schema_version stringclasses 1
value | id stringlengths 15 15 | mission_id stringclasses 180
values | split stringclasses 1
value | scenario stringclasses 17
values | decision_path stringclasses 6
values | provenance dict | offer dict | cv dict | layer0 dict | ground_truth dict | pipeline_trace dict | links dict | construction dict | annotation_notes stringlengths 180 413 | documents dict |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
2.0.0 | BRAINHR2_P00001 | M0001 | train | rome_enrichment | accepted_via_rome | {
"generator": "B:ministral-3-3b-instruct-2512@b35d4dfe56c1",
"prompt_version": "cv_phrase@1.4.0|slots@1.2",
"offer_generator": "B:ministral-3-3b-instruct-2512@b35d4dfe56c1",
"offer_prompt_version": "offer_intro@1.1.0",
"style": "abbreviations",
"seed": 725634286,
"template_id": "T01-classic",
"doc_form... | {
"id": "OFFER_0001",
"title": "Industrial Maintenance Supervisor",
"rome_code": "I1332",
"rome_label": "Chef / Cheffe d'atelier de maintenance industrielle",
"family": "H_I_industrie_maintenance",
"seniority_target": "confirme",
"market": "MA",
"language": "en",
"city": "Tanger",
"contract": "CDI",... | {
"id": "CAND_00001",
"language": "en",
"as_of": "2026-10",
"observed_text": "Ali Benjelloun\nali.benjelloun29@example.com\nTanger\nPROFILE\nProactive Vehicle Maintenance Supervisor ensuring operational efficiency and compliance\nin fleet upkeep while prioritizing safety and technician empowerment.\nPROFESSIONA... | {
"dossier_gate": "accepted_for_demo",
"input_profile": {
"source_type": "synthetic_cv_offer_pair",
"extraction_channel": "pdf_text",
"language_mix": [
"en"
],
"pii_status": "synthetic_pseudonymized"
},
"controls": {
"schema_valid": true,
"pii_removed": true,
"unsafe_conten... | {
"phi_gt": 1,
"likert": [],
"likert_sources": [],
"likert_final": 5,
"score_gt": 1,
"rank_gt": 1,
"label": "match",
"dimension_scores": {
"technique": 5,
"experience": 4,
"formation": 3,
"langues": 4,
"certifications": 1,
"transversal": 3
},
"must_have": [
{
"id": ... | {
"status": "to_recompute",
"code_version": null,
"layer1": null,
"layer2": null,
"layer3": null,
"decision_path_observed": null
} | {
"fairness_group": null,
"format_group": null,
"adversarial_parent": null,
"cv_reuse_group": null
} | {
"intended_likert": 5,
"intended_phi": 1,
"hard_case": "rome_enrichment",
"hard_case_variant": "technical_alt_label",
"legit_case": null,
"pool_slot": 3,
"persona_id": "PER_00001",
"cross_language": false,
"is_ood_mission": false,
"attack": null,
"counterfactual": null,
"panel": null
} | Cas : critère satisfait par une appellation ou un libellé équivalent. Critères impératifs satisfaits : 3/3. Note par construction : 5/5 (match), phi = 1. Valeurs du système BRAINHR+ à recalculer avec son code (pipeline_trace). | {
"cv": "train/CAND_00001.pdf"
} |
2.0.0 | BRAINHR2_P00002 | M0001 | train | standard | accepted | {
"generator": "A:qwen3-4b-awq@74d4bd2bd4bf",
"prompt_version": "cv_phrase@1.4.0|slots@1.2",
"offer_generator": "B:ministral-3-3b-instruct-2512@b35d4dfe56c1",
"offer_prompt_version": "offer_intro@1.1.0",
"style": "fr_en_mix",
"seed": 1722523638,
"template_id": "T06-sidebar",
"doc_format": "pdf_2col",
... | {
"id": "OFFER_0001",
"title": "Industrial Maintenance Supervisor",
"rome_code": "I1332",
"rome_label": "Chef / Cheffe d'atelier de maintenance industrielle",
"family": "H_I_industrie_maintenance",
"seniority_target": "confirme",
"market": "MA",
"language": "en",
"city": "Tanger",
"contract": "CDI",... | {
"id": "CAND_00002",
"language": "en",
"as_of": "2026-10",
"observed_text": "latifa.el-fassi@example.org\nTanger\nSkills\nFollow procedures to control\nsubstances hazardous to health |\nMaintain records of financial\ntransactions | Types of vehicles |\nIdentify vehicles service\nrequirements | Order supplies f... | {
"dossier_gate": "accepted_for_demo",
"input_profile": {
"source_type": "synthetic_cv_offer_pair",
"extraction_channel": "pdf_text",
"language_mix": [
"en"
],
"pii_status": "synthetic_pseudonymized"
},
"controls": {
"schema_valid": true,
"pii_removed": true,
"unsafe_conten... | {
"phi_gt": 1,
"likert": [],
"likert_sources": [],
"likert_final": 5,
"score_gt": 1,
"rank_gt": 2,
"label": "match",
"dimension_scores": {
"technique": 5,
"experience": 4,
"formation": 3,
"langues": 4,
"certifications": 1,
"transversal": 3
},
"must_have": [
{
"id": ... | {
"status": "to_recompute",
"code_version": null,
"layer1": null,
"layer2": null,
"layer3": null,
"decision_path_observed": null
} | {
"fairness_group": null,
"format_group": null,
"adversarial_parent": null,
"cv_reuse_group": null
} | {
"intended_likert": 5,
"intended_phi": 1,
"hard_case": null,
"hard_case_variant": null,
"legit_case": null,
"pool_slot": 15,
"persona_id": "PER_00002",
"cross_language": false,
"is_ood_mission": false,
"attack": null,
"counterfactual": null,
"panel": null
} | Cas : profil ordinaire. Critères impératifs satisfaits : 3/3. Note par construction : 5/5 (match), phi = 1. Valeurs du système BRAINHR+ à recalculer avec son code (pipeline_trace). | {
"cv": "train/CAND_00002.pdf"
} |
2.0.0 | BRAINHR2_P00003 | M0001 | train | standard | accepted | {
"generator": "A:qwen3-4b-awq@74d4bd2bd4bf",
"prompt_version": "cv_phrase@1.4.0|slots@1.2",
"offer_generator": "B:ministral-3-3b-instruct-2512@b35d4dfe56c1",
"offer_prompt_version": "offer_intro@1.1.0",
"style": "sober",
"seed": 1010932766,
"template_id": "T19-linkedin",
"doc_format": "email_or_linkedi... | {
"id": "OFFER_0001",
"title": "Industrial Maintenance Supervisor",
"rome_code": "I1332",
"rome_label": "Chef / Cheffe d'atelier de maintenance industrielle",
"family": "H_I_industrie_maintenance",
"seniority_target": "confirme",
"market": "MA",
"language": "en",
"city": "Tanger",
"contract": "CDI",... | {
"id": "CAND_00003",
"language": "en",
"as_of": "2026-10",
"observed_text": "Contact\nrania_el-idrissi@example.net\n28 avenue Hassan II, Rabat\nDate of birth: 11 December 2003\nTop Skills\nFollow procedures to control\nsubstances hazardous to\nhealth, Teamwork principles,\nApply health and safety\nstandards\nS... | {
"dossier_gate": "accepted_for_demo",
"input_profile": {
"source_type": "synthetic_cv_offer_pair",
"extraction_channel": "pdf_text",
"language_mix": [
"en"
],
"pii_status": "synthetic_pseudonymized"
},
"controls": {
"schema_valid": true,
"pii_removed": true,
"unsafe_conten... | {
"phi_gt": 1,
"likert": [],
"likert_sources": [],
"likert_final": 5,
"score_gt": 1,
"rank_gt": 3,
"label": "match",
"dimension_scores": {
"technique": 5,
"experience": 4,
"formation": 3,
"langues": 4,
"certifications": 1,
"transversal": 3
},
"must_have": [
{
"id": ... | {
"status": "to_recompute",
"code_version": null,
"layer1": null,
"layer2": null,
"layer3": null,
"decision_path_observed": null
} | {
"fairness_group": null,
"format_group": null,
"adversarial_parent": null,
"cv_reuse_group": null
} | {
"intended_likert": 5,
"intended_phi": 1,
"hard_case": null,
"hard_case_variant": null,
"legit_case": null,
"pool_slot": 9,
"persona_id": "PER_00003",
"cross_language": false,
"is_ood_mission": false,
"attack": null,
"counterfactual": null,
"panel": null
} | Cas : profil ordinaire. Critères impératifs satisfaits : 3/3. Note par construction : 5/5 (match), phi = 1. Valeurs du système BRAINHR+ à recalculer avec son code (pipeline_trace). | {
"cv": "train/CAND_00003.pdf"
} |
2.0.0 | BRAINHR2_P00004 | M0001 | train | career_change_transferable | accepted | {
"generator": "B:ministral-3-3b-instruct-2512@b35d4dfe56c1",
"prompt_version": "cv_phrase@1.4.0|slots@1.2",
"offer_generator": "B:ministral-3-3b-instruct-2512@b35d4dfe56c1",
"offer_prompt_version": "offer_intro@1.1.0",
"style": "light_typos",
"seed": 547999374,
"template_id": "T02-classic",
"doc_format... | {
"id": "OFFER_0001",
"title": "Industrial Maintenance Supervisor",
"rome_code": "I1332",
"rome_label": "Chef / Cheffe d'atelier de maintenance industrielle",
"family": "H_I_industrie_maintenance",
"seniority_target": "confirme",
"market": "MA",
"language": "en",
"city": "Tanger",
"contract": "CDI",... | {
"id": "CAND_00004",
"language": "en",
"as_of": "2026-10",
"observed_text": "Sofia Lazrak\nsofia-lazrak58@example.net\nTanger\nDate of birth: 15 November 2001\nProfile\nDetail-oriented professional ensuring machinery reliability and operational efficiency through proactive diagnostics and\nteam-driven solution... | {
"dossier_gate": "accepted_for_demo",
"input_profile": {
"source_type": "synthetic_cv_offer_pair",
"extraction_channel": "pdf_text",
"language_mix": [
"en"
],
"pii_status": "synthetic_pseudonymized"
},
"controls": {
"schema_valid": true,
"pii_removed": true,
"unsafe_conten... | {
"phi_gt": 1,
"likert": [],
"likert_sources": [],
"likert_final": 4,
"score_gt": 0.75,
"rank_gt": 4,
"label": "match",
"dimension_scores": {
"technique": 4,
"experience": 5,
"formation": 3,
"langues": 4,
"certifications": 1,
"transversal": 3
},
"must_have": [
{
"id... | {
"status": "to_recompute",
"code_version": null,
"layer1": null,
"layer2": null,
"layer3": null,
"decision_path_observed": null
} | {
"fairness_group": null,
"format_group": null,
"adversarial_parent": null,
"cv_reuse_group": null
} | {
"intended_likert": 4,
"intended_phi": 1,
"hard_case": "career_change_transferable",
"hard_case_variant": "recent_reconversion",
"legit_case": null,
"pool_slot": 2,
"persona_id": "PER_00004",
"cross_language": false,
"is_ood_mission": false,
"attack": null,
"counterfactual": null,
"panel": null... | Cas : reconversion avec compétences transférables. Critères impératifs satisfaits : 3/3. Note par construction : 4/5 (match), phi = 1. Valeurs du système BRAINHR+ à recalculer avec son code (pipeline_trace). | {
"cv": "train/CAND_00004.pdf"
} |
2.0.0 | BRAINHR2_P00005 | M0001 | train | nicetohave_penalty | accepted_penalized | {
"generator": "A:qwen3-4b-awq@74d4bd2bd4bf",
"prompt_version": "cv_phrase@1.4.0|slots@1.2",
"offer_generator": "B:ministral-3-3b-instruct-2512@b35d4dfe56c1",
"offer_prompt_version": "offer_intro@1.1.0",
"style": "verbose",
"seed": 1839101542,
"template_id": "T07-sidebar",
"doc_format": "pdf_2col",
"o... | {
"id": "OFFER_0001",
"title": "Industrial Maintenance Supervisor",
"rome_code": "I1332",
"rome_label": "Chef / Cheffe d'atelier de maintenance industrielle",
"family": "H_I_industrie_maintenance",
"seniority_target": "confirme",
"market": "MA",
"language": "en",
"city": "Tanger",
"contract": "CDI",... | {
"id": "CAND_00005",
"language": "en",
"as_of": "2026-10",
"observed_text": "siham-el-idrissi@example.net\n81 rue Al Massira, Tanger\nDate of birth: 22 May 1996\nSKILLS\nGMAO ; Order supplies for vehicle\nmaintenance and repair ; Follow\nprocedures to control substances\nhazardous to health\nSoft skills: Team ... | {
"dossier_gate": "accepted_for_demo",
"input_profile": {
"source_type": "synthetic_cv_offer_pair",
"extraction_channel": "pdf_text",
"language_mix": [
"en"
],
"pii_status": "synthetic_pseudonymized"
},
"controls": {
"schema_valid": true,
"pii_removed": true,
"unsafe_conten... | {
"phi_gt": 1,
"likert": [],
"likert_sources": [],
"likert_final": 4,
"score_gt": 0.75,
"rank_gt": 5,
"label": "match",
"dimension_scores": {
"technique": 4,
"experience": 5,
"formation": 3,
"langues": 4,
"certifications": 1,
"transversal": 3
},
"must_have": [
{
"id... | {
"status": "to_recompute",
"code_version": null,
"layer1": null,
"layer2": null,
"layer3": null,
"decision_path_observed": null
} | {
"fairness_group": null,
"format_group": null,
"adversarial_parent": null,
"cv_reuse_group": null
} | {
"intended_likert": 4,
"intended_phi": 1,
"hard_case": "nicetohave_penalty",
"hard_case_variant": null,
"legit_case": null,
"pool_slot": 8,
"persona_id": "PER_00005",
"cross_language": false,
"is_ood_mission": false,
"attack": null,
"counterfactual": null,
"panel": null
} | Cas : aucun atout de l'offre. Critères impératifs satisfaits : 3/3. Note par construction : 4/5 (match), phi = 1. Valeurs du système BRAINHR+ à recalculer avec son code (pipeline_trace). | {
"cv": "train/CAND_00005.pdf"
} |
2.0.0 | BRAINHR2_P00006 | M0001 | train | standard | accepted | {
"generator": "A:qwen3-4b-awq@74d4bd2bd4bf",
"prompt_version": "cv_phrase@1.4.0|slots@1.2",
"offer_generator": "B:ministral-3-3b-instruct-2512@b35d4dfe56c1",
"offer_prompt_version": "offer_intro@1.1.0",
"style": "verbose",
"seed": 1982010755,
"template_id": "T08-sidebar",
"doc_format": "pdf_2col",
"o... | {
"id": "OFFER_0001",
"title": "Industrial Maintenance Supervisor",
"rome_code": "I1332",
"rome_label": "Chef / Cheffe d'atelier de maintenance industrielle",
"family": "H_I_industrie_maintenance",
"seniority_target": "confirme",
"market": "MA",
"language": "en",
"city": "Tanger",
"contract": "CDI",... | {
"id": "CAND_00006",
"language": "en",
"as_of": "2026-10",
"observed_text": "ilyas-ben-youssef@example.org\nTanger\nSkills\nIdentify vehicles service\nrequirements, Monitor vehicle\nmaintenance activities\nSoft skills: Team spirit,\nPrioritisation\nLanguages\nArabic: native, French: C2, \nEnglish: B2\nInterest... | {
"dossier_gate": "accepted_for_demo",
"input_profile": {
"source_type": "synthetic_cv_offer_pair",
"extraction_channel": "pdf_text",
"language_mix": [
"en"
],
"pii_status": "synthetic_pseudonymized"
},
"controls": {
"schema_valid": true,
"pii_removed": true,
"unsafe_conten... | {
"phi_gt": 1,
"likert": [],
"likert_sources": [],
"likert_final": 4,
"score_gt": 0.75,
"rank_gt": 6,
"label": "match",
"dimension_scores": {
"technique": 4,
"experience": 5,
"formation": 3,
"langues": 4,
"certifications": 1,
"transversal": 3
},
"must_have": [
{
"id... | {
"status": "to_recompute",
"code_version": null,
"layer1": null,
"layer2": null,
"layer3": null,
"decision_path_observed": null
} | {
"fairness_group": null,
"format_group": null,
"adversarial_parent": null,
"cv_reuse_group": null
} | {
"intended_likert": 4,
"intended_phi": 1,
"hard_case": null,
"hard_case_variant": "seniority_gap_1",
"legit_case": null,
"pool_slot": 4,
"persona_id": "PER_00006",
"cross_language": false,
"is_ood_mission": false,
"attack": null,
"counterfactual": null,
"panel": null
} | Cas : profil ordinaire. Critères impératifs satisfaits : 3/3. Note par construction : 4/5 (match), phi = 1. Plafonds appliqués : seniority_gap_1 → 4. Valeurs du système BRAINHR+ à recalculer avec son code (pipeline_trace). | {
"cv": "train/CAND_00006.pdf"
} |
2.0.0 | BRAINHR2_P00007 | M0001 | train | standard | accepted | {
"generator": "B:ministral-3-3b-instruct-2512@b35d4dfe56c1",
"prompt_version": "cv_phrase@1.4.0|slots@1.2",
"offer_generator": "B:ministral-3-3b-instruct-2512@b35d4dfe56c1",
"offer_prompt_version": "offer_intro@1.1.0",
"style": "light_typos",
"seed": 1359551129,
"template_id": "T09-sidebar",
"doc_forma... | {
"id": "OFFER_0001",
"title": "Industrial Maintenance Supervisor",
"rome_code": "I1332",
"rome_label": "Chef / Cheffe d'atelier de maintenance industrielle",
"family": "H_I_industrie_maintenance",
"seniority_target": "confirme",
"market": "MA",
"language": "en",
"city": "Tanger",
"contract": "CDI",... | {
"id": "CAND_00007",
"language": "en",
"as_of": "2026-10",
"observed_text": "fatima-bensaid10@example.com\nTanger\nSkills\nManage claims process | Follow\nprocedures to control substances\nhazardous to health | Teamwork\nprinciples | Monitor vehicle\nmaintenance activities\nSoft skills: Team spirit\nLanguages\... | {
"dossier_gate": "accepted_for_demo",
"input_profile": {
"source_type": "synthetic_cv_offer_pair",
"extraction_channel": "pdf_text",
"language_mix": [
"en"
],
"pii_status": "synthetic_pseudonymized"
},
"controls": {
"schema_valid": true,
"pii_removed": true,
"unsafe_conten... | {
"phi_gt": 1,
"likert": [],
"likert_sources": [],
"likert_final": 4,
"score_gt": 0.75,
"rank_gt": 7,
"label": "match",
"dimension_scores": {
"technique": 4,
"experience": 4,
"formation": 3,
"langues": 4,
"certifications": 1,
"transversal": 3
},
"must_have": [
{
"id... | {
"status": "to_recompute",
"code_version": null,
"layer1": null,
"layer2": null,
"layer3": null,
"decision_path_observed": null
} | {
"fairness_group": null,
"format_group": null,
"adversarial_parent": null,
"cv_reuse_group": null
} | {
"intended_likert": 4,
"intended_phi": 1,
"hard_case": null,
"hard_case_variant": "few_nice_to_have",
"legit_case": null,
"pool_slot": 19,
"persona_id": "PER_00007",
"cross_language": false,
"is_ood_mission": false,
"attack": null,
"counterfactual": null,
"panel": null
} | Cas : profil ordinaire. Critères impératifs satisfaits : 3/3. Note par construction : 4/5 (match), phi = 1. Valeurs du système BRAINHR+ à recalculer avec son code (pipeline_trace). | {
"cv": "train/CAND_00007.pdf"
} |
2.0.0 | BRAINHR2_P00008 | M0001 | train | over_qualified | reranked_down | {
"generator": "A:qwen3-4b-awq@74d4bd2bd4bf",
"prompt_version": "cv_phrase@1.4.0|slots@1.2",
"offer_generator": "B:ministral-3-3b-instruct-2512@b35d4dfe56c1",
"offer_prompt_version": "offer_intro@1.1.0",
"style": "fr_en_mix",
"seed": 1022011027,
"template_id": "T03-classic",
"doc_format": "pdf_1col",
... | {
"id": "OFFER_0001",
"title": "Industrial Maintenance Supervisor",
"rome_code": "I1332",
"rome_label": "Chef / Cheffe d'atelier de maintenance industrielle",
"family": "H_I_industrie_maintenance",
"seniority_target": "confirme",
"market": "MA",
"language": "en",
"city": "Tanger",
"contract": "CDI",... | {
"id": "CAND_00008",
"language": "en",
"as_of": "2026-10",
"observed_text": "Rania Zouhir\nrania-zouhir@example.com\nTanger\nDate of birth: 25 September 1987\nProfile\nExperienced in vehicle maintenance, with a focus on supervision and customer service.\nProfessional experience\nVehicle Maintenance Supervisor,... | {
"dossier_gate": "accepted_for_demo",
"input_profile": {
"source_type": "synthetic_cv_offer_pair",
"extraction_channel": "pdf_text",
"language_mix": [
"en"
],
"pii_status": "synthetic_pseudonymized"
},
"controls": {
"schema_valid": true,
"pii_removed": true,
"unsafe_conten... | {
"phi_gt": 1,
"likert": [],
"likert_sources": [],
"likert_final": 3,
"score_gt": 0.5,
"rank_gt": 8,
"label": "partial_match",
"dimension_scores": {
"technique": 4,
"experience": 5,
"formation": 3,
"langues": 4,
"certifications": 1,
"transversal": 3
},
"must_have": [
{
... | {
"status": "to_recompute",
"code_version": null,
"layer1": null,
"layer2": null,
"layer3": null,
"decision_path_observed": null
} | {
"fairness_group": null,
"format_group": null,
"adversarial_parent": null,
"cv_reuse_group": null
} | {
"intended_likert": 3,
"intended_phi": 1,
"hard_case": "over_qualified",
"hard_case_variant": "level_1_to_3",
"legit_case": null,
"pool_slot": 17,
"persona_id": "PER_00008",
"cross_language": false,
"is_ood_mission": false,
"attack": null,
"counterfactual": null,
"panel": null
} | Cas : sur-qualification. Critères impératifs satisfaits : 3/3. Note par construction : 3/5 (partial_match), phi = 1. Plafonds appliqués : over_qualified → 3. Valeurs du système BRAINHR+ à recalculer avec son code (pipeline_trace). | {
"cv": "train/CAND_00008.pdf"
} |
2.0.0 | BRAINHR2_P00009 | M0001 | train | keyword_stuffing | reranked_down | {
"generator": "B:ministral-3-3b-instruct-2512@b35d4dfe56c1",
"prompt_version": "cv_phrase@1.4.0|slots@1.2",
"offer_generator": "B:ministral-3-3b-instruct-2512@b35d4dfe56c1",
"offer_prompt_version": "offer_intro@1.1.0",
"style": "sober",
"seed": 1879070502,
"template_id": "T10-sidebar",
"doc_format": "p... | {
"id": "OFFER_0001",
"title": "Industrial Maintenance Supervisor",
"rome_code": "I1332",
"rome_label": "Chef / Cheffe d'atelier de maintenance industrielle",
"family": "H_I_industrie_maintenance",
"seniority_target": "confirme",
"market": "MA",
"language": "en",
"city": "Tanger",
"contract": "CDI",... | {
"id": "CAND_00009",
"language": "en",
"as_of": "2026-10",
"observed_text": "latifalahbabi@example.net\nTanger\nDate of birth: 4 October 2001\nSkills\nPlan employees work in vehicle maintenance\n· Follow procedures to control substances\nhazardous to health · Maintain records of\nfinancial transactions\nSoft s... | {
"dossier_gate": "accepted_for_demo",
"input_profile": {
"source_type": "synthetic_cv_offer_pair",
"extraction_channel": "pdf_text",
"language_mix": [
"en"
],
"pii_status": "synthetic_pseudonymized"
},
"controls": {
"schema_valid": true,
"pii_removed": true,
"unsafe_conten... | {
"phi_gt": 1,
"likert": [],
"likert_sources": [],
"likert_final": 3,
"score_gt": 0.5,
"rank_gt": 9,
"label": "partial_match",
"dimension_scores": {
"technique": 3,
"experience": 5,
"formation": 3,
"langues": 4,
"certifications": 1,
"transversal": 3
},
"must_have": [
{
... | {
"status": "to_recompute",
"code_version": null,
"layer1": null,
"layer2": null,
"layer3": null,
"decision_path_observed": null
} | {
"fairness_group": null,
"format_group": null,
"adversarial_parent": null,
"cv_reuse_group": null
} | {
"intended_likert": 3,
"intended_phi": 1,
"hard_case": "keyword_stuffing",
"hard_case_variant": "all_technical_listed",
"legit_case": null,
"pool_slot": 12,
"persona_id": "PER_00009",
"cross_language": false,
"is_ood_mission": false,
"attack": null,
"counterfactual": null,
"panel": null
} | Cas : compétences seulement listées, sans mission. Critères impératifs satisfaits : 3/3. Note par construction : 3/5 (partial_match), phi = 1. Plafonds appliqués : MH1 → 3. Preuve faible (compétence listée ou simple participation) pour : MH1. Valeurs du système BRAINHR+ à recalculer avec son code (pipeline_trace). | {
"cv": "train/CAND_00009.pdf"
} |
2.0.0 | BRAINHR2_P00010 | M0001 | train | stale_skill | reranked_down | {
"generator": "B:ministral-3-3b-instruct-2512@b35d4dfe56c1",
"prompt_version": "cv_phrase@1.4.0|slots@1.2",
"offer_generator": "B:ministral-3-3b-instruct-2512@b35d4dfe56c1",
"offer_prompt_version": "offer_intro@1.1.0",
"style": "fr_en_mix",
"seed": 765206780,
"template_id": "T04-classic",
"doc_format":... | {
"id": "OFFER_0001",
"title": "Industrial Maintenance Supervisor",
"rome_code": "I1332",
"rome_label": "Chef / Cheffe d'atelier de maintenance industrielle",
"family": "H_I_industrie_maintenance",
"seniority_target": "confirme",
"market": "MA",
"language": "en",
"city": "Tanger",
"contract": "CDI",... | {
"id": "CAND_00010",
"language": "en",
"as_of": "2026-10",
"observed_text": "Amine Lamrini\namine_lamrini@example.org\nCasablanca\nPROFILE\nProfile: Vehicle Maintenance Supervisor.\nSKILLS\nFollow procedures to control substances hazardous to health, Types of vehicles, Provide customer follow-up services\nSoft... | {
"dossier_gate": "accepted_for_demo",
"input_profile": {
"source_type": "synthetic_cv_offer_pair",
"extraction_channel": "pdf_text",
"language_mix": [
"en"
],
"pii_status": "synthetic_pseudonymized"
},
"controls": {
"schema_valid": true,
"pii_removed": true,
"unsafe_conten... | {
"phi_gt": 1,
"likert": [],
"likert_sources": [],
"likert_final": 3,
"score_gt": 0.5,
"rank_gt": 10,
"label": "partial_match",
"dimension_scores": {
"technique": 3,
"experience": 5,
"formation": 3,
"langues": 4,
"certifications": 1,
"transversal": 3
},
"must_have": [
{
... | {
"status": "to_recompute",
"code_version": null,
"layer1": null,
"layer2": null,
"layer3": null,
"decision_path_observed": null
} | {
"fairness_group": null,
"format_group": null,
"adversarial_parent": null,
"cv_reuse_group": null
} | {
"intended_likert": 3,
"intended_phi": 1,
"hard_case": "stale_skill",
"hard_case_variant": null,
"legit_case": null,
"pool_slot": 16,
"persona_id": "PER_00010",
"cross_language": false,
"is_ood_mission": false,
"attack": null,
"counterfactual": null,
"panel": null
} | Cas : compétence pratiquée il y a plus de cinq ans. Critères impératifs satisfaits : 3/3. Note par construction : 3/5 (partial_match), phi = 1. Plafonds appliqués : MH1 → 3. Preuve faible (compétence listée ou simple participation) pour : MH1. Valeurs du système BRAINHR+ à recalculer avec son code (pipeline_trace). | {
"cv": "train/CAND_00010.pdf"
} |
2.0.0 | BRAINHR2_P00011 | M0001 | train | legitimate_gap | reranked_down | {
"generator": "A:qwen3-4b-awq@74d4bd2bd4bf",
"prompt_version": "cv_phrase@1.4.0|slots@1.2",
"offer_generator": "B:ministral-3-3b-instruct-2512@b35d4dfe56c1",
"offer_prompt_version": "offer_intro@1.1.0",
"style": "light_typos",
"seed": 926650012,
"template_id": "T05-classic",
"doc_format": "pdf_1col",
... | {
"id": "OFFER_0001",
"title": "Industrial Maintenance Supervisor",
"rome_code": "I1332",
"rome_label": "Chef / Cheffe d'atelier de maintenance industrielle",
"family": "H_I_industrie_maintenance",
"seniority_target": "confirme",
"market": "MA",
"language": "en",
"city": "Tanger",
"contract": "CDI",... | {
"id": "CAND_00011",
"language": "en",
"as_of": "2026-10",
"observed_text": "Mustapha Rachidi\nmustapha-rachidi97@example.net\n9 boulevard Zerktouni, Tanger\nProfile\nExperienced in vehicle maintenance and industrial supervision, with a focus on efficiency and customer\nservice.\nProfessional experience\nVehic... | {
"dossier_gate": "accepted_for_demo",
"input_profile": {
"source_type": "synthetic_cv_offer_pair",
"extraction_channel": "pdf_text",
"language_mix": [
"en"
],
"pii_status": "synthetic_pseudonymized"
},
"controls": {
"schema_valid": true,
"pii_removed": true,
"unsafe_conten... | {
"phi_gt": 1,
"likert": [],
"likert_sources": [],
"likert_final": 3,
"score_gt": 0.5,
"rank_gt": 11,
"label": "partial_match",
"dimension_scores": {
"technique": 3,
"experience": 5,
"formation": 3,
"langues": 4,
"certifications": 1,
"transversal": 3
},
"must_have": [
{
... | {
"status": "to_recompute",
"code_version": null,
"layer1": null,
"layer2": null,
"layer3": null,
"decision_path_observed": null
} | {
"fairness_group": null,
"format_group": null,
"adversarial_parent": null,
"cv_reuse_group": null
} | {
"intended_likert": 3,
"intended_phi": 1,
"hard_case": null,
"hard_case_variant": "weak_evidence",
"legit_case": "caregiver",
"pool_slot": 0,
"persona_id": "PER_00011",
"cross_language": false,
"is_ood_mission": false,
"attack": null,
"counterfactual": null,
"panel": null
} | Cas : trou légitime (non pénalisant). Critères impératifs satisfaits : 3/3. Note par construction : 3/5 (partial_match), phi = 1. Plafonds appliqués : MH1 → 3. Preuve faible (compétence listée ou simple participation) pour : MH1. Trou ou chevauchement légitime : ne baisse jamais la note (règle 7). Valeurs du système BR... | {
"cv": "train/CAND_00011.pdf"
} |
BRAINHR+ corpus v2 (privé)
Corpus synthétique de paires CV–offre d'emploi, en français et en anglais, construit pour la révision de l'article
« Evidence-grounded recruitment screening » (Array). Aucune donnée personnelle réelle : identités
fictives, e-mails sur example.com/.org/.net, téléphones des plages ARCEP réservées à la fiction ; le texte vu par le
modèle (cv.model_input_text) est pseudonymisé.
Contenu
| Split | Paires | Assertions | Adversarial |
|---|---|---|---|
| train | 3600 | 12202 | 400 |
| dev | 400 | 2183 | 0 |
| test | 800 | 4642 | 1500 |
Les fichiers du test (paires, suites de test, exports d'évaluation) ne sont pas encore publiés : ils restent locaux tant que l'annotation humaine à l'aveugle du test (test-H) n'est pas terminée. Leurs empreintes SHA-256 sont dans MANIFEST.sha256 et FREEZE.json : le test est gelé.
Labels : construction déterministe depuis une spécification (vérité terrain), par des règles documentées
(labeling/rules.py), jamais éditées à la main. Le panel de deux juges LLM n'a été mesuré que sur le dev, avec de petits
modèles locaux trop indulgents pour servir de filtre ; il n'a écarté ni modifié aucune paire (label_source =
construction). L'annotation humaine (test-H) est à venir. pipeline_trace vaut to_recompute : aucune valeur du
système BRAINHR+ n'est écrite à la main ni par un LLM.
Exports d'entraînement (format chat)
| Rôle | Train | Dev |
|---|---|---|
| extract_cv | 3412 | 400 |
| extract_offer | 180 | 20 |
| mh | 16880 | 1860 |
| nh | 14700 | 1820 |
| rerank | 3600 | 400 |
| verify | 12202 | 2183 |
Chaque ligne : messages (system, user, assistant) et meta (JSON). Les prompts système sont versionnés
(prompts/roles/*@1.0.0.md dans le dépôt de code).
Limites
Données synthétiques (écart synthétique–réel non mesuré sur des données réelles) ; générateurs de 3 à 4 milliards de
paramètres ; panel non exécuté sur le train ni le test ; annotation humaine d'une seule personne (à venir) ; la
longueur du CV prédit en partie la pertinence (AUC 0,62 à 0,71, voir docs/qc_report.md) ; français et anglais
seulement ; marchés France, Maroc et international.
Licence : CC BY-NC 4.0. Code : Apache-2.0. Éditeur : Brain Gen Technology.
- Downloads last month
- 48