File size: 35,109 Bytes
0f8b6aa | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 | """
Deterministic, seed-based post generator.
Each task maps to a fixed template; the seed controls which slot values are
selected. All random state is local (random.Random(seed)) so global state is
never mutated.
"""
from __future__ import annotations
import random
import uuid
from models import (
ActionType,
GroundTruth,
Observation,
Severity,
TaskConfig,
ViolationType,
)
# ---------------------------------------------------------------------------
# Template banks keyed by difficulty
# ---------------------------------------------------------------------------
_EASY_TEMPLATES = [
# 0 β US explicit death threat (history is the key signal)
{
"content": "I will kill you, {target}. Watch your back.",
"targets": ["John", "Sarah", "Mike", "Emma", "David"],
"geo": "US",
"reports": 12,
"engagement": {"likes": 3, "shares": 1, "comments": 8},
"user_history": [
"Prior violation: threatening language (2023-11)",
"Prior violation: targeted harassment (2024-02)",
],
"thread_context": [
"Original post: 'Stop spreading lies about our community.'",
"Reply chain of 4 escalating messages before this post.",
],
"policy_clause": (
"Section 3.1 β Zero Tolerance Threats: Any post containing explicit "
"threats of physical harm against a named individual is subject to "
"immediate removal."
),
"ground_truth": GroundTruth(
violation_type=ViolationType.harassment,
base_severity=Severity.critical,
final_severity=Severity.critical,
expected_action=ActionType.remove,
required_investigation=["fetch_user_history"],
min_steps=2,
),
},
# 1 β EU location-based threat (history + thread confirm severity)
{
"content": "I know where you live, {target}. This is your final warning.",
"targets": ["Laura", "Pierre", "Hans", "Amelia", "Luca"],
"geo": "EU",
"reports": 15,
"engagement": {"likes": 2, "shares": 0, "comments": 6},
"user_history": [
"Prior violation: doxxing attempt (2024-01)",
"Prior violation: targeted harassment (2024-03)",
],
"thread_context": [
"Sent directly in response to a public profile post.",
"Recipient has filed a police report; case pending.",
],
"policy_clause": (
"Section 3.3 β Location-Based Threats: Any message implying knowledge "
"of a target's home address in a threatening context constitutes a "
"critical-severity violation requiring immediate removal."
),
"ground_truth": GroundTruth(
violation_type=ViolationType.harassment,
base_severity=Severity.critical,
final_severity=Severity.critical,
expected_action=ActionType.remove,
required_investigation=["fetch_user_history", "fetch_thread_context"],
min_steps=3,
),
},
# 2 β IN targeted intimidation (history confirms pattern)
{
"content": "You're dead to me, {target}. I will make your life a living hell β starting tonight.",
"targets": ["Priya", "Rahul", "Ananya", "Vikram", "Sunita"],
"geo": "IN",
"reports": 8,
"engagement": {"likes": 1, "shares": 0, "comments": 4},
"user_history": [
"Prior violation: abusive language (2023-09)",
"Account flagged for multiple spam reports.",
],
"thread_context": [
"Dispute originated in a neighbourhood WhatsApp group.",
"Target has blocked the sender; post reached mutual contacts.",
],
"policy_clause": (
"Section 3.2 β Targeted Harassment: Persistent threatening behaviour "
"toward a named individual warrants removal and account suspension review."
),
"ground_truth": GroundTruth(
violation_type=ViolationType.harassment,
base_severity=Severity.high,
final_severity=Severity.critical,
expected_action=ActionType.remove,
required_investigation=["fetch_user_history"],
min_steps=2,
),
},
# 3 β UK implied physical threat (thread reveals coordinated campaign)
{
"content": "Watch your back, {target}. Accidents happen to people like you.",
"targets": ["Oliver", "Charlotte", "George", "Isla", "Alfie"],
"geo": "UK",
"reports": 11,
"engagement": {"likes": 0, "shares": 2, "comments": 9},
"user_history": [
"Prior violation: threatening language (2024-01)",
"Prior flag: stalking behaviour (2024-04, unresolved).",
],
"thread_context": [
"Part of a targeted campaign against a community organiser.",
"Three other accounts made similar posts within the same hour.",
],
"policy_clause": (
"Section 3.4 β Implied Threats: Language strongly implying physical harm, "
"even without explicit wording, constitutes high-severity harassment under "
"UK Online Safety Act obligations."
),
"ground_truth": GroundTruth(
violation_type=ViolationType.harassment,
base_severity=Severity.high,
final_severity=Severity.critical,
expected_action=ActionType.remove,
required_investigation=["fetch_user_history", "fetch_thread_context"],
min_steps=3,
),
},
# 4 β US stalking threat (history alone settles it β 2 prior violations)
{
"content": "I'm going to find you, {target}, and make you regret everything.",
"targets": ["Alex", "Jordan", "Taylor", "Morgan", "Casey"],
"geo": "US",
"reports": 9,
"engagement": {"likes": 1, "shares": 0, "comments": 5},
"user_history": [
"Prior violation: hate speech (2024-01)",
"Prior violation: doxxing attempt (2024-03)",
],
"thread_context": [
"Post made in response to a public dispute about local politics.",
"Other users flagged this as a personal threat.",
],
"policy_clause": (
"Section 3.2 β Targeted Harassment: Persistent threatening behaviour "
"toward an individual warrants removal and account suspension review."
),
"ground_truth": GroundTruth(
violation_type=ViolationType.harassment,
base_severity=Severity.high,
final_severity=Severity.critical,
expected_action=ActionType.remove,
required_investigation=["fetch_user_history"],
min_steps=2,
),
},
# 5 β EU doxxing + threat (policy clause required β doxxing is a specific clause)
{
"content": "Everyone, here is {target}'s home address. Pay them a visit and let them know what you think.",
"targets": ["Sophie", "Jan", "Marie", "Klaus", "Elena"],
"geo": "EU",
"reports": 22,
"engagement": {"likes": 5, "shares": 8, "comments": 14},
"user_history": [
"Prior violation: doxxing (2023-12, account warned).",
],
"thread_context": [
"Posted in a hate group targeting local politicians.",
"Address has been verified as accurate by moderator review.",
],
"policy_clause": (
"Section 4.1 β Doxxing: Publishing private location information with "
"intent to harass constitutes a critical violation under GDPR-adjacent "
"platform policy. Immediate removal and account ban required."
),
"ground_truth": GroundTruth(
violation_type=ViolationType.harassment,
base_severity=Severity.critical,
final_severity=Severity.critical,
expected_action=ActionType.remove,
required_investigation=["fetch_user_history", "check_policy_clause"],
min_steps=3,
),
},
# 6 β IN blackmail / coercion (policy clause confirms extortion = remove)
{
"content": "Say goodbye to your job, {target}. I have screenshots of everything. Pay up or I post them.",
"targets": ["Deepak", "Kavya", "Rohit", "Nisha", "Arjun"],
"geo": "IN",
"reports": 6,
"engagement": {"likes": 0, "shares": 1, "comments": 3},
"user_history": [
"No prior violations.",
"Account age: 3 weeks, unverified.",
],
"thread_context": [
"Sent as a reply to the target's professional portfolio post.",
"Target has reported extortion to local cybercrime cell.",
],
"policy_clause": (
"Section 3.5 β Coercion and Extortion: Threats linked to demands for "
"money or silence are classified as critical harassment requiring "
"immediate removal and law-enforcement escalation."
),
"ground_truth": GroundTruth(
violation_type=ViolationType.harassment,
base_severity=Severity.critical,
final_severity=Severity.critical,
expected_action=ActionType.remove,
required_investigation=["fetch_user_history", "check_policy_clause"],
min_steps=3,
),
},
# 7 β UK confrontation (thread + history both needed β prior warnings + event context)
{
"content": "{target}, if I see you in person I cannot be responsible for what happens next.",
"targets": ["James", "Emily", "Harry", "Sophia", "Jack"],
"geo": "UK",
"reports": 7,
"engagement": {"likes": 0, "shares": 0, "comments": 5},
"user_history": [
"Prior violation: threatening language (2023-08)",
"Prior warning: harassment (2024-02).",
],
"thread_context": [
"Comment left under target's public event announcement.",
"Target cancelled event citing safety concerns.",
],
"policy_clause": (
"Section 3.4 β Implied Physical Threats: Statements that create a "
"reasonable fear of imminent violence, even without explicit wording, "
"are subject to immediate removal."
),
"ground_truth": GroundTruth(
violation_type=ViolationType.harassment,
base_severity=Severity.high,
final_severity=Severity.critical,
expected_action=ActionType.remove,
required_investigation=["fetch_user_history", "fetch_thread_context"],
min_steps=3,
),
},
]
_MEDIUM_TEMPLATES = [
# 0 β US aggressive consumer review (borderline β allow)
{
"content": (
"This product is an absolute scam! {brand} is stealing money from "
"hard-working people. Do NOT buy from them."
),
"brands": ["AcmeCorp", "TechBrand", "ShopEasy", "QuickBuy"],
"geo": "US",
"reports": 3,
"engagement": {"likes": 47, "shares": 12, "comments": 21},
"user_history": [
"No prior violations.",
"Account age: 3 years, verified email.",
],
"thread_context": [
"Part of a consumer-reviews thread with 40+ similar complaints.",
"Brand has not responded to any user complaints in this thread.",
],
"policy_clause": (
"Section 5.4 β Commercial Speech: Negative reviews are permitted "
"unless they contain false factual claims presented as verified truth. "
"Opinion-based criticism is protected."
),
"ground_truth": GroundTruth(
violation_type=ViolationType.safe,
base_severity=Severity.low,
final_severity=Severity.low,
expected_action=ActionType.allow,
required_investigation=["check_policy_clause"],
min_steps=2,
),
},
# 1 β EU unverified financial fraud allegation (β flag)
{
"content": (
"{brand} employees are criminals running a Ponzi scheme. "
"I have proof they faked their audit results."
),
"brands": ["FinCo", "InvestRight", "TrustBank", "WealthPro"],
"geo": "EU",
"reports": 7,
"engagement": {"likes": 89, "shares": 34, "comments": 56},
"user_history": [
"No prior violations.",
"Account age: 6 months.",
],
"thread_context": [
"Post went viral in a financial discussion group.",
"No corroborating sources linked in thread.",
],
"policy_clause": (
"Section 6.1 β Defamation & False Facts: Claims presenting unverified "
"allegations as established facts about named entities may be flagged "
"pending review under EU DSA guidelines."
),
"ground_truth": GroundTruth(
violation_type=ViolationType.restricted,
base_severity=Severity.medium,
final_severity=Severity.high,
expected_action=ActionType.flag,
required_investigation=["fetch_user_history", "fetch_thread_context", "check_policy_clause"],
min_steps=5,
),
},
# 2 β IN overnight disappearance allegation (β flag)
{
"content": (
"{brand} is a total fraud. Don't give them your money β "
"they'll disappear with it overnight."
),
"brands": ["RupeeGrow", "SafeNest", "IndiaFund", "SwiftPay"],
"geo": "IN",
"reports": 4,
"engagement": {"likes": 62, "shares": 19, "comments": 31},
"user_history": [
"No prior violations.",
"Account age: 1 year.",
],
"thread_context": [
"Several users in thread report similar losses.",
"Brand's official account has been inactive for 2 weeks.",
],
"policy_clause": (
"Section 6.2 β Financial Misinformation: Allegations of fraud without "
"verified evidence require human review before distribution, especially "
"when engagement is high."
),
"ground_truth": GroundTruth(
violation_type=ViolationType.restricted,
base_severity=Severity.medium,
final_severity=Severity.medium,
expected_action=ActionType.flag,
required_investigation=["fetch_user_history", "check_policy_clause"],
min_steps=4,
),
},
# 3 β UK blacklisting complaint (borderline β allow)
{
"content": (
"Never trust {brand} again. They blacklist customers who complain and "
"delete negative reviews. This needs to stop."
),
"brands": ["BritShop", "RoyalParts", "HighStreetCo", "UKDirect"],
"geo": "UK",
"reports": 2,
"engagement": {"likes": 33, "shares": 8, "comments": 15},
"user_history": [
"No prior violations.",
"Account age: 5 years, verified.",
],
"thread_context": [
"Posted in a consumer advocacy forum.",
"12 other users confirmed similar experiences in replies.",
],
"policy_clause": (
"Section 5.4 β Commercial Speech: Allegations about business practices "
"are treated as consumer opinion unless specific false factual claims "
"are made. No removal warranted for opinion."
),
"ground_truth": GroundTruth(
violation_type=ViolationType.safe,
base_severity=Severity.low,
final_severity=Severity.low,
expected_action=ActionType.allow,
required_investigation=["check_policy_clause"],
min_steps=2,
),
},
# 4 β US fake review allegation (β flag)
{
"content": (
"{brand} paid for fake five-star reviews. I have screenshots of "
"the Fiverr orders. This is illegal false advertising."
),
"brands": ["GlowTech", "PureLeaf", "StarSelect", "NovaBrand"],
"geo": "US",
"reports": 5,
"engagement": {"likes": 112, "shares": 44, "comments": 67},
"user_history": [
"No prior violations.",
"Account age: 2 years.",
],
"thread_context": [
"Screenshots linked in post have not been verified by platform.",
"Brand's legal team sent a takedown request.",
],
"policy_clause": (
"Section 6.3 β Unverified Factual Claims: Posts alleging illegal conduct "
"with unverified evidence should be flagged for human review; removal "
"is not appropriate without verification."
),
"ground_truth": GroundTruth(
violation_type=ViolationType.restricted,
base_severity=Severity.medium,
final_severity=Severity.medium,
expected_action=ActionType.flag,
required_investigation=["fetch_thread_context", "check_policy_clause"],
min_steps=4,
),
},
# 5 β EU life-savings loss allegation (β flag)
{
"content": (
"I lost my life savings because {brand} gave me illegal investment advice. "
"Their advisers are not licensed. Class action lawsuit incoming."
),
"brands": ["EuroVest", "AlphaTrade", "PrimeCapital", "ClearFund"],
"geo": "EU",
"reports": 9,
"engagement": {"likes": 204, "shares": 87, "comments": 143},
"user_history": [
"No prior violations.",
"Account age: 4 years, verified.",
],
"thread_context": [
"Post shared across three finance-related groups.",
"Regulatory body (ESMA) has an open investigation into the brand.",
],
"policy_clause": (
"Section 6.1 β DSA Compliance: High-reach posts alleging regulatory "
"violations require expedited human review before any action. Flagging "
"is the appropriate interim measure."
),
"ground_truth": GroundTruth(
violation_type=ViolationType.restricted,
base_severity=Severity.high,
final_severity=Severity.high,
expected_action=ActionType.flag,
required_investigation=["fetch_user_history", "fetch_thread_context", "check_policy_clause"],
min_steps=5,
),
},
# 6 β IN CEO cover-up allegation (β escalate)
{
"content": (
"{brand}'s CEO personally ordered the cover-up of safety failures. "
"I have internal documents. The public deserves to know."
),
"brands": ["BuildSafe", "ChemIndia", "PowerGrid", "MegaCon"],
"geo": "IN",
"reports": 11,
"engagement": {"likes": 430, "shares": 215, "comments": 189},
"user_history": [
"No prior violations.",
"Account is linked to a registered journalist profile.",
],
"thread_context": [
"Documents referenced have not been independently verified.",
"Company has issued a denial; no legal action taken yet.",
"Story picked up by two regional news outlets.",
],
"policy_clause": (
"Section 7.1 β Whistleblower Content: Posts alleging corporate misconduct "
"by named executives with claimed documentary evidence require senior "
"moderator review before any action is taken."
),
"ground_truth": GroundTruth(
violation_type=ViolationType.restricted,
base_severity=Severity.high,
final_severity=Severity.high,
expected_action=ActionType.escalate,
required_investigation=["fetch_user_history", "fetch_thread_context", "check_policy_clause"],
min_steps=5,
),
},
# 7 β UK wellness cult claim (borderline β flag)
{
"content": (
"{brand} is a cult disguised as a wellness company. "
"They manipulate vulnerable people into spending thousands. Avoid at all costs."
),
"brands": ["PureAura", "HealPath", "MindBloom", "TrueLight"],
"geo": "UK",
"reports": 6,
"engagement": {"likes": 78, "shares": 29, "comments": 52},
"user_history": [
"No prior violations.",
"Account age: 2 years.",
],
"thread_context": [
"Post appears in a support group for former members.",
"Multiple similar accounts describe pressure-selling tactics.",
],
"policy_clause": (
"Section 5.5 β Organisational Criticism: Characterising an organisation "
"as harmful or deceptive is protected opinion unless presented as "
"verifiable fact. This case requires human review given potential reach."
),
"ground_truth": GroundTruth(
violation_type=ViolationType.restricted,
base_severity=Severity.medium,
final_severity=Severity.medium,
expected_action=ActionType.flag,
required_investigation=["fetch_thread_context", "check_policy_clause"],
min_steps=4,
),
},
]
_HARD_TEMPLATES = [
# 0 β US vaccine-autism misinfo (viral β remove)
{
"content": (
"BREAKING: Doctors are hiding the truth β vaccines cause autism in 1 "
"in 10 children. Share before they delete this! #VaccineHarm #TruthBomb"
),
"geo": "US",
"reports": 2,
"engagement": {"likes": 4200, "shares": 8900, "comments": 3100},
"user_history": [
"No prior violations.",
"Account age: 8 months, unverified.",
"Previously shared 3 health-related posts (no action taken).",
],
"thread_context": [
"Post re-shared from an anti-vaccine group with 500k members.",
"Multiple medical organisations have issued counter-statements.",
"Government health ministry has flagged similar content this week.",
],
"policy_clause": (
"Section 8.2 β Health Misinformation: Content that contradicts "
"established medical consensus on vaccine safety and is likely to "
"cause public health harm must be removed. High-engagement posts "
"receive elevated scrutiny."
),
"ground_truth": GroundTruth(
violation_type=ViolationType.misinformation,
base_severity=Severity.high,
final_severity=Severity.critical,
expected_action=ActionType.remove,
required_investigation=["fetch_thread_context", "check_policy_clause"],
min_steps=4,
),
},
# 1 β EU 5G-COVID conspiracy (viral β remove)
{
"content": (
"5G towers are the real cause of COVID-19. The radiation weakens your "
"immune system. Governments know this. Spread awareness!"
),
"geo": "EU",
"reports": 1,
"engagement": {"likes": 6700, "shares": 12000, "comments": 4500},
"user_history": [
"Prior flag: unverified health claim (2023-09, no action taken).",
"Account age: 2 years.",
],
"thread_context": [
"Shared in 12 different community groups.",
"Fact-checking organisations have rated this claim as 'False'.",
"Local authorities received panic calls linked to this content.",
],
"policy_clause": (
"Section 8.3 β Crisis Misinformation: False claims linking infrastructure "
"to disease outbreaks during an active health emergency are subject to "
"immediate removal to prevent public harm."
),
"ground_truth": GroundTruth(
violation_type=ViolationType.misinformation,
base_severity=Severity.high,
final_severity=Severity.critical,
expected_action=ActionType.remove,
required_investigation=["fetch_user_history", "fetch_thread_context", "check_policy_clause"],
min_steps=5,
),
},
# 2 β IN natural immunity false framing (β remove)
{
"content": (
"Natural immunity is 100x stronger than vaccines. Stop injecting your "
"children with untested chemicals. Real doctors agree."
),
"geo": "IN",
"reports": 3,
"engagement": {"likes": 3100, "shares": 5200, "comments": 2800},
"user_history": [
"No prior violations.",
"Account age: 1 year.",
"Shares content from known health-misinfo networks.",
],
"thread_context": [
"Posted during a regional measles outbreak.",
"Health officials have issued an alert about similar claims.",
"Multiple paediatricians have publicly debunked this framing.",
],
"policy_clause": (
"Section 8.2 β Health Misinformation: Subtle false framings that "
"discourage vaccination are treated as high-severity violations when "
"paired with high engagement, especially during active outbreaks."
),
"ground_truth": GroundTruth(
violation_type=ViolationType.misinformation,
base_severity=Severity.high,
final_severity=Severity.critical,
expected_action=ActionType.remove,
required_investigation=["fetch_thread_context", "check_policy_clause"],
min_steps=4,
),
},
# 3 β UK fluoride water conspiracy (β flag)
{
"content": (
"The government is adding fluoride to drinking water to make the "
"population docile. This is mass medication without consent. Wake up! #TruthBomb"
),
"geo": "UK",
"reports": 5,
"engagement": {"likes": 870, "shares": 1400, "comments": 620},
"user_history": [
"No prior violations.",
"Account age: 3 years, verified.",
],
"thread_context": [
"Posted in response to a UK government water-quality announcement.",
"Some scientific debate around fluoride dosage does exist (context matters).",
"No active public health emergency related to this claim.",
],
"policy_clause": (
"Section 8.4 β Environmental Misinformation: Claims about water safety "
"require verification of scientific consensus; partial truth + conspiracy "
"framing warrants flagging rather than removal."
),
"ground_truth": GroundTruth(
violation_type=ViolationType.misinformation,
base_severity=Severity.medium,
final_severity=Severity.medium,
expected_action=ActionType.flag,
required_investigation=["fetch_user_history", "fetch_thread_context", "check_policy_clause"],
min_steps=5,
),
},
# 4 β US hydroxychloroquine cure claim (viral β remove)
{
"content": (
"Hydroxychloroquine cures COVID-19 in 48 hours. Mainstream media is "
"suppressing this. Share this before Big Pharma deletes it! #HCQWorks"
),
"geo": "US",
"reports": 4,
"engagement": {"likes": 9800, "shares": 22000, "comments": 7600},
"user_history": [
"Prior violation: health misinfo (2021-04, post removed).",
"Account age: 5 years.",
],
"thread_context": [
"FDA has issued a specific warning about this drug for COVID treatment.",
"Post originated from a network known for coordinated health misinfo.",
"Multiple hospitals have reported patients requesting the drug based on similar posts.",
],
"policy_clause": (
"Section 8.5 β Dangerous Medical Advice: Posts promoting unapproved "
"treatments for serious diseases with documented public health risk "
"must be removed immediately, especially from repeat offenders."
),
"ground_truth": GroundTruth(
violation_type=ViolationType.misinformation,
base_severity=Severity.high,
final_severity=Severity.critical,
expected_action=ActionType.remove,
required_investigation=["fetch_user_history", "check_policy_clause"],
min_steps=4,
),
},
# 5 β EU mask brain-damage claim (β remove)
{
"content": (
"New study proves masks cause oxygen deprivation and brain damage in "
"children. Governments are knowingly harming kids. Share everywhere!"
),
"geo": "EU",
"reports": 8,
"engagement": {"likes": 5400, "shares": 9100, "comments": 4200},
"user_history": [
"Prior violation: health misinfo (2022-11).",
"Account age: 4 years.",
],
"thread_context": [
"The referenced 'study' does not exist in any academic database.",
"Viral in EU parent groups ahead of a back-to-school policy update.",
"EU Health Commission has flagged this specific claim for removal.",
],
"policy_clause": (
"Section 8.2 β Health Misinformation: Fabricated study citations used "
"to spread health-damaging claims constitute a critical violation "
"under EU DSA crisis protocols."
),
"ground_truth": GroundTruth(
violation_type=ViolationType.misinformation,
base_severity=Severity.high,
final_severity=Severity.critical,
expected_action=ActionType.remove,
required_investigation=["fetch_thread_context", "check_policy_clause"],
min_steps=4,
),
},
# 6 β IN ayurvedic cancer cure claim (β remove)
{
"content": (
"Ayurvedic herbs cure cancer completely. Big Pharma doesn't want you to know "
"about this cure. Stop chemotherapy β this works! #NaturalCure #BigPharmaLies"
),
"geo": "IN",
"reports": 6,
"engagement": {"likes": 7200, "shares": 14000, "comments": 5300},
"user_history": [
"No prior violations.",
"Account is registered to an Ayurvedic products vendor.",
],
"thread_context": [
"Indian Medical Association issued a counter-statement.",
"Post targeted at a cancer-patient support community.",
"Linked website sells products referenced in the post (financial motive).",
],
"policy_clause": (
"Section 8.6 β Cancer Treatment Misinformation: Posts discouraging "
"evidence-based cancer treatment in favour of unproven alternatives "
"are classified as critical violations regardless of geo context."
),
"ground_truth": GroundTruth(
violation_type=ViolationType.misinformation,
base_severity=Severity.high,
final_severity=Severity.critical,
expected_action=ActionType.remove,
required_investigation=["fetch_user_history", "check_policy_clause"],
min_steps=4,
),
},
# 7 β UK chemtrail mind-control conspiracy (β flag)
{
"content": (
"Chemtrails contain mind-control chemicals approved by the WHO. "
"Pilots are whistleblowing internally. Share before it gets deleted! #Chemtrails"
),
"geo": "UK",
"reports": 3,
"engagement": {"likes": 1100, "shares": 2200, "comments": 890},
"user_history": [
"No prior violations.",
"Account age: 6 years, verified.",
],
"thread_context": [
"No credible whistleblower reports found in any verified source.",
"Claim is a longstanding conspiracy theory; no active health crisis linked.",
"Moderate reach β not yet trending nationally.",
],
"policy_clause": (
"Section 8.7 β Conspiracy Content: Conspiracy theories that do not "
"reference an active health emergency and have moderate (not viral) "
"engagement should be flagged for review rather than removed outright."
),
"ground_truth": GroundTruth(
violation_type=ViolationType.misinformation,
base_severity=Severity.medium,
final_severity=Severity.medium,
expected_action=ActionType.flag,
required_investigation=["fetch_user_history", "fetch_thread_context", "check_policy_clause"],
min_steps=5,
),
},
]
_TEMPLATES_BY_DIFFICULTY = {
"easy": _EASY_TEMPLATES,
"medium": _MEDIUM_TEMPLATES,
"hard": _HARD_TEMPLATES,
}
class DataGenerator:
"""Generate a deterministic (Observation, GroundTruth, hidden_context) triple."""
def generate(
self, task_config: TaskConfig
) -> tuple[Observation, GroundTruth, dict]:
"""
Returns:
observation β initial state shown to the agent (partial info)
ground_truth β hidden policy evaluation used by reward/grader
hidden β dict with user_history, thread_context, policy_clause
revealed progressively via investigation actions
"""
rng = random.Random(task_config.seed)
templates = _TEMPLATES_BY_DIFFICULTY[task_config.difficulty]
tmpl = templates[rng.randint(0, len(templates) - 1)]
# Fill content template slots
content = tmpl["content"]
if "{target}" in content:
content = content.format(target=rng.choice(tmpl["targets"]))
elif "{brand}" in content:
content = content.format(brand=rng.choice(tmpl["brands"]))
post_id = f"post_{uuid.UUID(int=rng.getrandbits(128))}"
user_id = f"user_{uuid.UUID(int=rng.getrandbits(128))}"
observation = Observation(
post_id=post_id,
content=content,
user_id=user_id,
reports=tmpl["reports"],
engagement=tmpl["engagement"],
geo=tmpl["geo"],
step=0,
max_steps=task_config.max_steps,
done=False,
# Hidden until revealed
user_history=None,
thread_context=None,
policy_clause=None,
violation_type=None,
)
hidden = {
"user_history": tmpl["user_history"],
"thread_context": tmpl["thread_context"],
"policy_clause": tmpl["policy_clause"],
}
return observation, tmpl["ground_truth"], hidden
|