Spaces:
Running
Running
File size: 32,306 Bytes
ff293b1 ee21104 ff293b1 ee21104 ff293b1 ee21104 ff293b1 ee21104 ff293b1 ee21104 ff293b1 ee21104 ff293b1 ee21104 ff293b1 ee21104 ff293b1 ee21104 ff293b1 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 | # Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
#
# This source code is licensed under the BSD-style license found in the
# LICENSE file in the root directory of this source tree.
"""
GhostExec simulated world, agent step (Phases 2–3), and reward (Phase 4).
Scenario payloads load from scenarios/*.json. Observations are plain-text briefings.
Invalid actions return a structured error in observation metadata without raising.
Rewards aggregate conflict / relationship / task scores and log each step to outputs/logs/.
"""
from __future__ import annotations
import json
import os
from datetime import datetime, timedelta, timezone
from pathlib import Path
from typing import Any
from uuid import uuid4
from openenv.core.env_server.interfaces import Environment
from openenv.core.env_server.types import State
try:
from ..models import (
Contact,
Email,
GhostexecAction,
GhostexecObservation,
Meeting,
Mood,
RewardBreakdown,
Task,
TaskStatus,
WorldState,
)
except ImportError:
from models import (
Contact,
Email,
GhostexecAction,
GhostexecObservation,
Meeting,
Mood,
RewardBreakdown,
Task,
TaskStatus,
WorldState,
)
try:
from . import reward as _reward
except ImportError:
try:
from server import reward as _reward
except ImportError:
import reward as _reward # type: ignore[no-redef]
_PRIORITY_RANK: dict[str, int] = {"critical": 0, "high": 1, "normal": 2, "low": 3}
_REL_DISPLAY: dict[str, str] = {
"board_member": "Board",
"spouse": "Spouse",
"investor": "Investor",
"direct_report": "Direct report",
"client": "Client",
"friend": "Friend",
"team_member": "Team",
}
_INVALID_ACTION_REWARD = -0.25
_DEFAULT_STEP_REWARD = 0.0
_MOOD_ORDER: tuple[Mood, ...] = ("furious", "angry", "annoyed", "neutral", "happy")
def _default_scenario_path() -> Path:
return Path(__file__).resolve().parent.parent / "scenarios" / "phase2_core.json"
def _parse_dt(value: str) -> datetime:
if value.endswith("Z"):
return datetime.fromisoformat(value[:-1]).replace(tzinfo=timezone.utc)
dt = datetime.fromisoformat(value)
if dt.tzinfo is None:
return dt.replace(tzinfo=timezone.utc)
return dt
def _meeting_end(m: Meeting) -> datetime:
start = _parse_dt(m.start)
return start + timedelta(minutes=m.duration_minutes)
def _windows_overlap(a_start: datetime, a_end: datetime, b_start: datetime, b_end: datetime) -> bool:
return a_start < b_end and b_start < a_end
class GhostexecEnvironment(Environment):
"""Inbox, calendar, contacts, tasks, actions, briefings, and Phase 4 rewards."""
SUPPORTS_CONCURRENT_SESSIONS: bool = True
def __init__(
self,
scenario_path: str | Path | None = None,
schema_drift_events_path: str | Path | None = None,
reward_mode: str | None = None,
) -> None:
self._scenario_path = Path(scenario_path) if scenario_path else _default_scenario_path()
self._drift_events_path = (
Path(schema_drift_events_path) if schema_drift_events_path is not None else None
)
self._drift_events: list[dict[str, Any]] = []
if self._drift_events_path and self._drift_events_path.is_file():
drift_raw = json.loads(self._drift_events_path.read_text(encoding="utf-8"))
self._drift_events = list(drift_raw.get("events", []))
self._reply_relationship_suppressed: set[str] = set()
self._reward_log_path = (
Path(__file__).resolve().parent.parent / "outputs" / "logs" / "episode_rewards.jsonl"
)
self._world: WorldState | None = None
self._base_stress: int = 0
self._state = State(episode_id=str(uuid4()), step_count=0)
self._last_step_ok: bool = True
self._last_step_error: str | None = None
self._last_step_detail: str = ""
self._last_reward_breakdown: RewardBreakdown | None = None
self._reward_mode = (reward_mode or os.getenv("GHOSTEXEC_REWARD_MODE", "full")).strip().lower()
if self._reward_mode not in {"full", "base", "shaping"}:
self._reward_mode = "full"
# --- lifecycle ---
def reset(self) -> GhostexecObservation: # type: ignore[override]
self._world = self.load_world_from_json(self._scenario_path)
self._base_stress = self._world.stress
self._rebuild_conflict_list()
self._state = State(episode_id=str(uuid4()), step_count=0)
self._last_step_ok = True
self._last_step_error = None
self._last_step_detail = "Episode started."
self._reply_relationship_suppressed.clear()
self._last_reward_breakdown = None
self._ensure_reward_log_dir()
briefing = self.build_briefing_text()
return self._observation_from_briefing(
briefing,
reward=_DEFAULT_STEP_REWARD,
done=False,
reward_breakdown=None,
)
def step(self, action: GhostexecAction) -> GhostexecObservation: # type: ignore[override]
if self._world is None:
# OpenEnv HTTP uses a new env per request; prime the world so this step still
# runs the requested action (invalid actions get step_ok False, rewards apply).
self.reset()
assert self._world is not None
if not self._world.episode_active:
self._last_step_ok = False
self._last_step_error = "Episode is already finished."
bd = RewardBreakdown(
final=_INVALID_ACTION_REWARD,
invalid_step_adjustment=_INVALID_ACTION_REWARD,
)
self._last_reward_breakdown = bd
return self._observation_from_briefing(
self.build_briefing_text(),
reward=bd.final,
done=True,
reward_breakdown=bd,
)
self._state.step_count += 1
self._maybe_apply_schema_drift_events()
if action.message.strip():
self._world.action_log.append(f"note: {action.message.strip()}")
before = self.world.model_copy(deep=True)
action_ok = self._apply_action(action)
self._apply_post_action_dynamics(action, action_ok=action_ok)
self._rebuild_conflict_list()
episode_done = False
if self._state.step_count >= self._world.max_episode_steps:
episode_done = True
self._world.episode_active = False
self._world.episode_end_reason = self._world.episode_end_reason or "step_limit"
breakdown = _reward.compute_step_reward(
before,
self.world,
action,
action_ok=action_ok,
episode_done=episode_done,
relationship_suppressed_for_email_to=frozenset(self._reply_relationship_suppressed),
reward_mode=self._reward_mode,
step_index=self._state.step_count,
max_steps=self.world.max_episode_steps,
)
self._last_reward_breakdown = breakdown
self._append_reward_log(breakdown, episode_done, action)
briefing = self.build_briefing_text()
return self._observation_from_briefing(
briefing,
reward=breakdown.final,
done=episode_done,
reward_breakdown=breakdown,
)
@property
def state(self) -> State:
return self._state
@property
def world(self) -> WorldState:
if self._world is None:
raise RuntimeError("World not initialised; call reset() first.")
return self._world
# --- Phase 3 briefing (plain text for LLM) ---
def build_briefing_text(self) -> str:
w = self.world
now = _parse_dt(w.simulation_time)
header = now.strftime("=== GHOSTEXEC BRIEFING — %a %d %b %Y %H:%M ===")
unread = self.get_unread_emails_sorted()
email_lines = [
f"- [{e.priority.upper()}] From: {e.sender} ({_REL_DISPLAY.get(e.sender_relationship, e.sender_relationship)}) — "
f'"{e.subject}"\n Preview: {(e.body[:100] + ("…" if len(e.body) > 100 else "")).replace(chr(10), " ")}'
for e in unread[:20]
]
email_block = "\n".join(email_lines) if email_lines else "(none)"
horizon = now + timedelta(hours=4)
conflict_lines: list[str] = []
for row in self.detect_meeting_conflicts():
o0 = _parse_dt(row["overlap_start"])
o1 = _parse_dt(row["overlap_end"])
if o1 <= now or o0 >= horizon:
continue
ma = self._meeting_by_id(row["meeting_a"])
mb = self._meeting_by_id(row["meeting_b"])
if not ma or not mb or ma.cancelled or mb.cancelled:
continue
conflict_lines.append(
f"- {_fmt_meeting_line(ma)} CLASHES WITH -> {_fmt_meeting_line(mb)}"
)
conflict_block = "\n".join(conflict_lines) if conflict_lines else "(none in next 4 hours)"
top_contacts = sorted(w.contacts, key=lambda c: (-c.importance, c.name))[:5]
contact_lines = [
f"- {c.name}: {c.mood.upper()} — {_REL_DISPLAY.get(c.relationship_type, c.relationship_type)}; "
f"prefers {c.communication_preference}"
for c in top_contacts
]
contact_block = "\n".join(contact_lines) if contact_lines else "(none)"
soon = now + timedelta(hours=24)
task_lines: list[str] = []
for t in w.tasks:
if t.status == "done":
continue
dl = _parse_dt(t.deadline)
if dl < now or (now <= dl <= soon):
flag = "OVERDUE" if dl < now else "due soon"
task_lines.append(f"- [{flag}] {t.description} (deadline {t.deadline}, owner {t.owner})")
task_block = "\n".join(task_lines[:15]) if task_lines else "(none)"
remaining = max(0, w.max_episode_steps - self._state.step_count)
parts = [
header,
"",
f"UNREAD EMAILS ({len(unread)} unread):",
email_block,
"",
"CALENDAR CONFLICTS IN NEXT 4 HOURS:",
conflict_block,
"",
"CONTACTS TO WATCH (top 5 by importance):",
contact_block,
"",
"OVERDUE OR DUE-SOON TASKS (next 24h window):",
task_block,
"",
f"EXEC STRESS LEVEL: {w.stress}/100",
f"STEPS REMAINING: {remaining}",
]
if self._last_step_error:
parts += ["", f"LAST ACTION: ERROR — {self._last_step_error}"]
elif self._last_step_detail:
parts += ["", f"LAST ACTION: OK — {self._last_step_detail}"]
return "\n".join(parts)
def _meeting_by_id(self, mid: str) -> Meeting | None:
for m in self.world.meetings:
if m.id == mid:
return m
return None
# --- scenario IO ---
@staticmethod
def load_world_from_json(path: str | Path) -> WorldState:
raw = Path(path).read_text(encoding="utf-8")
data = json.loads(raw)
return WorldState.model_validate(data)
@staticmethod
def world_to_json(world: WorldState) -> str:
return world.model_dump_json()
@staticmethod
def world_from_json(blob: str) -> WorldState:
return WorldState.model_validate_json(blob)
# --- inbox ---
def get_unread_emails_sorted(self) -> list[Email]:
w = self.world
unread = [e for e in w.emails if not e.read]
return sorted(
unread,
key=lambda e: (_PRIORITY_RANK.get(e.priority, 99), e.id),
)
def mark_email_read(self, email_id: str) -> bool:
for i, e in enumerate(self.world.emails):
if e.id == email_id:
self.world.emails[i] = e.model_copy(update={"read": True})
return True
return False
def mark_email_replied(self, email_id: str) -> bool:
for i, e in enumerate(self.world.emails):
if e.id == email_id:
self.world.emails[i] = e.model_copy(update={"read": True, "replied": True})
return True
return False
# --- calendar ---
def detect_meeting_conflicts(self) -> list[dict[str, Any]]:
active = [m for m in self.world.meetings if not m.cancelled]
out: list[dict[str, Any]] = []
for i, a in enumerate(active):
a_start = _parse_dt(a.start)
a_end = _meeting_end(a)
for b in active[i + 1 :]:
b_start = _parse_dt(b.start)
b_end = _meeting_end(b)
if _windows_overlap(a_start, a_end, b_start, b_end):
overlap_start = max(a_start, b_start)
overlap_end = min(a_end, b_end)
out.append(
{
"meeting_a": a.id,
"meeting_b": b.id,
"overlap_start": overlap_start.isoformat(),
"overlap_end": overlap_end.isoformat(),
}
)
return out
def _reschedule_causes_overlap(self, meeting_id: str, new_start_iso: str) -> bool:
idx = next((i for i, m in enumerate(self.world.meetings) if m.id == meeting_id), None)
if idx is None:
return True
cand = self.world.meetings[idx].model_copy(update={"start": new_start_iso})
c_start = _parse_dt(cand.start)
c_end = _meeting_end(cand)
for m in self.world.meetings:
if m.cancelled or m.id == meeting_id:
continue
if _windows_overlap(c_start, c_end, _parse_dt(m.start), _meeting_end(m)):
return True
return False
def reschedule_meeting(self, meeting_id: str, new_start_iso: str) -> bool:
for i, m in enumerate(self.world.meetings):
if m.id == meeting_id and not m.cancelled:
self.world.meetings[i] = m.model_copy(update={"start": new_start_iso})
self._rebuild_conflict_list()
return True
return False
def cancel_meeting(self, meeting_id: str) -> bool:
for i, m in enumerate(self.world.meetings):
if m.id == meeting_id:
self.world.meetings[i] = m.model_copy(update={"cancelled": True})
self._rebuild_conflict_list()
return True
return False
def add_meeting(self, meeting: Meeting) -> None:
self.world.meetings.append(meeting)
self._rebuild_conflict_list()
# --- contacts ---
def get_contact(self, name: str) -> Contact | None:
for c in self.world.contacts:
if c.name == name:
return c
return None
def update_contact_mood(self, name: str, mood: Mood) -> bool:
for i, c in enumerate(self.world.contacts):
if c.name == name:
self.world.contacts[i] = c.model_copy(update={"mood": mood})
return True
return False
# --- tasks ---
def update_task_status(self, task_id: str, status: TaskStatus) -> bool:
for i, t in enumerate(self.world.tasks):
if t.id == task_id:
self.world.tasks[i] = t.model_copy(update={"status": status})
return True
return False
def overdue_tasks_at(self, simulation_iso: str) -> list[Task]:
now = _parse_dt(simulation_iso)
out: list[Task] = []
for t in self.world.tasks:
if t.status in ("done",):
continue
if _parse_dt(t.deadline) < now:
out.append(t)
return out
def set_simulation_time(self, simulation_iso: str) -> None:
self.world.simulation_time = simulation_iso
self._reapply_task_overdue_flags()
self._rebuild_conflict_list()
# --- Phase 3 action execution ---
def _apply_action(self, action: GhostexecAction) -> bool:
self._last_step_ok = True
self._last_step_error = None
self._last_step_detail = ""
at = action.action_type
if at == "do_nothing":
self._last_step_detail = "No action taken."
return True
if at == "reply_email":
if not action.email_id:
return self._fail("reply_email requires email_id")
if not any(e.id == action.email_id for e in self.world.emails):
return self._fail(f"Unknown email_id {action.email_id!r}")
if not action.message_body.strip():
return self._fail("reply_email requires non-empty message_body")
self.mark_email_replied(action.email_id)
self._last_step_detail = f"Replied to email {action.email_id}."
return True
if at == "archive_email":
if not action.email_id:
return self._fail("archive_email requires email_id")
if not self.mark_email_read(action.email_id):
return self._fail(f"Unknown email_id {action.email_id!r}")
self._last_step_detail = f"Archived (read) email {action.email_id}."
return True
if at == "reschedule_meeting":
if not action.meeting_id or not action.new_time:
return self._fail("reschedule_meeting requires meeting_id and new_time")
if not any(m.id == action.meeting_id for m in self.world.meetings):
return self._fail(f"Unknown meeting_id {action.meeting_id!r}")
if self._reschedule_causes_overlap(action.meeting_id, action.new_time):
return self._fail("Target time overlaps another active meeting.")
if not self.reschedule_meeting(action.meeting_id, action.new_time):
return self._fail("Could not reschedule meeting.")
self._last_step_detail = f"Rescheduled {action.meeting_id} to {action.new_time}."
return True
if at == "cancel_meeting":
if not action.meeting_id:
return self._fail("cancel_meeting requires meeting_id")
if not any(m.id == action.meeting_id for m in self.world.meetings):
return self._fail(f"Unknown meeting_id {action.meeting_id!r}")
if not self.cancel_meeting(action.meeting_id):
return self._fail("Could not cancel meeting.")
reason = action.reason.strip() or "(no reason given)"
self._world.action_log.append(f"cancelled {action.meeting_id}: {reason}")
self._last_step_detail = f"Cancelled meeting {action.meeting_id}."
return True
if at == "complete_task":
if not action.task_id:
return self._fail("complete_task requires task_id")
t = next((x for x in self.world.tasks if x.id == action.task_id), None)
if not t:
return self._fail(f"Unknown task_id {action.task_id!r}")
if t.status == "done":
return self._fail("Task is already done.")
self.update_task_status(action.task_id, "done")
self._last_step_detail = f"Completed task {action.task_id}."
return True
if at == "delegate_task":
if not action.task_id or not action.contact_name.strip():
return self._fail("delegate_task requires task_id and contact_name")
if not any(t.id == action.task_id for t in self.world.tasks):
return self._fail(f"Unknown task_id {action.task_id!r}")
if not self.get_contact(action.contact_name.strip()):
return self._fail(f"Unknown contact {action.contact_name.strip()!r}")
for i, t in enumerate(self.world.tasks):
if t.id == action.task_id:
self.world.tasks[i] = t.model_copy(
update={
"delegated_to": action.contact_name.strip(),
"status": "in-progress",
}
)
break
self._last_step_detail = f"Delegated {action.task_id} to {action.contact_name.strip()}."
return True
if at == "send_message":
name = action.contact_name.strip()
if not name:
return self._fail("send_message requires contact_name")
if not self.get_contact(name):
return self._fail(f"Unknown contact {name!r}")
if not action.message_body.strip():
return self._fail("send_message requires non-empty message_body")
self._world.action_log.append(f"message to {name}: {action.message_body.strip()[:500]}")
self._last_step_detail = f"Message sent to {name}."
return True
return self._fail(f"Unsupported action_type {at!r}")
def _fail(self, msg: str) -> bool:
self._last_step_ok = False
self._last_step_error = msg
self._last_step_detail = ""
self._world.action_log.append(f"error: {msg}")
return False
def _advance_clock(self, minutes: int) -> None:
now = _parse_dt(self.world.simulation_time)
new_now = (now + timedelta(minutes=minutes)).replace(tzinfo=None)
self.world.simulation_time = new_now.isoformat(timespec="seconds")
self._reapply_task_overdue_flags()
def _shift_contact_mood(self, name: str, delta: int) -> None:
if delta == 0:
return
c = self.get_contact(name)
if c is None:
return
idx = _MOOD_ORDER.index(c.mood)
next_idx = max(0, min(len(_MOOD_ORDER) - 1, idx + delta))
if next_idx != idx:
self.update_contact_mood(name, _MOOD_ORDER[next_idx])
def _apply_post_action_dynamics(self, action: GhostexecAction, *, action_ok: bool) -> None:
# Step-level world progression adds realistic pressure dynamics while
# remaining deterministic and learnable for policy optimization.
self._advance_clock(minutes=20)
now = _parse_dt(self.world.simulation_time)
if action_ok and action.action_type == "reply_email" and action.email_id:
em = next((e for e in self.world.emails if e.id == action.email_id), None)
if em:
self._shift_contact_mood(em.sender, +1)
if action_ok and action.action_type == "send_message" and action.contact_name:
self._shift_contact_mood(action.contact_name.strip(), +1)
if action_ok and action.action_type == "cancel_meeting" and action.meeting_id:
mtg = next((m for m in self.world.meetings if m.id == action.meeting_id), None)
if mtg:
for attendee in mtg.attendees:
self._shift_contact_mood(attendee, -1)
# Pressure escalation only on idle/invalid behavior to keep
# action-quality separation sharp for learning.
if (not action_ok) or action.action_type == "do_nothing":
critical_pending = [e for e in self.world.emails if e.priority == "critical" and not e.replied]
if critical_pending:
self._shift_contact_mood(critical_pending[0].sender, -1)
# Meetings that have already ended without cancellation and still overlap
# indicate unresolved calendar debt; this increases stress pressure.
unresolved_past_conflicts = 0
for row in self.detect_meeting_conflicts():
overlap_end = _parse_dt(row["overlap_end"])
if overlap_end <= now:
unresolved_past_conflicts += 1
if unresolved_past_conflicts > 0:
self.world.action_log.append(
f"pressure: {unresolved_past_conflicts} unresolved past overlap(s) increased stress pressure."
)
def _ensure_reward_log_dir(self) -> None:
self._reward_log_path.parent.mkdir(parents=True, exist_ok=True)
def _append_reward_log(
self,
breakdown: RewardBreakdown,
episode_done: bool,
action: GhostexecAction,
) -> None:
self._ensure_reward_log_dir()
w = self.world
crit_open = sum(1 for e in w.emails if e.priority == "critical" and not e.replied)
overdue_n = len(self.overdue_tasks_at(w.simulation_time))
line = {
"episode_id": self._state.episode_id,
"step": self._state.step_count,
"action_type": action.action_type,
"step_ok": self._last_step_ok,
"reward": breakdown.final,
"conflict_raw": breakdown.conflict_raw,
"critical_queue_bonus": breakdown.critical_queue_bonus,
"conflict": breakdown.conflict,
"relationship": breakdown.relationship,
"task": breakdown.task,
"weighted_base": breakdown.weighted_base,
"output_scale": breakdown.output_scale,
"shaping_synergy": breakdown.shaping_synergy,
"shaping_tradeoff": breakdown.shaping_tradeoff,
"shaping_potential": breakdown.shaping_potential,
"shaping_scaffold": breakdown.shaping_scaffold,
"shaping_quality": breakdown.shaping_quality,
"shaping_total": breakdown.shaping_total,
"shaping_to_base_ratio": breakdown.shaping_to_base_ratio,
"invalid_step_adjustment": breakdown.invalid_step_adjustment,
"episode_completion_bonus": breakdown.episode_completion_bonus,
"catastrophic_penalty": breakdown.catastrophic_penalty,
"episode_done": episode_done,
"calendar_overlap_pairs": len(self.detect_meeting_conflicts()),
"critical_unreplied": crit_open,
"overdue_tasks": overdue_n,
"reward_mode": self._reward_mode,
}
with self._reward_log_path.open("a", encoding="utf-8") as fh:
fh.write(json.dumps(line) + "\n")
def _maybe_apply_schema_drift_events(self) -> None:
if not self._world or not self._drift_events:
return
step = self._state.step_count
for ev in self._drift_events:
if ev.get("after_step") != step:
continue
if "shift_all_meetings_hours" in ev:
delta = int(ev["shift_all_meetings_hours"])
for i, m in enumerate(self._world.meetings):
new_start = (_parse_dt(m.start) + timedelta(hours=delta)).replace(tzinfo=None)
self._world.meetings[i] = m.model_copy(
update={"start": new_start.isoformat(timespec="seconds")}
)
self._world.action_log.append(
f"schema drift: shifted all meeting starts by {delta:+d} hour(s) (calendar TZ policy)."
)
pref = ev.get("set_contact_preference")
if isinstance(pref, dict):
name = str(pref.get("name", ""))
comm = str(pref.get("communication_preference", "text"))
for i, c in enumerate(self._world.contacts):
if c.name == name:
self._world.contacts[i] = c.model_copy(
update={"communication_preference": comm} # type: ignore[arg-type]
)
break
self._world.action_log.append(
f"schema drift: contact {name!r} now prefers {comm} only (relationship channel change)."
)
td = ev.get("set_task_deadline")
if isinstance(td, dict):
tid = str(td.get("task_id", ""))
dl = str(td.get("deadline", ""))
for i, t in enumerate(self._world.tasks):
if t.id == tid:
self._world.tasks[i] = t.model_copy(update={"deadline": dl})
break
self._world.action_log.append(
f"schema drift: task {tid!r} deadline moved earlier to {dl!r}."
)
for name in ev.get("suppress_reply_relationship_for_senders", []) or []:
self._reply_relationship_suppressed.add(str(name))
self._world.action_log.append(
f"schema drift: replies to emails from {name!r} yield zero relationship score this episode."
)
scm = ev.get("set_contact_mood")
if isinstance(scm, dict):
cname = str(scm.get("name", ""))
mood_raw = str(scm.get("mood", "neutral"))
allowed: tuple[Mood, ...] = ("happy", "neutral", "annoyed", "angry", "furious")
if cname and mood_raw in allowed and self.update_contact_mood(cname, mood_raw):
self._world.action_log.append(
f"schema drift: stakeholder {cname!r} mood is now {mood_raw} (external pressure)."
)
if any(ev.get("after_step") == step for ev in self._drift_events):
self._rebuild_conflict_list()
# --- internals ---
def _reapply_task_overdue_flags(self) -> None:
now = _parse_dt(self.world.simulation_time)
for i, t in enumerate(self.world.tasks):
if t.status == "done":
continue
if _parse_dt(t.deadline) < now and t.status != "overdue":
self.world.tasks[i] = t.model_copy(update={"status": "overdue"})
def _rebuild_conflict_list(self) -> None:
lines: list[str] = []
for row in self.detect_meeting_conflicts():
lines.append(
f"Calendar overlap: {row['meeting_a']} vs {row['meeting_b']} "
f"({row['overlap_start']} – {row['overlap_end']})"
)
for e in self.world.emails:
if e.priority == "critical" and not e.replied:
lines.append(f"Unanswered critical email {e.id}: {e.subject}")
bump = min(35, len(lines) * 2)
self.world.active_conflicts = lines
self.world.stress = min(100, self._base_stress + bump)
def _observation_from_briefing(
self,
briefing: str,
reward: float,
done: bool,
reward_breakdown: RewardBreakdown | None = None,
) -> GhostexecObservation:
w = self.world
unread_sorted = self.get_unread_emails_sorted()
meta: dict[str, Any] = {
"simulation_time": w.simulation_time,
"stress": w.stress,
"unread_email_count": sum(1 for e in w.emails if not e.read),
"calendar_conflict_pairs": len(self.detect_meeting_conflicts()),
"episode_step": self._state.step_count,
"max_episode_steps": w.max_episode_steps,
"episode_active": w.episode_active,
"episode_end_reason": w.episode_end_reason,
"step_ok": self._last_step_ok,
"step_error": self._last_step_error,
"step_detail": self._last_step_detail,
# Compact ids for remote trainers / Colab (briefing stays plain text).
"critical_unreplied_email_ids": [
e.id for e in w.emails if e.priority == "critical" and not e.replied
][:12],
"unread_email_ids": [e.id for e in unread_sorted[:15]],
"overdue_task_ids": [t.id for t in self.overdue_tasks_at(w.simulation_time)][:12],
"active_meeting_ids": [m.id for m in w.meetings if not m.cancelled][:20],
}
if reward_breakdown is not None:
meta["reward_breakdown"] = reward_breakdown.model_dump()
cap = 48_000
text = briefing if len(briefing) <= cap else briefing[: cap - 1] + "…"
return GhostexecObservation(
echoed_message=text,
message_length=len(text),
done=done,
reward=reward,
metadata=meta,
)
def _fmt_meeting_line(m: Meeting) -> str:
st = _parse_dt(m.start)
return f"{st.strftime('%H:%M')}: {m.title} ({m.duration_minutes}min)"
|