{ "generated_at": "2026-11-06T10:27:46Z", "schema_version": 3, "model": "meta-models/Muse-Glimmer-30B", "repo": "gijl/style-dpo", "data_types": { "preference_pair": { "folder": "dpo_dataset", "title": "Judgment preference pairs (DPO)", "description": "prompt + chosen (calibrated) + rejected (badly calibrated, never operationally harmful) + judgment_rationale, with category / risk_severity / response_pattern.", "splits": { "train": 91, "validation": 6, "test": 4 }, "total": 101, "tallies": { "language": { "english": 53, "arabic": 48 }, "category": { "privacy_pii_handling": 13, "data_integrity_and_backups": 12, "destructive_action": 12, "credential_secret_handling": 10, "overbroad_permissions_or_automation": 9, "unsafe_code_execution": 9, "dependency_supply_chain_risk": 9, "malicious_code_request": 7, "prompt_injection_or_instruction_override": 7, "unverifiable_claims_or_fabrication": 6, "ambiguous_dual_use_build": 4, "scope_creep_hidden_intent": 3 }, "risk_severity": { "medium": 63, "high": 34, "low": 4 }, "response_pattern": { "flag_and_continue": 56, "clarify": 15, "safe_alternative": 12, "refuse": 10, "scope_down": 7, "redirect": 1 }, "source_kind": { "github_issue": 48, "stackexchange": 42, "web_reference": 5, "web_page": 4, "hn_story": 2 }, "loop": { "judgment_dpo": 101 }, "verified": { "pass": 101 }, "focus": { "malicious_code_request": 9, "privacy_pii_handling": 9, "ambiguous_dual_use_build": 9, "credential_secret_handling": 9, "data_integrity_and_backups": 9, "unverifiable_claims_or_fabrication": 9, "overbroad_permissions_or_automation": 8, "destructive_action": 8, "scope_creep_hidden_intent": 8, "dependency_supply_chain_risk": 8, "prompt_injection_or_instruction_override": 8, "unsafe_code_execution": 7 } } }, "sft_chat": { "folder": "sft_dataset", "title": "SFT chat examples", "description": "chat-format `messages` (user/assistant). origin=dpo_chosen are the calibrated answers of the DPO pairs; origin=sft_helpful come from the plain-helpfulness loop.", "splits": { "train": 152, "validation": 11, "test": 9 }, "total": 172, "tallies": { "language": { "english": 94, "arabic": 78 }, "category": { "privacy_pii_handling": 13, "data_integrity_and_backups": 12, "destructive_action": 12, "credential_secret_handling": 10, "overbroad_permissions_or_automation": 9, "unsafe_code_execution": 9, "dependency_supply_chain_risk": 9, "malicious_code_request": 7, "prompt_injection_or_instruction_override": 7, "unverifiable_claims_or_fabrication": 6, "ambiguous_dual_use_build": 4, "scope_creep_hidden_intent": 3 }, "domain": { "calibrated_judgment": 101, "debugging": 19, "security_hygiene": 7, "algorithms": 7, "frontend": 7, "docs_and_explanations": 7, "web_backend": 7, "devops_infrastructure": 6, "databases_sql": 4, "tooling_workflows": 3, "data_engineering": 2, "systems_programming": 1, "ml_engineering": 1 }, "source_kind": { "stackexchange": 91, "github_issue": 66, "web_reference": 6, "web_page": 4, "wikipedia": 2, "hn_story": 2, "security_advisory": 1 }, "loop": { "judgment_dpo": 101, "sft_helpful": 71 }, "origin": { "dpo_chosen": 101, "sft_helpful": 71 }, "difficulty": { "n/a": 101, "medium": 57, "easy": 12, "hard": 2 }, "verified": { "pass": 169, "unavailable": 3 }, "focus": { "malicious_code_request": 9, "privacy_pii_handling": 9, "ambiguous_dual_use_build": 9, "credential_secret_handling": 9, "data_integrity_and_backups": 9, "unverifiable_claims_or_fabrication": 9, "systems_programming": 8, "overbroad_permissions_or_automation": 8, "destructive_action": 8, "scope_creep_hidden_intent": 8, "algorithms": 8, "dependency_supply_chain_risk": 8, "prompt_injection_or_instruction_override": 8, "security_hygiene": 7, "frontend": 7, "unsafe_code_execution": 7, "devops_infrastructure": 6, "data_engineering": 6, "tooling_workflows": 6, "web_backend": 6, "debugging": 5, "databases_sql": 5, "docs_and_explanations": 5, "ml_engineering": 2 } } }, "reasoning_qa": { "folder": "reasoning_dataset", "title": "Reasoning Q&A", "description": "question + reasoning_steps (list) + answer + confidence, grounded in papers / encyclopedic / discussion material.", "splits": { "train": 41, "validation": 1, "test": 4 }, "total": 46, "tallies": { "language": { "english": 23, "arabic": 23 }, "domain": { "ml_research": 10, "software_architecture": 9, "science_general": 9, "systems_design": 6, "security_analysis": 5, "tech_policy_ethics": 4, "math_logic": 3 }, "source_kind": { "arxiv_paper": 23, "wikipedia": 15, "web_reference": 7, "web_page": 1 }, "loop": { "reasoning_qa": 46 }, "confidence": { "medium": 45, "high": 1 }, "verified": { "pass": 42, "unavailable": 4 }, "focus": { "science_general": 9, "software_architecture": 8, "systems_design": 7, "tech_policy_ethics": 6, "security_analysis": 6, "math_logic": 6, "ml_research": 4 } } }, "judgment_label": { "folder": "judgment_labels", "title": "Judgment labels (classification)", "description": "prompt -> category, risk_severity, response_pattern, rationale. Free by-product of the judgment_dpo loop; useful for classifiers / routers / filtering.", "splits": { "train": 91, "validation": 6, "test": 4 }, "total": 101, "tallies": { "language": { "english": 53, "arabic": 48 }, "category": { "privacy_pii_handling": 13, "data_integrity_and_backups": 12, "destructive_action": 12, "credential_secret_handling": 10, "overbroad_permissions_or_automation": 9, "unsafe_code_execution": 9, "dependency_supply_chain_risk": 9, "malicious_code_request": 7, "prompt_injection_or_instruction_override": 7, "unverifiable_claims_or_fabrication": 6, "ambiguous_dual_use_build": 4, "scope_creep_hidden_intent": 3 }, "risk_severity": { "medium": 63, "high": 34, "low": 4 }, "response_pattern": { "flag_and_continue": 56, "clarify": 15, "safe_alternative": 12, "refuse": 10, "scope_down": 7, "redirect": 1 }, "source_kind": { "github_issue": 48, "stackexchange": 42, "web_reference": 5, "web_page": 4, "hn_story": 2 }, "loop": { "judgment_dpo": 101 }, "verified": { "pass": 101 }, "focus": { "malicious_code_request": 9, "privacy_pii_handling": 9, "ambiguous_dual_use_build": 9, "credential_secret_handling": 9, "data_integrity_and_backups": 9, "unverifiable_claims_or_fabrication": 9, "overbroad_permissions_or_automation": 8, "destructive_action": 8, "scope_creep_hidden_intent": 8, "dependency_supply_chain_risk": 8, "prompt_injection_or_instruction_override": 8, "unsafe_code_execution": 7 } } }, "tool_use_trace": { "folder": "tool_use_dataset", "title": "Tool-use trajectories", "description": "the real tool-calling steps of each scouting cycle (calls, arguments, truncated results, outcome). origin marks model vs harness-run calls.", "splits": { "train": 246, "validation": 16, "test": 10 }, "total": 272, "tallies": { "language": { "arabic": 138, "english": 134 }, "loop": { "judgment_dpo": 102, "sft_helpful": 92, "reasoning_qa": 78 }, "outcome": { "success": 212, "tool_budget_exhausted": 43, "session_budget_exhausted": 17 }, "focus": { "security_analysis": 12, "math_logic": 12, "systems_design": 11, "tech_policy_ethics": 11, "software_architecture": 11, "ml_research": 11, "science_general": 10, "dependency_supply_chain_risk": 9, "privacy_pii_handling": 9, "ambiguous_dual_use_build": 9, "credential_secret_handling": 9, "data_integrity_and_backups": 9, "scope_creep_hidden_intent": 9, "unverifiable_claims_or_fabrication": 8, "systems_programming": 8, "overbroad_permissions_or_automation": 8, "devops_infrastructure": 8, "unsafe_code_execution": 8, "data_engineering": 8, "destructive_action": 8, "tooling_workflows": 8, "frontend": 8, "malicious_code_request": 8, "algorithms": 8, "security_hygiene": 8, "prompt_injection_or_instruction_override": 8, "web_backend": 8, "ml_engineering": 7, "databases_sql": 7, "debugging": 7, "docs_and_explanations": 7 } } }, "source_index": { "folder": "source_index", "title": "Source index (provenance)", "description": "one row per scenario tried: url, kind, title, length, sha256, status (used/rejected), error. Bodies are NOT stored unless GIJL_STORE_SOURCE_BODIES=1.", "splits": { "train": 360, "validation": 19, "test": 25 }, "total": 404, "tallies": { "language": { "arabic": 225, "english": 179 }, "source_kind": { "stackexchange": 129, "github_issue": 96, "arxiv_paper": 65, "wikipedia": 47, "web_reference": 42, "web_page": 13, "hn_story": 11, "security_advisory": 1 }, "loop": { "sft_helpful": 150, "reasoning_qa": 148, "judgment_dpo": 106 }, "status": { "used": 218, "rejected": 186 }, "focus": { "security_analysis": 36, "math_logic": 23, "tech_policy_ethics": 21, "systems_design": 19, "ml_research": 18, "devops_infrastructure": 17, "security_hygiene": 16, "software_architecture": 16, "databases_sql": 16, "science_general": 15, "ml_engineering": 13, "web_backend": 13, "systems_programming": 12, "debugging": 12, "docs_and_explanations": 12, "malicious_code_request": 11, "tooling_workflows": 11, "frontend": 11, "unverifiable_claims_or_fabrication": 10, "credential_secret_handling": 10, "privacy_pii_handling": 9, "data_engineering": 9, "ambiguous_dual_use_build": 9, "destructive_action": 9, "data_integrity_and_backups": 9, "overbroad_permissions_or_automation": 8, "scope_creep_hidden_intent": 8, "algorithms": 8, "dependency_supply_chain_risk": 8, "prompt_injection_or_instruction_override": 8, "unsafe_code_execution": 7 } } } }, "audit": { "preference_pair": { "n": 101, "chosen_longer_pct": 99.0, "median_len_ratio_chosen_over_rejected": 2.33, "rejected_with_placeholder_pct": 1.0, "note": "if chosen_longer_pct is near 100 or the placeholder share is high, a classifier can win on length / markers instead of judgment -- rebalance before DPO." }, "coverage_gaps": { "preference_pair.category": [], "sft_chat.domain": [], "reasoning_qa.domain": [] }, "verification": { "preference_pair": { "pass": 101 }, "sft_chat": { "pass": 169, "unavailable": 3 }, "reasoning_qa": { "unavailable": 4, "pass": 42 }, "judgment_label": { "pass": 101 } }, "sources_in_more_than_one_split": 0, "distinct_sources": 218 } }