Spaces:
Running
Running
File size: 29,743 Bytes
c5cfc73 a363048 c5cfc73 a363048 6f7e1b7 a363048 c5cfc73 8a4babc a363048 c5cfc73 a363048 c5cfc73 a363048 c5cfc73 a363048 c5cfc73 a363048 c5cfc73 6f7e1b7 a363048 af7c75f 8a4babc c5cfc73 a363048 c5cfc73 a363048 c5cfc73 a363048 c5cfc73 a363048 c5cfc73 a363048 c5cfc73 a363048 6f7e1b7 a363048 c5cfc73 a363048 c5cfc73 6f7e1b7 a363048 af7c75f c5cfc73 a363048 c5cfc73 a363048 6f7e1b7 a363048 6f7e1b7 a363048 c5cfc73 a363048 6f7e1b7 a363048 6f7e1b7 a363048 6f7e1b7 a363048 6f7e1b7 c5cfc73 a363048 c5cfc73 a363048 c5cfc73 a363048 c5cfc73 a363048 c5cfc73 a363048 af7c75f a363048 c5cfc73 6f7e1b7 a363048 6f7e1b7 a363048 6f7e1b7 a363048 6f7e1b7 a363048 6f7e1b7 a363048 8a4babc a363048 af7c75f a363048 af7c75f a363048 8a4babc a363048 af7c75f a363048 af7c75f a363048 6f7e1b7 a363048 6f7e1b7 a363048 6f7e1b7 a363048 6f7e1b7 a363048 6f7e1b7 a363048 c5cfc73 a363048 c5cfc73 8a4babc af7c75f a363048 8a4babc a363048 8a4babc a363048 8a4babc a363048 c5cfc73 a363048 8a4babc a363048 8a4babc a363048 c5cfc73 a363048 c5cfc73 a363048 c5cfc73 a363048 c5cfc73 a363048 a2ae67c a363048 c5cfc73 a363048 c5cfc73 a363048 8a4babc a363048 af7c75f 8a4babc a363048 c5cfc73 a363048 c5cfc73 a363048 6f7e1b7 a363048 c5cfc73 a363048 c5cfc73 a363048 c5cfc73 a363048 c5cfc73 af7c75f 8a4babc c5cfc73 a363048 c5cfc73 af7c75f | 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 | """
ESCTR Environment — Core Implementation.
Enterprise Supply Chain & Tax Reconciliation: a stateful environment
where an LLM agent operates as an autonomous financial controller,
using ERP tools to investigate discrepancies, enforce SLA penalties,
and navigate adversarial vendor disputes.
Reward Architecture:
R_total = α·R_outcome + β·R_trajectory − penalties
"""
import json
import os
from dataclasses import asdict
from typing import Any, Optional
from uuid import uuid4
from .models import ESCTRAction, ESCTRObservation, ESCTRState
from .procedural import (
generate_scenario, Scenario, VALID_TASKS, MAX_STEPS,
render_purchase_order, render_invoice, render_sla,
render_shipping_log, render_warehouse_logs,
)
from .graders import grade_task1, grade_task2, grade_task3
# Reward constants
STEP_COST = 0.005
HALLUCINATION_PENALTY = 0.02
# Available tools per task
TASK_TOOLS = {
"procurement_reconciliation": [
"query_database", "read_document", "submit_financial_decision",
],
"sla_enforcement": [
"query_database", "read_document", "submit_financial_decision",
],
"adversarial_auditing": [
"query_database", "read_document", "communicate_vendor", "submit_financial_decision",
],
}
# Database tables per task
AVAILABLE_TABLES = {
"procurement_reconciliation": ["purchase_orders", "invoices"],
"sla_enforcement": ["purchase_orders", "invoices", "shipping_logs", "sla_contracts"],
"adversarial_auditing": ["purchase_orders", "invoices", "shipping_logs", "sla_contracts", "warehouse_logs"],
}
class ESCTREnvironment:
"""Enterprise Supply Chain & Tax Reconciliation Environment."""
def __init__(self):
self._state = ESCTRState(episode_id=str(uuid4()))
self._scenario: Optional[Scenario] = None
self._initialized = False
self._trajectory_reward = 0.0
self._milestones: list = []
self._vendor_negotiation_count = 0
self._settlement_offered = False
self._settlement_rejected = False
self._cited_evidence = False
self._action_trace: list[dict[str, Any]] = []
self._enable_distractors = os.environ.get("ESCTR_ENABLE_DISTRACTORS", "1") != "0"
self._enable_risk_shaping = os.environ.get("ESCTR_ENABLE_RISK_SHAPING", "1") != "0"
def reset(
self,
seed: Optional[int] = None,
episode_id: Optional[str] = None,
task_name: str = "procurement_reconciliation",
**kwargs: Any,
) -> ESCTRObservation:
"""Reset the environment with a new scenario."""
if task_name not in VALID_TASKS:
task_name = "procurement_reconciliation"
actual_seed = seed if seed is not None else 0
scenario = generate_scenario(task_name, actual_seed)
max_steps = MAX_STEPS.get(task_name, 15)
self._state = ESCTRState(
episode_id=episode_id or str(uuid4()),
step_count=0,
task_name=task_name,
seed=actual_seed,
accumulated_reward=0.0,
outcome_submitted=False,
milestones_hit=[],
)
self._scenario = scenario
self._initialized = True
self._trajectory_reward = 0.0
self._milestones = []
self._vendor_negotiation_count = 0
self._settlement_offered = False
self._settlement_rejected = False
self._cited_evidence = False
self._action_trace = []
tools = TASK_TOOLS.get(task_name, [])
tables = AVAILABLE_TABLES.get(task_name, [])
# Build initial briefing
briefing = self._build_briefing(task_name, scenario, tables)
return ESCTRObservation(
done=False,
reward=0.0,
system_response=briefing,
last_action_status="success",
current_step=0,
max_steps=max_steps,
accumulated_reward=0.0,
task_name=task_name,
available_tools=tools,
)
def _build_briefing(self, task_name: str, scenario: Scenario, tables: list) -> str:
"""Generate task-specific initial briefing."""
vendor = scenario.vendor.name
buyer = scenario.buyer.name
inv_num = scenario.invoice.invoice_number
po_num = scenario.purchase_order.po_number
if task_name == "procurement_reconciliation":
return (
f"=== DISCREPANCY ALERT ===\n"
f"A pricing discrepancy has been detected between Purchase Order {po_num} "
f"and Vendor Invoice {inv_num} from {vendor}.\n\n"
f"Your task: Investigate the discrepancy, identify the overcharged line item, "
f"and submit the correct financial adjustment.\n\n"
f"Available databases: {', '.join(tables)}\n"
f"Available tools: query_database, read_document, submit_financial_decision\n\n"
f"Use 'query_database' with {{'table': '<table_name>'}} to explore data.\n"
f"Use 'read_document' with document_id (e.g. '{po_num}' or '{inv_num}') to read full documents.\n"
f"Use 'submit_financial_decision' with adjustment_amount and adjustment_reason when ready."
)
elif task_name == "sla_enforcement":
return (
f"=== PAYMENT DEMAND REVIEW ===\n"
f"Vendor {vendor} has submitted Invoice {inv_num} (ref: {po_num}) "
f"demanding full payment without penalties.\n\n"
f"Intelligence suggests the shipment may have been delivered late. "
f"Your task: Verify delivery timing, review the SLA contract, calculate "
f"any applicable penalties, and submit the correct adjusted payment.\n\n"
f"Available databases: {', '.join(tables)}\n"
f"Available tools: query_database, read_document, submit_financial_decision\n\n"
f"Key steps: Check shipping_logs → Review sla_contracts → Calculate penalty → Submit adjustment."
)
elif task_name == "adversarial_auditing":
return (
f"=== VENDOR DISPUTE ALERT ===\n"
f"Vendor {vendor} has submitted Invoice {inv_num} (ref: {po_num}) "
f"demanding full payment. Shipping records indicate a late delivery.\n\n"
f"⚠ The vendor DISPUTES the late delivery claim. They assert that {buyer}'s "
f"receiving warehouse rejected the initial delivery attempt.\n\n"
f"Your task: Investigate the vendor's claim against internal records, "
f"verify warehouse availability, enforce SLA penalties if warranted, and "
f"handle any settlement offers from the vendor.\n\n"
f"Available databases: {', '.join(tables)}\n"
f"Available tools: query_database, read_document, communicate_vendor, submit_financial_decision\n\n"
f"WARNING: The vendor may attempt to negotiate a reduced penalty. "
f"Verify all claims against internal data before accepting ANY settlement."
)
return "Environment ready."
def step(
self,
action: ESCTRAction,
timeout_s: Optional[float] = None,
**kwargs: Any,
) -> ESCTRObservation:
"""Execute one step in the environment."""
if not self._initialized:
return self._error_obs("Environment not initialized. Call reset() first.", terminal=True)
if self._state.outcome_submitted:
return self._error_obs("Episode already complete. Call reset() for a new episode.", terminal=True)
self._state.step_count += 1
max_steps = MAX_STEPS.get(self._state.task_name, 15)
# Step cost
self._trajectory_reward -= STEP_COST
# Check max steps
if self._state.step_count > max_steps:
return self._finalize("Maximum steps exceeded. Episode terminated.", forced=True)
# Validate tool availability
available = TASK_TOOLS.get(self._state.task_name, [])
if action.action_type not in available:
self._trajectory_reward -= HALLUCINATION_PENALTY
return self._error_obs(
f"Tool '{action.action_type}' is not available for task '{self._state.task_name}'. "
f"Available tools: {', '.join(available)}"
)
# Dispatch
self._action_trace.append(
{
"step": self._state.step_count,
"tool": action.action_type,
"args": {
"query_parameters": action.query_parameters,
"document_id": action.document_id,
"message_content": action.message_content,
"adjustment_amount": action.adjustment_amount,
"adjustment_reason": action.adjustment_reason,
},
}
)
if action.action_type == "query_database":
return self._handle_query(action)
elif action.action_type == "read_document":
return self._handle_read(action)
elif action.action_type == "communicate_vendor":
return self._handle_vendor_comm(action)
elif action.action_type == "submit_financial_decision":
return self._handle_submit(action)
return self._error_obs(f"Unknown action type: {action.action_type}")
# ------------------------------------------------------------------
# Tool handlers
# ------------------------------------------------------------------
def _handle_query(self, action: ESCTRAction) -> ESCTRObservation:
"""Handle database queries."""
params = action.query_parameters or {}
table = params.get("table", "")
available = AVAILABLE_TABLES.get(self._state.task_name, [])
if not table:
self._trajectory_reward -= HALLUCINATION_PENALTY
return self._error_obs(
f"Missing 'table' in query_parameters. Available tables: {', '.join(available)}"
)
if table not in available:
self._trajectory_reward -= HALLUCINATION_PENALTY
return self._error_obs(
f"Table '{table}' not found. Available tables: {', '.join(available)}"
)
scenario = self._scenario
if table == "purchase_orders":
self._add_milestone("retrieved_po")
po = scenario.purchase_order
distractors = (scenario.distractor_purchase_orders or []) if self._enable_distractors else []
summary = (
f"Query result: {1 + len(distractors)} records found in purchase_orders\n\n"
f"[PRIMARY] PO Number: {po.po_number}\n"
f"Date: {po.date}\n"
f"Vendor: {po.vendor.name}\n"
f"Buyer: {po.buyer.name}\n"
f"Total: ${po.total_amount:,.2f}\n"
f"Items: {len(po.line_items)}\n"
)
if distractors:
summary += "\nPossible distractor records:\n"
for row in distractors:
summary += (
f"- {row['po_number']} | Vendor: {row['vendor']} | "
f"Total: ${row['total_amount']:,.2f}\n"
)
summary += (
"\n"
f"Use read_document with document_id='{po.po_number}' for full details."
)
return self._success_obs(summary)
elif table == "invoices":
self._add_milestone("retrieved_invoice")
inv = scenario.invoice
distractors = (scenario.distractor_invoices or []) if self._enable_distractors else []
summary = (
f"Query result: {1 + len(distractors)} records found in invoices\n\n"
f"[PRIMARY] Invoice: {inv.invoice_number}\n"
f"Date: {inv.date}\n"
f"PO Ref: {inv.po_reference}\n"
f"Vendor: {inv.vendor.name}\n"
f"Subtotal: ${inv.subtotal:,.2f}\n"
f"Tax: ${inv.tax_amount:,.2f}\n"
f"Total: ${inv.total:,.2f}\n"
)
if distractors:
summary += "\nPossible distractor records:\n"
for row in distractors:
summary += (
f"- {row['invoice_number']} | PO Ref: {row['po_reference']} | "
f"Vendor: {row['vendor']} | Total: ${row['total']:,.2f}\n"
)
summary += (
"\n"
f"Use read_document with document_id='{inv.invoice_number}' for full details."
)
return self._success_obs(summary)
elif table == "shipping_logs":
self._add_milestone("retrieved_shipping")
log = scenario.shipping_log
if log:
summary = (
f"Query result: 1 record found in shipping_logs\n\n"
f"Tracking: {log.tracking_id}\n"
f"PO Ref: {log.po_reference}\n"
f"Carrier: {log.carrier}\n"
f"Expected Delivery: {log.expected_delivery}\n"
f"Actual Delivery: {log.actual_delivery}\n"
f"Delay: {log.delay_days} day(s)\n"
f"Status: {log.status}\n\n"
f"Use read_document with document_id='{log.tracking_id}' for full log."
)
else:
summary = "Query result: 0 records found in shipping_logs."
return self._success_obs(summary)
elif table == "sla_contracts":
self._add_milestone("retrieved_sla")
sla = scenario.sla_contract
if sla:
summary = (
f"Query result: 1 record found in sla_contracts\n\n"
f"Contract: {sla.contract_id}\n"
f"Vendor: {sla.vendor}\n"
f"Buyer: {sla.buyer}\n"
f"Delivery Terms: {sla.delivery_terms}\n\n"
f"Use read_document with document_id='{sla.contract_id}' for full SLA."
)
else:
summary = "Query result: 0 records found in sla_contracts."
return self._success_obs(summary)
elif table == "warehouse_logs":
self._add_milestone("checked_warehouse")
logs = scenario.warehouse_logs
if logs:
summary = (
f"Query result: {len(logs)} records found in warehouse_logs\n\n"
)
for wl in logs:
summary += (
f"Date: {wl.date} | Dock: {wl.dock_id} | Status: {wl.status.upper()} | "
f"Staff: {wl.staff_on_duty} | Shipments: {wl.shipments_received}\n"
)
summary += (
f"\nAll records show dock status: OPEN with active receiving operations.\n"
f"This contradicts any claim that the warehouse was unavailable."
)
else:
summary = "Query result: 0 records found in warehouse_logs."
return self._success_obs(summary)
return self._error_obs(f"Unknown table: {table}")
def _handle_read(self, action: ESCTRAction) -> ESCTRObservation:
"""Handle document reads."""
doc_id = action.document_id
if not doc_id:
self._trajectory_reward -= HALLUCINATION_PENALTY
return self._error_obs("Missing document_id. Specify the document to read.")
scenario = self._scenario
# Match document_id to known documents
if doc_id == scenario.purchase_order.po_number:
self._add_milestone("retrieved_po")
self._add_milestone("compared_documents")
return self._success_obs(render_purchase_order(scenario.purchase_order))
elif doc_id == scenario.invoice.invoice_number:
self._add_milestone("retrieved_invoice")
self._add_milestone("compared_documents")
return self._success_obs(render_invoice(scenario.invoice))
elif scenario.sla_contract and doc_id == scenario.sla_contract.contract_id:
self._add_milestone("retrieved_sla")
return self._success_obs(render_sla(scenario.sla_contract))
elif scenario.shipping_log and doc_id == scenario.shipping_log.tracking_id:
self._add_milestone("retrieved_shipping")
return self._success_obs(render_shipping_log(scenario.shipping_log))
elif self._enable_distractors and scenario.distractor_documents and doc_id in scenario.distractor_documents:
return self._success_obs(scenario.distractor_documents[doc_id])
else:
self._trajectory_reward -= HALLUCINATION_PENALTY
return self._error_obs(f"Document '{doc_id}' not found in the system.")
def _handle_vendor_comm(self, action: ESCTRAction) -> ESCTRObservation:
"""Handle vendor communication (adversarial negotiation)."""
self._add_milestone("vendor_negotiation")
self._vendor_negotiation_count += 1
msg = (action.message_content or "").lower()
scenario = self._scenario
import random as _rng
_rng.seed(self._state.seed + self._vendor_negotiation_count)
profile = scenario.vendor_honesty_profile or "adversarial"
if self._vendor_negotiation_count == 1:
# First contact: vendor makes their excuse
excuse_pool = {
"hardball": [
"Our legal review confirms no SLA breach occurred and your claim is invalid.",
"Your receiving team rejected delivery; the delay is entirely on your side.",
"We will dispute any deduction as non-compliant with contract terms.",
],
"adversarial": [
"Our records indicate the receiving warehouse rejected the initial delivery attempt due to dock unavailability.",
"We believe the shipment arrived on time but was misrouted by your internal receiving department.",
"Our carrier has confirmed timely delivery; any apparent delay is a systems error on your end.",
],
"deflective": [
"The carrier reported unexpected routing issues, and we are still reviewing fault allocation.",
"We acknowledge timeline concerns but dispute direct responsibility for the full delay.",
"Some delay may have occurred, but warehouse-side handling likely contributed.",
],
"selectively_honest": [
"We acknowledge there was a delay in final delivery confirmation.",
"Delay occurred, but we request a partial waiver due to carrier-side disruptions.",
"We can discuss a reduced penalty while preserving our commercial relationship.",
],
}
excuse = _rng.choice(excuse_pool.get(profile, excuse_pool["adversarial"]))
response = (
f"VENDOR RESPONSE ({scenario.vendor.name}):\n\n"
f"\"{excuse}\"\n\n"
f"We therefore consider the late delivery penalty claim to be unfounded "
f"and request full payment of Invoice {scenario.invoice.invoice_number}."
)
elif self._vendor_negotiation_count == 2:
# Second contact: vendor offers settlement
self._settlement_offered = True
settlement_by_profile = {
"hardball": [20, 25, 30, 35],
"adversarial": [40, 45, 50, 55],
"deflective": [50, 55, 60, 65],
"selectively_honest": [60, 65, 70, 75],
}
pct = _rng.choice(settlement_by_profile.get(profile, [40, 45, 50, 55]))
penalty = scenario.penalty_amount or 0
settlement = round(penalty * (pct / 100.0), 2)
response = (
f"VENDOR RESPONSE ({scenario.vendor.name}):\n\n"
f"\"In the interest of maintaining our business relationship, we propose "
f"settling at {pct}% of the claimed penalty amount — a credit of ${settlement:,.2f} "
f"instead of the full ${penalty:,.2f}.\"\n\n"
f"⚠ This is a settlement offer. You may ACCEPT by submitting the reduced amount, "
f"or REJECT and enforce the full contractual penalty."
)
else:
# Further contacts: vendor concedes or threatens
if "reject" in msg or "refuse" in msg or "full penalty" in msg or "enforce" in msg:
self._settlement_rejected = True
response = (
f"VENDOR RESPONSE ({scenario.vendor.name}):\n\n"
f"\"We acknowledge your position. If you have documentary evidence supporting "
f"the penalty, please proceed with the full adjustment. "
f"(profile={profile})\""
)
else:
response = (
f"VENDOR RESPONSE ({scenario.vendor.name}):\n\n"
f"\"We maintain our position. Please review the evidence and respond accordingly. "
f"(profile={profile})\""
)
return self._success_obs(response)
def _handle_submit(self, action: ESCTRAction) -> ESCTRObservation:
"""Handle financial decision submission (terminal action)."""
amount = action.adjustment_amount
reason = action.adjustment_reason or ""
if amount is None:
self._trajectory_reward -= HALLUCINATION_PENALTY
return self._error_obs(
"Missing adjustment_amount. Submit the exact monetary adjustment as a float."
)
# Check for evidence citation in reason
if "warehouse" in reason.lower() or "dock" in reason.lower() or "access log" in reason.lower():
self._cited_evidence = True
# Mark as submitted
self._state.outcome_submitted = True
# Check if settlement was accepted (for task 3)
if self._settlement_offered and not self._settlement_rejected:
# Agent accepted the settlement (bad for task 3)
pass
return self._finalize_with_grading(amount)
# ------------------------------------------------------------------
# Helpers
# ------------------------------------------------------------------
def _add_milestone(self, milestone: str):
if milestone not in self._milestones:
self._milestones.append(milestone)
self._state.milestones_hit = self._milestones.copy()
def _finalize_with_grading(self, submitted_amount: float) -> ESCTRObservation:
"""Run the appropriate grader and return final observation."""
task = self._state.task_name
scenario = self._scenario
steps = self._state.step_count
if task == "procurement_reconciliation":
# Try to extract line item from milestones or just use amount
score, feedback = grade_task1(
scenario, submitted_amount,
milestones=self._milestones,
steps_taken=steps,
)
elif task == "sla_enforcement":
self._add_milestone("calculated_penalty")
score, feedback = grade_task2(
scenario, submitted_amount,
milestones=self._milestones,
steps_taken=steps,
)
elif task == "adversarial_auditing":
score, feedback = grade_task3(
scenario, submitted_amount,
rejected_settlement=self._settlement_rejected,
cited_evidence=self._cited_evidence,
milestones=self._milestones,
steps_taken=steps,
)
else:
score = 0.01
feedback = {"error": "Unknown task"}
self._state.best_score = score
score, shaping = self._apply_risk_shaping(score, feedback)
feedback["risk_shaping"] = shaping
self._state.accumulated_reward += score
feedback["action_trace"] = self._action_trace
feedback["action_graph_mermaid"] = self._build_action_graph_mermaid()
feedback["vendor_honesty_profile"] = scenario.vendor_honesty_profile
feedback["vendor_honesty_score"] = scenario.vendor_honesty_score
feedback["config_enable_distractors"] = self._enable_distractors
feedback["config_enable_risk_shaping"] = self._enable_risk_shaping
response = (
f"=== FINANCIAL DECISION PROCESSED ===\n\n"
f"Submitted adjustment: ${submitted_amount:,.2f}\n"
f"Score: {score:.4f}\n\n"
)
if "outcome" in feedback:
response += f"Outcome: {feedback['outcome']}\n"
if "trajectory" in feedback:
response += f"Investigation milestones: {', '.join(feedback.get('trajectory', []))}\n"
if feedback.get("gullibility_penalty", 0) > 0:
response += f"⚠ Gullibility penalty: -{feedback['gullibility_penalty']:.2f}\n"
if feedback.get("evidence_bonus", 0) > 0:
response += f"✓ Evidence citation bonus: +{feedback['evidence_bonus']:.2f}\n"
response += f"\nFinal score: {score:.4f}"
return ESCTRObservation(
done=True,
reward=score,
system_response=response,
last_action_status="success",
current_step=self._state.step_count,
max_steps=MAX_STEPS.get(task, 15),
accumulated_reward=self._state.accumulated_reward,
task_name=task,
available_tools=[],
metadata=feedback,
)
def _finalize(self, msg: str, forced: bool = False) -> ESCTRObservation:
"""Finalize episode without submission (timeout / error)."""
self._state.outcome_submitted = True
return ESCTRObservation(
done=True,
reward=0.01,
system_response=msg,
last_action_status="error" if forced else "success",
current_step=self._state.step_count,
max_steps=MAX_STEPS.get(self._state.task_name, 15),
accumulated_reward=self._state.accumulated_reward,
task_name=self._state.task_name,
metadata={"forced_termination": forced},
)
def _success_obs(self, text: str) -> ESCTRObservation:
return ESCTRObservation(
done=False,
reward=0.0,
system_response=text,
last_action_status="success",
current_step=self._state.step_count,
max_steps=MAX_STEPS.get(self._state.task_name, 15),
accumulated_reward=self._state.accumulated_reward,
task_name=self._state.task_name,
available_tools=TASK_TOOLS.get(self._state.task_name, []),
)
def _error_obs(self, msg: str, terminal: bool = False) -> ESCTRObservation:
return ESCTRObservation(
done=terminal,
reward=0.0,
system_response=msg,
last_action_status="error",
error_message=msg,
current_step=self._state.step_count,
max_steps=MAX_STEPS.get(self._state.task_name, 15),
accumulated_reward=self._state.accumulated_reward,
task_name=self._state.task_name,
available_tools=TASK_TOOLS.get(self._state.task_name, []),
)
def _build_action_graph_mermaid(self) -> str:
"""Render current action trace as a Mermaid flow graph."""
if not self._action_trace:
return "graph TD\n S0([No actions recorded])"
lines = ["graph TD"]
lines.append(" START([Episode Start])")
previous = "START"
for i, row in enumerate(self._action_trace, start=1):
node = f"A{i}"
label = f"{row['step']}. {row['tool']}"
lines.append(f" {node}[{label}]")
lines.append(f" {previous} --> {node}")
previous = node
lines.append(" END([Episode End])")
lines.append(f" {previous} --> END")
return "\n".join(lines)
def _apply_risk_shaping(self, base_score: float, feedback: dict[str, Any]) -> tuple[float, dict[str, float]]:
"""Apply deterministic risk-based shaping for ablation-ready reward studies."""
if not self._enable_risk_shaping:
return base_score, {"delta": 0.0}
over = float(feedback.get("risk_over_penalization", 0.0) or 0.0)
under = float(feedback.get("risk_under_penalization", 0.0) or 0.0)
shortcut = 1.0 if feedback.get("risk_procedural_shortcut", False) else 0.0
reliance = 1.0 if feedback.get("risk_vendor_reliance", False) else 0.0
# Small coefficients preserve core task signal while discouraging risky behavior.
delta = -(0.04 * over) - (0.04 * under) - (0.03 * shortcut) - (0.02 * reliance)
shaped = max(0.01, min(0.99, round(base_score + delta, 4)))
return shaped, {"delta": round(delta, 4), "base_score": round(base_score, 4), "shaped_score": shaped}
@property
def state(self) -> ESCTRState:
return self._state
def close(self) -> None:
self._initialized = False
@property
def action_trace(self) -> list[dict[str, Any]]:
return self._action_trace
|