Spaces:
Configuration error
Configuration error
File size: 15,560 Bytes
d745a13 b43600a d745a13 b43600a d745a13 b43600a d745a13 b43600a d745a13 b43600a d745a13 b43600a d745a13 b43600a d745a13 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 | """
Utilities for reconstructing an AI decision trace from a logged game session.
The analyzer intentionally works from files already written by older sessions:
prompt_N.json, response_N.json, intermediate tool-call responses, optional
tool follow-up prompts, and tool_executions.json.
"""
from __future__ import annotations
import copy
import json
from datetime import datetime
from pathlib import Path
from typing import Any, Dict, List, Optional
def build_decision_analysis(
source_session: Path,
decision: Dict[str, Any],
action_result: Optional[Dict[str, Any]] = None,
) -> Dict[str, Any]:
"""Build a human-readable decision trace for one replay decision."""
session_dir = _resolve_session_for_decision(source_session, decision)
player_name = str(decision.get("player_name") or "")
request_number = int(decision.get("request_number") or 0)
prompt_doc = _load_prompt(session_dir, player_name, request_number)
response_doc = _load_response(session_dir, player_name, request_number)
if not isinstance(prompt_doc, dict):
prompt_doc = {}
if not isinstance(response_doc, dict):
response_doc = {}
parsed = (
copy.deepcopy(response_doc.get("parsed"))
if isinstance(response_doc, dict) and isinstance(response_doc.get("parsed"), dict)
else copy.deepcopy(decision.get("parsed") or {})
)
prompt = prompt_doc.get("prompt") if isinstance(prompt_doc, dict) else {}
if not isinstance(prompt, dict):
prompt = {}
tool_trace = _load_tool_trace(session_dir, player_name, request_number)
memory_before = copy.deepcopy(prompt.get("memory") or {})
social_context = copy.deepcopy(prompt.get("social_context") or {})
constraints = copy.deepcopy(prompt.get("constraints") or {})
allowed_actions = copy.deepcopy(prompt_doc.get("allowed_actions") or constraints.get("allowed_actions") or [])
compact_state_text = prompt.get("game_state") or ""
compact_state_json = _extract_embedded_json(compact_state_text)
observed_facts = _build_observed_facts(
compact_state_json,
allowed_actions,
prompt.get("task_context") or {},
)
action_type = parsed.get("action_type") or (parsed.get("action") or {}).get("type")
action_parameters = parsed.get("parameters")
if action_parameters is None and isinstance(parsed.get("action"), dict):
action_parameters = parsed["action"].get("parameters")
return {
"available": bool(prompt_doc or response_doc or parsed),
"session": session_dir.name if session_dir else "",
"session_path": str(session_dir) if session_dir else "",
"player_name": player_name,
"request_number": request_number,
"timestamp": (
response_doc.get("timestamp")
or prompt_doc.get("timestamp")
or decision.get("timestamp")
or ""
),
"label": f"{player_name} #{request_number}: {action_type or 'decision'}",
"worldview": {
"task_context": copy.deepcopy(prompt.get("task_context") or {}),
"memory_before": memory_before,
"social_context": social_context,
"constraints": constraints,
"compact_game_state": compact_state_text,
"compact_game_state_json": compact_state_json,
"observed_facts": observed_facts,
"allowed_actions": allowed_actions,
},
"tool_trace": tool_trace,
"thinking": parsed.get("internal_thinking") or "",
"memory_write": parsed.get("note_to_self") or "",
"say_outloud": parsed.get("say_outloud") or "",
"action": {
"type": action_type,
"parameters": action_parameters,
},
"engine_result": copy.deepcopy(action_result or {}),
"raw": {
"prompt": prompt_doc,
"response": response_doc,
},
}
def build_turn_flow(
source_session: Path,
decisions: List[Dict[str, Any]],
) -> List[Dict[str, Any]]:
"""Build lightweight summaries for every decision in the selected turn."""
flow: List[Dict[str, Any]] = []
for item in decisions:
decision = item.get("decision") or {}
action_result = item.get("action_result") or {}
parsed = decision.get("parsed") or {}
action_type = parsed.get("action_type") or (parsed.get("action") or {}).get("type")
response_doc = _load_response(
_resolve_session_for_decision(source_session, decision),
str(decision.get("player_name") or ""),
int(decision.get("request_number") or 0),
)
if isinstance(response_doc.get("parsed"), dict):
parsed = response_doc["parsed"]
action_type = parsed.get("action_type") or (parsed.get("action") or {}).get("type")
flow.append({
"snapshot_index": item.get("snapshot_index"),
"label": item.get("label") or "",
"player_name": decision.get("player_name") or "",
"request_number": decision.get("request_number") or 0,
"action_type": action_type,
"say_outloud": parsed.get("say_outloud") or "",
"memory_write": parsed.get("note_to_self") or "",
"success": action_result.get("success"),
"message": action_result.get("message") or "",
"turn_number": action_result.get("turn_number"),
})
return flow
def _load_tool_trace(session_dir: Path, player_name: str, request_number: int) -> List[Dict[str, Any]]:
intermediate_responses = _load_intermediate_responses(session_dir, player_name, request_number)
followups = _load_tool_followups(session_dir, player_name, request_number)
execution_batches = _load_tool_executions(session_dir)
used_batch_indexes: set[int] = set()
trace = []
for intermediate in intermediate_responses:
iteration = int(intermediate.get("iteration") or 0)
tool_calls = copy.deepcopy(intermediate.get("tool_calls") or [])
followup = next((item for item in followups if int(item.get("iteration") or 0) == iteration), {})
batch_index = _match_tool_execution_batch(
execution_batches,
intermediate,
tool_calls,
used_batch_indexes,
)
batch = execution_batches[batch_index] if batch_index is not None else {}
if batch_index is not None:
used_batch_indexes.add(batch_index)
trace.append({
"iteration": iteration,
"timestamp": intermediate.get("timestamp") or followup.get("timestamp") or batch.get("timestamp") or "",
"tool_calls": tool_calls,
"tool_results_text": followup.get("tool_results") or _format_batch_results(batch),
"execution_batch": batch,
"followup_context_available": bool(followup.get("full_context_sent")),
"full_context_sent": followup.get("full_context_sent") or "",
})
for followup in followups:
iteration = int(followup.get("iteration") or 0)
if any(item["iteration"] == iteration for item in trace):
continue
trace.append({
"iteration": iteration,
"timestamp": followup.get("timestamp") or "",
"tool_calls": [],
"tool_results_text": followup.get("tool_results") or "",
"execution_batch": {},
"followup_context_available": bool(followup.get("full_context_sent")),
"full_context_sent": followup.get("full_context_sent") or "",
})
trace.sort(key=lambda item: item.get("iteration") or 0)
return trace
def _resolve_session_for_decision(source_session: Path, decision: Dict[str, Any]) -> Path:
source_file = decision.get("source_file")
if source_file:
response_path = Path(source_file)
if response_path.exists():
return response_path.parent.parent.parent
return Path(source_session)
def _player_dir(session_dir: Path, player_name: str) -> Optional[Path]:
direct = session_dir / player_name
if direct.exists():
return direct
wanted = player_name.lower()
for child in session_dir.iterdir() if session_dir.exists() else []:
if child.is_dir() and child.name.lower() == wanted:
return child
return None
def _load_prompt(session_dir: Path, player_name: str, request_number: int) -> Dict[str, Any]:
player_dir = _player_dir(session_dir, player_name)
if not player_dir:
return {}
return _read_json(player_dir / "prompts" / f"prompt_{request_number}.json")
def _load_response(session_dir: Path, player_name: str, request_number: int) -> Dict[str, Any]:
player_dir = _player_dir(session_dir, player_name)
if not player_dir:
return {}
return _read_json(player_dir / "responses" / f"response_{request_number}.json")
def _load_intermediate_responses(session_dir: Path, player_name: str, request_number: int) -> List[Dict[str, Any]]:
player_dir = _player_dir(session_dir, player_name)
if not player_dir:
return []
intermediate_dir = player_dir / "responses" / "intermediate"
items = []
for path in sorted(intermediate_dir.glob(f"response_{request_number}_iter*.json")):
data = _read_json(path)
if data:
items.append(data)
return items
def _load_tool_followups(session_dir: Path, player_name: str, request_number: int) -> List[Dict[str, Any]]:
player_dir = _player_dir(session_dir, player_name)
if not player_dir:
return []
iterations_dir = player_dir / "prompts" / "iterations"
items = []
for path in sorted(iterations_dir.glob(f"prompt_{request_number}_iter*.json")):
data = _read_json(path)
if data:
items.append(data)
return items
def _load_tool_executions(session_dir: Path) -> List[Dict[str, Any]]:
data = _read_json(session_dir / "tool_executions.json")
return data if isinstance(data, list) else []
def _match_tool_execution_batch(
batches: List[Dict[str, Any]],
intermediate: Dict[str, Any],
tool_calls: List[Dict[str, Any]],
used_batch_indexes: set[int],
) -> Optional[int]:
expected_names = [str(call.get("name") or "") for call in tool_calls]
intermediate_ts = _parse_timestamp(intermediate.get("timestamp"))
best_index = None
best_delta = None
for index, batch in enumerate(batches):
if index in used_batch_indexes:
continue
batch_names = [str(call.get("name") or "") for call in batch.get("calls") or []]
if expected_names and batch_names[: len(expected_names)] != expected_names:
continue
batch_ts = _parse_timestamp(batch.get("timestamp"))
if intermediate_ts and batch_ts and batch_ts < intermediate_ts:
continue
delta = (
(batch_ts - intermediate_ts).total_seconds()
if intermediate_ts and batch_ts
else float(index)
)
if best_delta is None or delta < best_delta:
best_delta = delta
best_index = index
return best_index
def _format_batch_results(batch: Dict[str, Any]) -> str:
calls = batch.get("calls") or []
if not calls:
return ""
lines = ["=== Tool Results ===\n"]
for call in calls:
lines.append(f"Tool: {call.get('name', '')}")
lines.append(f"Parameters: {json.dumps(call.get('parameters') or {}, indent=2, ensure_ascii=False)}")
if call.get("success", True):
lines.append("Result:")
lines.append(json.dumps(call.get("result"), indent=2, ensure_ascii=False))
else:
lines.append(f"Error: {call.get('error') or ''}")
lines.append("---\n")
return "\n".join(lines)
def _extract_embedded_json(text: str) -> Optional[Dict[str, Any]]:
marker = "JSON:"
if not isinstance(text, str) or marker not in text:
return None
candidate = text.split(marker, 1)[1].strip()
try:
return json.loads(candidate)
except Exception:
return None
def _build_observed_facts(
compact_state: Optional[Dict[str, Any]],
allowed_actions: List[Dict[str, Any]],
task_context: Dict[str, Any],
) -> Dict[str, Any]:
"""Extract the high-signal facts that were visible in compact game_state."""
if not isinstance(compact_state, dict):
return {
"expected_action": _expected_action_from_allowed(allowed_actions),
"prompt_warnings": _prompt_consistency_warnings(allowed_actions, task_context),
}
meta = compact_state.get("meta") or {}
dice = meta.get("dice")
dice_total = sum(dice) if isinstance(dice, list) and all(isinstance(x, (int, float)) for x in dice) else None
current_player = meta.get("curr")
players = compact_state.get("players") or {}
current_player_state = (
copy.deepcopy(players.get(current_player) or {})
if current_player is not None and isinstance(players, dict)
else {}
)
return {
"current_player": current_player,
"phase": meta.get("phase"),
"robber_hex": meta.get("robber"),
"dice": dice,
"dice_total": dice_total,
"expected_action": _expected_action_from_allowed(allowed_actions),
"prompt_warnings": _prompt_consistency_warnings(allowed_actions, task_context),
"current_player_state": current_player_state,
"players": copy.deepcopy(players) if isinstance(players, dict) else {},
}
def _allowed_types(allowed_actions: List[Dict[str, Any]]) -> set[str]:
result = set()
for action in allowed_actions or []:
if isinstance(action, dict):
value = action.get("type")
else:
value = str(action)
if value:
result.add(str(value).lower())
return result
def _expected_action_from_allowed(allowed_actions: List[Dict[str, Any]]) -> str:
allowed = _allowed_types(allowed_actions)
if "roll_dice" in allowed and allowed <= {"roll_dice", "use_dev_card"}:
if "use_dev_card" in allowed:
return "Start the turn: roll dice, or optionally use a development card before rolling."
return "Start the turn: roll dice."
if {"build_settlement", "build_city", "build_road", "trade_propose", "trade_bank", "buy_dev_card", "end_turn"} & allowed:
return "Post-roll actions: build, trade, buy/use development card, or end turn."
if allowed:
return "Allowed now: " + ", ".join(sorted(allowed))
return ""
def _prompt_consistency_warnings(
allowed_actions: List[Dict[str, Any]],
task_context: Dict[str, Any],
) -> List[str]:
allowed = _allowed_types(allowed_actions)
what_happened = str((task_context or {}).get("what_just_happened") or "").lower()
warnings = []
if "roll_dice" in allowed and allowed <= {"roll_dice", "use_dev_card"}:
if "build, trade, or end" in what_happened:
warnings.append(
"The prompt text says build/trade/end, but the allowed actions show this is a pre-roll decision."
)
return warnings
def _parse_timestamp(value: Any) -> Optional[datetime]:
if not value:
return None
try:
return datetime.fromisoformat(str(value))
except Exception:
return None
def _read_json(path: Path) -> Any:
try:
if path.exists():
return json.loads(path.read_text(encoding="utf-8"))
except Exception:
return {}
return {}
|