2 nemotron + 2 qwen = 4 NPCs.
    config = replace(config, npcs=NpcConfig(count=count))
    world = create_world(config)

    if beast:
        # Drop a beast right next to the nemotron cluster (spawns near x=-80)
        # so we can watch whether NPCs notice and react this tick.
        anchor = next(
            (npc.position for npc in world.npcs if npc.country_id == "nemotron"),
            Vec3(x=-80.0, y=0.0, z=0.0),
        )
        world.beasts.append(
            Beast(
                id="beast_test_1",
                position=Vec3(x=anchor.x + 6.0, y=0.0, z=anchor.z),
                health=60.0,
                damage=9.0,
            )
        )
        print(f"INJECTED beast_test_1 near nemotron at x={anchor.x + 6.0:g} z={anchor.z:g}")

    deterministic = DeterministicWorldSimulator()

    # Real endpoints: no chat_completer override -> _complete_with_openai.
    routes = {
        connector_id: OpenAICompatibleWorldSimulator(
            connector_cfg,
            fallback=deterministic,
            connector_id_filter=connector_id,
        )
        for connector_id, connector_cfg in config.secondary_connectors.items()
        if connector_cfg.type == "openai_compatible"
    }
    default = OpenAICompatibleWorldSimulator(
        config.connector,
        fallback=deterministic,
        connector_id_filter=None,
    )
    simulator = RoutingWorldSimulator(routes=routes, default=default)
    return GameRuntime(world=world, simulator=simulator, config=config)


def main() -> None:
    parser = argparse.ArgumentParser(description="Local real-model play.")
    parser.add_argument("--ticks", type=int, default=20)
    parser.add_argument("--count", type=int, default=2, help="NPCs per faction.")
    parser.add_argument(
        "--config", type=Path, default=REPO_ROOT / "config" / "game.modal.local.json"
    )
    parser.add_argument("--beast", action="store_true", help="Inject a beast near the nemotron NPCs.")
    args = parser.parse_args()

    runtime = build_runtime(args.config, args.count, beast=args.beast)
    print(
        f"LOCALPLAY start config={args.config.name} ticks={args.ticks} "
        f"count_per_faction={args.count} simulator={runtime.simulator_name}"
    )

    for _ in range(args.ticks):
        status, _payload = runtime.tick()
        if int(status) != 200:
            print(f"tick failed: {status}")
            break

    ledger_path = runtime._ledger.ledger_path  # noqa: SLF001
    print(f"LOCALPLAY done. ledger={ledger_path}")
    _summarize(ledger_path)


# Words that mark speech as actual discontent / a demand for resources back,
# as opposed to friendly cross-country chatter.
DISCONTENT_KEYWORDS = (
    "give back", "give it back", "give them back", "return", "give us back",
    "stole", "stolen", "steal", "theft", "thief", "thieves", "robbed", "rob",
    "took our", "took from", "our coins", "our food", "our gold", "our wealth",
    "demand", "owe", "repay", "pay back", "or else", "surrender", "what is ours",
    "avenge", "revenge", "raid", "you took", "they took", "back or",
)


def _treasury_country(object_id: object) -> str | None:
    """Country that owns a treasury id like 'treasury_qwen'."""
    if isinstance(object_id, str) and object_id.startswith("treasury_"):
        return object_id[len("treasury_"):]
    return None


def _is_beast(entity_id: object) -> bool:
    return isinstance(entity_id, str) and entity_id.startswith("beast")


def _summarize(ledger_path: Path) -> None:
    verdicts: Counter[str] = Counter()
    actions: Counter[str] = Counter()
    models_by_npc: dict[str, set[str]] = {}
    country_by_npc: dict[str, str] = {}
    fallback_reasons: Counter[str] = Counter()
    speech: list[str] = []
    # Inter-country timeline entries: (tick, kind, text).
    timeline: list[tuple[int, str, str]] = []
    tick = 0

    with ledger_path.open("r", encoding="utf-8") as handle:
        for line in handle:
            if not line.strip():
                continue
            rec = json.loads(line)
            phase = rec.get("phase", "?")
            if phase == "npc_request":
                models_by_npc.setdefault(str(rec.get("npc_id")), set()).add(
                    f"{rec.get('model_profile')}::{rec.get('model')}"
                )
                if rec.get("country_id"):
                    country_by_npc[str(rec.get("npc_id"))] = str(rec.get("country_id"))
                # Every ledger record carries the real tick; the briefing payload
                # is natural-language text (not JSON), so read the tick directly.
                tick = int(rec.get("tick", tick))
            if phase == "npc_response":
                tick = int(rec.get("tick", tick))
                verdict = rec.get("validator_verdict") or {}
                verdicts[str(verdict.get("status"))] += 1
                parsed = rec.get("parsed_action") or {}
                if isinstance(parsed, dict) and parsed.get("action"):
                    actions[str(parsed["action"])] += 1
                    actor = str(parsed.get("npc_id"))
                    actor_country = country_by_npc.get(actor)
                    target = parsed.get("target_npc_id") or parsed.get("target_entity_id")
                    target_country = (
                        country_by_npc.get(str(target)) or _treasury_country(target)
                    )
                    cross = bool(
                        actor_country and target_country and actor_country != target_country
                    )
                    if parsed.get("action") == "speak" and parsed.get("message"):
                        message = str(parsed["message"])
                        speech.append(f"  t{tick} {actor}: \"{message}\"")
                        # A cross-country DEMAND only counts as discontent when the
                        # words actually demand resources back / threaten the rival.
                        if cross and any(k in message.lower() for k in DISCONTENT_KEYWORDS):
                            timeline.append((tick, "DEMAND",
                                f"{actor}({actor_country}) -> {target}({target_country}): "
                                f"\"{message}\""))
                    # Intent signals: what the model *tried*, even if the engine did
                    # not execute it. Display-only; never counted toward the goal.
                    if parsed.get("take") and cross:
                        timeline.append((tick, "intent-steal",
                            f"{actor}({actor_country}) tried to steal from {target}({target_country})"))
                    if parsed.get("action") == "attack" and cross and not _is_beast(target):
                        timeline.append((tick, "intent-attack",
                            f"{actor}({actor_country}) tried to attack {target}({target_country})"))
            if phase == "engine_events":
                for event in rec.get("events") or []:
                    etick = int(event.get("tick", rec.get("tick", tick)))
                    etype = event.get("type")
                    # Engine-confirmed cross-country theft.
                    if etype == "treasury_stolen":
                        thief = str(event.get("actor_id"))
                        thief_country = country_by_npc.get(thief)
                        victim_country = _treasury_country(event.get("object_id"))
                        if (thief_country and victim_country
                                and thief_country != victim_country):
                            timeline.append((etick, "STEAL",
                                f"{thief}({thief_country}) stole from "
                                f"treasury_{victim_country}: {event.get('summary')}"))
                    # Engine-confirmed cross-country NPC attack (beasts excluded).
                    elif etype == "npc_attack":
                        attacker = str(event.get("actor_id"))
                        victim = event.get("target_id")
                        a_country = country_by_npc.get(attacker)
                        v_country = country_by_npc.get(str(victim))
                        if (not _is_beast(victim) and a_country and v_country
                                and a_country != v_country):
                            timeline.append((etick, "ATTACK",
                                f"{attacker}({a_country}) attacked "
                                f"{victim}({v_country}): {event.get('summary')}"))
            if phase == "npc_fallback":
                fallback_reasons[str(rec.get("reason"))] += 1

    print("\n== model per NPC ==")
    for npc_id in sorted(models_by_npc):
        print(f"  {npc_id}: {sorted(models_by_npc[npc_id])}")
    print("\n== validator verdicts ==")
    for status, count in verdicts.most_common():
        print(f"  {status}: {count}")
    print("\n== parsed LLM actions ==")
    for action, count in actions.most_common():
        print(f"  {action}: {count}")
    print("\n== fallback reasons ==")
    for reason, count in fallback_reasons.most_common() or [("(none)", 0)]:
        print(f"  {reason}: {count}")
    print(f"\n== speech ({len(speech)}) ==")
    for line in speech[-30:]:
        print(line)

    timeline.sort(key=lambda item: item[0])
    print(f"\n== INTER-COUNTRY TIMELINE ({len(timeline)}) ==")
    for tk, kind, text in timeline:
        print(f"  t{tk:03d} [{kind}] {text}")
    _goal_verdict(timeline)


def _goal_verdict(timeline: list[tuple[int, str, str]]) -> None:
    """Goal: a STEAL, then a DEMAND at a rival, then a cross-country ATTACK."""
    steal = next((t for t, k, _ in timeline if k == "STEAL"), None)
    demand = next((t for t, k, _ in timeline if k == "DEMAND" and steal is not None and t >= steal), None)
    attack = next(
        (t for t, k, _ in timeline if k == "ATTACK" and demand is not None and t >= demand), None
    )
    print("\n== GOAL CHECK ==")
    print(f"  steal at tick:  {steal}")
    print(f"  discontent at:  {demand} (after steal)")
    print(f"  attack at tick: {attack} (after discontent)")
    met = steal is not None and demand is not None and attack is not None
    print(f"  GOAL {'MET' if met else 'NOT met'}")


if __name__ == "__main__":
    main()