Dataset Viewer
Auto-converted to Parquet Duplicate
adversarial
string
annotations
string
art9_special_categories
string
behavioral_signals
string
context_length_tier
string
context_preservation
string
data_type
string
difficulty_level
string
dimensions
string
document_type
string
domain
string
entity_tracking
string
language
string
language_family
string
primary_dimension
string
privacy_risk
string
provenance
string
query_context
string
record_id
string
reg_ccpa_deidentified
string
reg_gdpr
string
reg_hipaa_expert_determination
string
reg_hipaa_safe_harbor
string
reg_pci_dss
string
resource_level
string
schema_version
string
script
string
text
string
tier3_evaluation
string
token_count
int64
version
string
{"difficulty": "clean", "techniques": [], "type": null}
[{"anonymization_strategy": "pseudonymize", "category": "identity_demographics", "cluster_id": null, "context_dependency": "low", "end": 75, "entity_id": "e0", "entity_type": "PERSON_NAME", "information_anchor_score": 0.78, "mention_variant": "full_name", "nested_entities": null, "sensitivity_class": "direct_identifier...
["HEALTH_CONDITION", "HEALTH_INSURANCE_ID", "MEDICAL_RECORD_NUMBER", "PROCEDURE_NAME"]
{"behavioral_signal_density": 0.7333, "interest_topics": {"indicators": ["topic:tech_enthusiast:1", "topic:fitness:1"], "present": true, "uniqueness": "high"}, "location_signals": {"indicators": [], "present": false, "uniqueness": "none"}, "personal_anecdote": {"indicators": [], "present": false, "uniqueness": "none"},...
medium
{"_caveat": "No variant is demonstrated anonymised under GDPR Recital 26. anonymized_pseudonymized and anonymized_llm_sanitized remain PERSONAL DATA (Art.4(5)). residual_quasi_identifier flags a partial identifier (e.g. ***-**-NNNN SSN tail) retained in a variant — a quasi-identifier, not a safe endpoint. Synthetic-onl...
unstructured_text
hard
["context_preservation", "diverse_pii_types"]
insurance_claim
clinical
{"coreference_chains": [], "num_distinct_persons": 3, "num_repeated_entities": 0, "tracking_difficulty": "none"}
en
Germanic
diverse_pii_types
{"_caveat": "k_anonymity_estimate is a quasi-identifier-TYPE-count proxy (1->100, 2->20, 3->5, 4+->2), NOT an equivalence-class k computed over QI values; honest alias: qid_type_count_risk_label; null where quasi-identifiers are absent.", "estimated_reid_recall": 0.6233, "k_anonymity_estimate": 100, "qid_type_count_ris...
{"generation_seed": 42, "license": "CC0-1.0", "source_type": "synthetic", "v1_3_0_record_id": "0f70e07a-5ead-45f8-ae9e-da1cd7d3b293", "v1_record_id": null}
null
e67353b0-dd23-5f8e-89af-8607cf02957a
out_of_scope_of_dataset
in_scope
in_scope
in_scope
out_of_scope_of_dataset
high
2.2.0
Latn
INSURANCE CLAIM NARRATIVE Claim #: CLM-2025-871477 Patient: Margaret Perez | DOB: 1964-10-22 | MRN: MRN-2454875 Insurance ID: INS-233150837 | Policy: POL-8303037 Provider: Dr. Amanda King | NPI: 7126214331 NARRATIVE: Margaret Perez presented on 2024-01-19 with symptoms consistent with Asthma. After examination and di...
{"_caveat": "re_identification_resistance_score is a HEURISTIC PRIOR (closed-form over behavioral_signal_density), NOT a measured attack; exposure_index_prior is its honest alias. Exposure index is a DETERMINISTIC pre-screen PRIOR computed from behavioral-signal annotations. It is NOT measured RRS and MUST NOT be repor...
78
2.2.0
{"difficulty": "clean", "techniques": [], "type": null}
[{"anonymization_strategy": "pseudonymize", "category": "identity_demographics", "cluster_id": null, "context_dependency": "low", "end": 27, "entity_id": "e0", "entity_type": "PERSON_NAME", "information_anchor_score": 0.84, "mention_variant": "full_name", "sensitivity_class": "direct_identifier", "start": 11, "text": "...
[]
{"behavioral_signal_density": 0.1, "interest_topics": {"indicators": [], "present": false, "uniqueness": "none"}, "location_signals": {"indicators": [], "present": false, "uniqueness": "none"}, "personal_anecdote": {"indicators": [], "present": false, "uniqueness": "none"}, "professional_domain": {"indicators": [], "pr...
short
{"_caveat": "No variant is demonstrated anonymised under GDPR Recital 26. anonymized_pseudonymized and anonymized_llm_sanitized remain PERSONAL DATA (Art.4(5)). residual_quasi_identifier flags a partial identifier (e.g. ***-**-NNNN SSN tail) retained in a variant — a quasi-identifier, not a safe endpoint. Synthetic-onl...
unstructured_text
easy
["context_preservation", "diverse_pii_types", "multilingual"]
null
general
{"coreference_chains": [], "num_distinct_persons": 1, "num_repeated_entities": 0, "tracking_difficulty": "none"}
ru
Slavic
diverse_pii_types
{"_caveat": "k_anonymity_estimate is a quasi-identifier-TYPE-count proxy (1->100, 2->20, 3->5, 4+->2), NOT an equivalence-class k computed over QI values; honest alias: qid_type_count_risk_label; null where quasi-identifiers are absent.", "estimated_reid_recall": 0.085, "k_anonymity_estimate": 100, "qid_type_count_risk...
{"generation_seed": 4482037211297612090, "lattice_cell_id": "LxE:entity_type=BANK_ROUTING_NUMBER|language=ru", "lattice_seed": 91237, "lattice_tier": "critical", "lattice_version": "1.0.0", "license": "CC0-1.0", "source_type": "synthetic_lattice_enrichment", "v1_3_0_record_id": "2f3b427a-d136-474f-94dd-ef90bbe00796", "...
null
29940551-9ea4-5a9b-9206-cda59f97b7c1
out_of_scope_of_dataset
in_scope
out_of_scope_of_dataset
out_of_scope_of_dataset
out_of_scope_of_dataset
high
2.2.0
Cyrl
Record for Светлана Морозов, Bank Routing Number: 564607960, contact светлана.морозов7@hotmail.com. [#3e336518c97a013a-1122]
{"_caveat": "re_identification_resistance_score is a HEURISTIC PRIOR (closed-form over behavioral_signal_density), NOT a measured attack; exposure_index_prior is its honest alias. Exposure index is a DETERMINISTIC pre-screen PRIOR computed from behavioral-signal annotations. It is NOT measured RRS and MUST NOT be repor...
11
2.2.0
{"difficulty": "hard", "techniques": ["multi_token_entity", "compound_name"], "type": "multi_token"}
[{"anonymization_strategy": "pseudonymize", "category": "identity_demographics", "cluster_id": null, "context_dependency": "low", "end": 34, "entity_id": "e0", "entity_type": "PERSON_NAME", "information_anchor_score": 0.87, "mention_variant": null, "nested_entities": null, "sensitivity_class": "direct_identifier", "sta...
[]
{"behavioral_signal_density": 0.11, "interest_topics": {"indicators": [], "present": false, "uniqueness": "none"}, "location_signals": {"indicators": [], "present": false, "uniqueness": "none"}, "personal_anecdote": {"indicators": [], "present": false, "uniqueness": "none"}, "professional_domain": {"indicators": [], "p...
short
{"_caveat": "No variant is demonstrated anonymised under GDPR Recital 26. anonymized_pseudonymized and anonymized_llm_sanitized remain PERSONAL DATA (Art.4(5)). residual_quasi_identifier flags a partial identifier (e.g. ***-**-NNNN SSN tail) retained in a variant — a quasi-identifier, not a safe endpoint. Synthetic-onl...
unstructured_text
challenging
["context_preservation", "edge_cases"]
null
general
{"coreference_chains": [], "num_distinct_persons": 1, "num_repeated_entities": 0, "tracking_difficulty": "none"}
en
Germanic
edge_cases
{"_caveat": "k_anonymity_estimate is a quasi-identifier-TYPE-count proxy (1->100, 2->20, 3->5, 4+->2), NOT an equivalence-class k computed over QI values; honest alias: qid_type_count_risk_label; null where quasi-identifiers are absent.", "estimated_reid_recall": 0.0935, "k_anonymity_estimate": null, "qid_type_count_ri...
{"generation_seed": 42, "license": "CC0-1.0", "source_type": "synthetic", "v1_3_0_record_id": "bd8d2205-ddb0-49b6-82b5-a5f917de9ae9", "v1_record_id": null}
null
8186cd85-7c47-5991-9da8-36a50e903faf
in_scope
in_scope
out_of_scope_of_dataset
out_of_scope_of_dataset
out_of_scope_of_dataset
high
2.2.0
Latn
Patient Jean-Pierre de la Fontaine (jessica.wilson26@mail.com) was seen today. Phone: +1 (887) 229-8651. SSN: 507-86-1262.
{"_caveat": "re_identification_resistance_score is a HEURISTIC PRIOR (closed-form over behavioral_signal_density), NOT a measured attack; exposure_index_prior is its honest alias. Exposure index is a DETERMINISTIC pre-screen PRIOR computed from behavioral-signal annotations. It is NOT measured RRS and MUST NOT be repor...
15
2.2.0
{"difficulty": "clean", "techniques": [], "type": null}
[{"anonymization_strategy": "pseudonymize", "category": "identity_demographics", "cluster_id": null, "context_dependency": "low", "end": 22, "entity_id": "e0", "entity_type": "PERSON_NAME", "information_anchor_score": 0.77, "mention_variant": "full_name", "nested_entities": null, "sensitivity_class": "direct_identifier...
[]
{"behavioral_signal_density": 0.1, "interest_topics": {"indicators": [], "present": false, "uniqueness": "none"}, "location_signals": {"indicators": [], "present": false, "uniqueness": "none"}, "personal_anecdote": {"indicators": [], "present": false, "uniqueness": "none"}, "professional_domain": {"indicators": [], "pr...
short
{"_caveat": "No variant is demonstrated anonymised under GDPR Recital 26. anonymized_pseudonymized and anonymized_llm_sanitized remain PERSONAL DATA (Art.4(5)). residual_quasi_identifier flags a partial identifier (e.g. ***-**-NNNN SSN tail) retained in a variant — a quasi-identifier, not a safe endpoint. Synthetic-onl...
unstructured_text
moderate
["context_preservation", "diverse_pii_types", "multilingual"]
null
general
{"coreference_chains": [], "num_distinct_persons": 1, "num_repeated_entities": 0, "tracking_difficulty": "none"}
ar
Semitic
diverse_pii_types
{"_caveat": "k_anonymity_estimate is a quasi-identifier-TYPE-count proxy (1->100, 2->20, 3->5, 4+->2), NOT an equivalence-class k computed over QI values; honest alias: qid_type_count_risk_label; null where quasi-identifiers are absent.", "estimated_reid_recall": 0.085, "k_anonymity_estimate": null, "qid_type_count_ris...
{"generation_seed": 6301330019468755174, "lattice_cell_id": "LxE:entity_type=CREDIT_CARD_NUMBER|language=ar", "lattice_seed": 91237, "lattice_tier": "critical", "lattice_version": "1.0.0", "license": "CC0-1.0", "source_type": "synthetic_lattice_enrichment", "v1_3_0_record_id": "50efa640-072f-45b3-9e18-6f70c6969821", "v...
null
77bff9df-ad9c-50a7-9f4c-7cf2a0fd06db
out_of_scope_of_dataset
in_scope
out_of_scope_of_dataset
out_of_scope_of_dataset
in_scope
high
2.2.0
Arab
Record for ليلى السعيد, Credit Card Number: 3752 3999 1187 4965, contact ليلى.السعيد39@zoho.com. [#5772d2444ef0c0e6-1155]
{"_caveat": "re_identification_resistance_score is a HEURISTIC PRIOR (closed-form over behavioral_signal_density), NOT a measured attack; exposure_index_prior is its honest alias. Exposure index is a DETERMINISTIC pre-screen PRIOR computed from behavioral-signal annotations. It is NOT measured RRS and MUST NOT be repor...
14
2.2.0
{"difficulty": "moderate", "techniques": ["mixed_patterns"], "type": "mixed"}
[{"anonymization_strategy": "pseudonymize", "category": "identity_demographics", "cluster_id": null, "context_dependency": "low", "end": 11, "entity_id": "e0", "entity_type": "PERSON_NAME", "information_anchor_score": 0.88, "mention_variant": "full_name", "nested_entities": null, "sensitivity_class": "direct_identifier...
[]
{"behavioral_signal_density": 0.1, "interest_topics": {"indicators": [], "present": false, "uniqueness": "none"}, "location_signals": {"indicators": [], "present": false, "uniqueness": "none"}, "personal_anecdote": {"indicators": [], "present": false, "uniqueness": "none"}, "professional_domain": {"indicators": [], "pr...
short
{"_caveat": "No variant is demonstrated anonymised under GDPR Recital 26. anonymized_pseudonymized and anonymized_llm_sanitized remain PERSONAL DATA (Art.4(5)). residual_quasi_identifier flags a partial identifier (e.g. ***-**-NNNN SSN tail) retained in a variant — a quasi-identifier, not a safe endpoint. Synthetic-onl...
unstructured_text
challenging
["context_preservation", "edge_cases"]
null
general
{"coreference_chains": [], "num_distinct_persons": 1, "num_repeated_entities": 0, "tracking_difficulty": "none"}
en
Germanic
edge_cases
{"_caveat": "k_anonymity_estimate is a quasi-identifier-TYPE-count proxy (1->100, 2->20, 3->5, 4+->2), NOT an equivalence-class k computed over QI values; honest alias: qid_type_count_risk_label; null where quasi-identifiers are absent.", "estimated_reid_recall": 0.085, "k_anonymity_estimate": null, "qid_type_count_ris...
{"generation_seed": 42, "license": "CC0-1.0", "source_type": "synthetic", "v1_3_0_record_id": "dd55e08a-e78d-44ba-8f2f-2a2cba168814", "v1_record_id": null}
{"query": "What email was used for registration?", "relevant_entity_ids": ["e1", "e3"]}
4d03c442-f89a-5cd8-b54e-9d46fade1dc4
in_scope
in_scope
out_of_scope_of_dataset
out_of_scope_of_dataset
out_of_scope_of_dataset
high
2.2.0
Latn
Susan Allen, susan.allen16@yahoo.com, +1 (762) 640-5270, 633-13-7100
{"_caveat": "re_identification_resistance_score is a HEURISTIC PRIOR (closed-form over behavioral_signal_density), NOT a measured attack; exposure_index_prior is its honest alias. Exposure index is a DETERMINISTIC pre-screen PRIOR computed from behavioral-signal annotations. It is NOT measured RRS and MUST NOT be repor...
7
2.2.0
{"difficulty": "severe", "techniques": ["zero_width_insertion"], "type": "zero_width_char"}
[{"anonymization_strategy": "pseudonymize", "category": "identity_demographics", "cluster_id": null, "context_dependency": "low", "end": 25, "entity_id": "e0", "entity_type": "PERSON_NAME", "information_anchor_score": 0.79, "mention_variant": null, "nested_entities": null, "sensitivity_class": "direct_identifier", "sta...
[]
{"behavioral_signal_density": 0.1, "interest_topics": {"indicators": [], "present": false, "uniqueness": "none"}, "location_signals": {"indicators": [], "present": false, "uniqueness": "none"}, "personal_anecdote": {"indicators": [], "present": false, "uniqueness": "none"}, "professional_domain": {"indicators": [], "pr...
short
{"_caveat": "No variant is demonstrated anonymised under GDPR Recital 26. anonymized_pseudonymized and anonymized_llm_sanitized remain PERSONAL DATA (Art.4(5)). residual_quasi_identifier flags a partial identifier (e.g. ***-**-NNNN SSN tail) retained in a variant — a quasi-identifier, not a safe endpoint. Synthetic-onl...
unstructured_text
challenging
["context_preservation", "edge_cases"]
null
general
{"coreference_chains": [], "num_distinct_persons": 1, "num_repeated_entities": 0, "tracking_difficulty": "none"}
en
Germanic
edge_cases
{"_caveat": "k_anonymity_estimate is a quasi-identifier-TYPE-count proxy (1->100, 2->20, 3->5, 4+->2), NOT an equivalence-class k computed over QI values; honest alias: qid_type_count_risk_label; null where quasi-identifiers are absent.", "estimated_reid_recall": 0.085, "k_anonymity_estimate": null, "qid_type_count_ris...
{"generation_seed": 42, "license": "CC0-1.0", "source_type": "synthetic", "v1_3_0_record_id": "3adb7c4d-c78b-4971-b45c-bd4e57f687d9", "v1_record_id": null}
null
ed2be924-aeaf-53d4-b114-65d950a557ec
in_scope
in_scope
out_of_scope_of_dataset
out_of_scope_of_dataset
out_of_scope_of_dataset
high
2.2.0
Latn
Contact James ‍Rodrigue​z at james.rodriguez50@mail.com or +1 (851) 679-6726. SSN: 689-20-8674.
{"_caveat": "re_identification_resistance_score is a HEURISTIC PRIOR (closed-form over behavioral_signal_density), NOT a measured attack; exposure_index_prior is its honest alias. Exposure index is a DETERMINISTIC pre-screen PRIOR computed from behavioral-signal annotations. It is NOT measured RRS and MUST NOT be repor...
11
2.2.0
{"difficulty": "moderate", "techniques": ["base64_encoding"], "type": "base64_encoding"}
[{"anonymization_strategy": "pseudonymize", "category": "identity_demographics", "cluster_id": null, "context_dependency": "low", "end": 26, "entity_id": "e0", "entity_type": "PERSON_NAME", "information_anchor_score": 0.84, "mention_variant": "full_name", "nested_entities": null, "sensitivity_class": "direct_identifier...
[]
{"behavioral_signal_density": 0.1, "interest_topics": {"indicators": [], "present": false, "uniqueness": "none"}, "location_signals": {"indicators": [], "present": false, "uniqueness": "none"}, "personal_anecdote": {"indicators": [], "present": false, "uniqueness": "none"}, "professional_domain": {"indicators": [], "pr...
short
{"_caveat": "No variant is demonstrated anonymised under GDPR Recital 26. anonymized_pseudonymized and anonymized_llm_sanitized remain PERSONAL DATA (Art.4(5)). residual_quasi_identifier flags a partial identifier (e.g. ***-**-NNNN SSN tail) retained in a variant — a quasi-identifier, not a safe endpoint. Synthetic-onl...
unstructured_text
easy
["context_preservation", "edge_cases"]
null
general
{"coreference_chains": [], "num_distinct_persons": 1, "num_repeated_entities": 0, "tracking_difficulty": "none"}
en
Germanic
edge_cases
{"_caveat": "k_anonymity_estimate is a quasi-identifier-TYPE-count proxy (1->100, 2->20, 3->5, 4+->2), NOT an equivalence-class k computed over QI values; honest alias: qid_type_count_risk_label; null where quasi-identifiers are absent.", "estimated_reid_recall": 0.085, "k_anonymity_estimate": null, "qid_type_count_ris...
{"generation_seed": 14579636618877719490, "lattice_cell_id": "AxE:adversarial=base64_encoding|entity_type=TAX_ID", "lattice_seed": 91237, "lattice_tier": "standard", "lattice_version": "1.0.0", "license": "CC0-1.0", "source_type": "synthetic_lattice_enrichment", "v1_3_0_record_id": "4c18e2b5-07f9-417b-b2d0-c936014a69d0...
null
56bc5917-b90b-5a63-9fcf-5d56484f4f80
out_of_scope_of_dataset
in_scope
out_of_scope_of_dataset
out_of_scope_of_dataset
out_of_scope_of_dataset
high
2.2.0
Latn
Record for Patricia Wilson, Tax Id: ODgtMzY1MTEyMg==, contact patricia.wilson81@fastmail.com. [#ca554642367bd3c2-490]
{"_caveat": "re_identification_resistance_score is a HEURISTIC PRIOR (closed-form over behavioral_signal_density), NOT a measured attack; exposure_index_prior is its honest alias. Exposure index is a DETERMINISTIC pre-screen PRIOR computed from behavioral-signal annotations. It is NOT measured RRS and MUST NOT be repor...
10
2.2.0
{"difficulty": "clean", "techniques": [], "type": null}
[{"anonymization_strategy": "pseudonymize", "category": "identity_demographics", "cluster_id": null, "context_dependency": "low", "end": 25, "entity_id": "e0", "entity_type": "PERSON_NAME", "information_anchor_score": 0.8, "mention_variant": "full_name", "nested_entities": null, "sensitivity_class": "direct_identifier"...
[]
{"behavioral_signal_density": 0.2667, "interest_topics": {"indicators": ["topic:fitness:1"], "present": true, "uniqueness": "moderate"}, "location_signals": {"indicators": [], "present": false, "uniqueness": "none"}, "personal_anecdote": {"indicators": [], "present": false, "uniqueness": "none"}, "professional_domain":...
short
{"_caveat": "No variant is demonstrated anonymised under GDPR Recital 26. anonymized_pseudonymized and anonymized_llm_sanitized remain PERSONAL DATA (Art.4(5)). residual_quasi_identifier flags a partial identifier (e.g. ***-**-NNNN SSN tail) retained in a variant — a quasi-identifier, not a safe endpoint. Synthetic-onl...
unstructured_text
moderate
["context_preservation", "diverse_pii_types", "multilingual"]
null
general
{"coreference_chains": [], "num_distinct_persons": 1, "num_repeated_entities": 0, "tracking_difficulty": "none"}
de
Germanic
diverse_pii_types
{"_caveat": "k_anonymity_estimate is a quasi-identifier-TYPE-count proxy (1->100, 2->20, 3->5, 4+->2), NOT an equivalence-class k computed over QI values; honest alias: qid_type_count_risk_label; null where quasi-identifiers are absent.", "estimated_reid_recall": 0.2267, "k_anonymity_estimate": null, "qid_type_count_ri...
{"generation_seed": 5505032973486469843, "lattice_cell_id": "LxE:entity_type=CVV|language=de", "lattice_seed": 91237, "lattice_tier": "critical", "lattice_version": "1.0.0", "license": "CC0-1.0", "source_type": "synthetic_lattice_enrichment", "v1_3_0_record_id": "1e507597-8244-4192-b223-4f03e65907e2", "v1_record_id": n...
null
bd25a465-2e83-5f55-a666-29c96ccf07ff
out_of_scope_of_dataset
in_scope
out_of_scope_of_dataset
out_of_scope_of_dataset
out_of_scope_of_dataset
high
2.2.0
Latn
Record for Brigitte Klein, CVV: 949, contact brigitte.klein49@protonmail.com. [#4c65ce528d7932d3-776]
{"_caveat": "re_identification_resistance_score is a HEURISTIC PRIOR (closed-form over behavioral_signal_density), NOT a measured attack; exposure_index_prior is its honest alias. Exposure index is a DETERMINISTIC pre-screen PRIOR computed from behavioral-signal annotations. It is NOT measured RRS and MUST NOT be repor...
9
2.2.0
{"difficulty": "clean", "techniques": [], "type": null}
[{"anonymization_strategy": "pseudonymize", "category": "identity_demographics", "cluster_id": null, "context_dependency": "low", "end": 23, "entity_id": "e0", "entity_type": "PERSON_NAME", "information_anchor_score": 0.89, "mention_variant": "full_name", "sensitivity_class": "direct_identifier", "start": 11, "text": "...
[]
{"behavioral_signal_density": 0.1, "interest_topics": {"indicators": [], "present": false, "uniqueness": "none"}, "location_signals": {"indicators": [], "present": false, "uniqueness": "none"}, "personal_anecdote": {"indicators": [], "present": false, "uniqueness": "none"}, "professional_domain": {"indicators": [], "pr...
short
{"_caveat": "No variant is demonstrated anonymised under GDPR Recital 26. anonymized_pseudonymized and anonymized_llm_sanitized remain PERSONAL DATA (Art.4(5)). residual_quasi_identifier flags a partial identifier (e.g. ***-**-NNNN SSN tail) retained in a variant — a quasi-identifier, not a safe endpoint. Synthetic-onl...
unstructured_text
moderate
["context_preservation", "diverse_pii_types", "multilingual"]
null
general
{"coreference_chains": [], "num_distinct_persons": 1, "num_repeated_entities": 0, "tracking_difficulty": "none"}
ru
Slavic
diverse_pii_types
{"_caveat": "k_anonymity_estimate is a quasi-identifier-TYPE-count proxy (1->100, 2->20, 3->5, 4+->2), NOT an equivalence-class k computed over QI values; honest alias: qid_type_count_risk_label; null where quasi-identifiers are absent.", "estimated_reid_recall": 0.085, "k_anonymity_estimate": null, "qid_type_count_ris...
{"generation_seed": 11513162144825209156, "lattice_cell_id": "LxE:entity_type=DRIVER_LICENSE_NUMBER|language=ru", "lattice_seed": 91237, "lattice_tier": "critical", "lattice_version": "1.0.0", "license": "CC0-1.0", "source_type": "synthetic_lattice_enrichment", "v1_3_0_record_id": "d87e5377-b52d-4581-931c-a5f9265f83e3"...
null
fddad8c1-a024-5c5e-9a0e-7a6c18c36d5a
in_scope
in_scope
out_of_scope_of_dataset
out_of_scope_of_dataset
out_of_scope_of_dataset
high
2.2.0
Cyrl
Record for Иван Смирнов, Driver License Number: YO-123475, contact иван.смирнов32@outlook.com. [#9fc6f7f8abc7f144-236]
{"_caveat": "re_identification_resistance_score is a HEURISTIC PRIOR (closed-form over behavioral_signal_density), NOT a measured attack; exposure_index_prior is its honest alias. Exposure index is a DETERMINISTIC pre-screen PRIOR computed from behavioral-signal annotations. It is NOT measured RRS and MUST NOT be repor...
11
2.2.0
{"difficulty": "clean", "techniques": [], "type": null}
[{"anonymization_strategy": "pseudonymize", "category": "identity_demographics", "cluster_id": null, "context_dependency": "low", "end": 22, "entity_id": "e0", "entity_type": "PERSON_NAME", "information_anchor_score": 0.8, "mention_variant": "full_name", "nested_entities": null, "sensitivity_class": "direct_identifier"...
[]
{"behavioral_signal_density": 0.1, "interest_topics": {"indicators": [], "present": false, "uniqueness": "none"}, "location_signals": {"indicators": [], "present": false, "uniqueness": "none"}, "personal_anecdote": {"indicators": [], "present": false, "uniqueness": "none"}, "professional_domain": {"indicators": [], "pr...
short
{"_caveat": "No variant is demonstrated anonymised under GDPR Recital 26. anonymized_pseudonymized and anonymized_llm_sanitized remain PERSONAL DATA (Art.4(5)). residual_quasi_identifier flags a partial identifier (e.g. ***-**-NNNN SSN tail) retained in a variant — a quasi-identifier, not a safe endpoint. Synthetic-onl...
unstructured_text
easy
["context_preservation", "diverse_pii_types", "multilingual"]
null
general
{"coreference_chains": [], "num_distinct_persons": 1, "num_repeated_entities": 0, "tracking_difficulty": "none"}
it
Romance
diverse_pii_types
{"_caveat": "k_anonymity_estimate is a quasi-identifier-TYPE-count proxy (1->100, 2->20, 3->5, 4+->2), NOT an equivalence-class k computed over QI values; honest alias: qid_type_count_risk_label; null where quasi-identifiers are absent.", "estimated_reid_recall": 0.085, "k_anonymity_estimate": null, "qid_type_count_ris...
{"generation_seed": 1641656076268693148, "lattice_cell_id": "LxE:entity_type=BANK_ACCOUNT_NUMBER|language=it", "lattice_seed": 91237, "lattice_tier": "critical", "lattice_version": "1.0.0", "license": "CC0-1.0", "source_type": "synthetic_lattice_enrichment", "v1_3_0_record_id": "7d706cd6-78fb-4d6f-b429-b5fdd2f62ab4", "...
null
fe15315e-6ccd-5fb9-a434-b68306dc0d48
out_of_scope_of_dataset
in_scope
out_of_scope_of_dataset
out_of_scope_of_dataset
out_of_scope_of_dataset
high
2.2.0
Latn
Record for Elena Bruno, Bank Account Number: 9418863045, contact elena.bruno18@gmail.com. [#16c85580e55b669c-92]
{"_caveat": "re_identification_resistance_score is a HEURISTIC PRIOR (closed-form over behavioral_signal_density), NOT a measured attack; exposure_index_prior is its honest alias. Exposure index is a DETERMINISTIC pre-screen PRIOR computed from behavioral-signal annotations. It is NOT measured RRS and MUST NOT be repor...
11
2.2.0
{"difficulty": "clean", "techniques": [], "type": null}
[{"anonymization_strategy": "pseudonymize", "category": "identity_demographics", "cluster_id": null, "context_dependency": "low", "end": 21, "entity_id": "e0", "entity_type": "PERSON_NAME", "information_anchor_score": 0.79, "mention_variant": "full_name", "nested_entities": null, "sensitivity_class": "direct_identifier...
[]
{"behavioral_signal_density": 0.1, "interest_topics": {"indicators": [], "present": false, "uniqueness": "none"}, "location_signals": {"indicators": [], "present": false, "uniqueness": "none"}, "personal_anecdote": {"indicators": [], "present": false, "uniqueness": "none"}, "professional_domain": {"indicators": [], "pr...
short
{"_caveat": "No variant is demonstrated anonymised under GDPR Recital 26. anonymized_pseudonymized and anonymized_llm_sanitized remain PERSONAL DATA (Art.4(5)). residual_quasi_identifier flags a partial identifier (e.g. ***-**-NNNN SSN tail) retained in a variant — a quasi-identifier, not a safe endpoint. Synthetic-onl...
unstructured_text
hard
["context_preservation", "diverse_pii_types", "multilingual"]
null
general
{"coreference_chains": [], "num_distinct_persons": 1, "num_repeated_entities": 0, "tracking_difficulty": "none"}
hi
Indo-Aryan
diverse_pii_types
{"_caveat": "k_anonymity_estimate is a quasi-identifier-TYPE-count proxy (1->100, 2->20, 3->5, 4+->2), NOT an equivalence-class k computed over QI values; honest alias: qid_type_count_risk_label; null where quasi-identifiers are absent.", "estimated_reid_recall": 0.085, "k_anonymity_estimate": null, "qid_type_count_ris...
{"generation_seed": 17262775379323070824, "lattice_cell_id": "LxE:entity_type=PASSWORD|language=hi", "lattice_seed": 91237, "lattice_tier": "critical", "lattice_version": "1.0.0", "license": "CC0-1.0", "source_type": "synthetic_lattice_enrichment", "v1_3_0_record_id": "85400e85-adbb-4ca6-ab57-a68f3d79eac9", "v1_record_...
null
92273cef-5da6-5fcc-94c2-096bbd6dd468
out_of_scope_of_dataset
in_scope
out_of_scope_of_dataset
out_of_scope_of_dataset
out_of_scope_of_dataset
high
2.2.0
Deva
Record for Arun Reddy, Password: mr%XcyRxqWVU!rXi, contact arun.reddy42@icloud.com. [#ef91b2bed1e92968-1226]
{"_caveat": "re_identification_resistance_score is a HEURISTIC PRIOR (closed-form over behavioral_signal_density), NOT a measured attack; exposure_index_prior is its honest alias. Exposure index is a DETERMINISTIC pre-screen PRIOR computed from behavioral-signal annotations. It is NOT measured RRS and MUST NOT be repor...
9
2.2.0
{"difficulty": "clean", "techniques": [], "type": null}
[{"anonymization_strategy": "pseudonymize", "category": "identity_demographics", "cluster_id": null, "context_dependency": "low", "end": 22, "entity_id": "e0", "entity_type": "PERSON_NAME", "information_anchor_score": 0.87, "mention_variant": "full_name", "nested_entities": null, "sensitivity_class": "direct_identifier...
["HEALTH_INSURANCE_ID"]
{"behavioral_signal_density": 0.1, "interest_topics": {"indicators": [], "present": false, "uniqueness": "none"}, "location_signals": {"indicators": [], "present": false, "uniqueness": "none"}, "personal_anecdote": {"indicators": [], "present": false, "uniqueness": "none"}, "professional_domain": {"indicators": [], "pr...
short
{"_caveat": "No variant is demonstrated anonymised under GDPR Recital 26. anonymized_pseudonymized and anonymized_llm_sanitized remain PERSONAL DATA (Art.4(5)). residual_quasi_identifier flags a partial identifier (e.g. ***-**-NNNN SSN tail) retained in a variant — a quasi-identifier, not a safe endpoint. Synthetic-onl...
unstructured_text
easy
["context_preservation", "diverse_pii_types", "multilingual"]
null
general
{"coreference_chains": [], "num_distinct_persons": 1, "num_repeated_entities": 0, "tracking_difficulty": "none"}
it
Romance
diverse_pii_types
{"_caveat": "k_anonymity_estimate is a quasi-identifier-TYPE-count proxy (1->100, 2->20, 3->5, 4+->2), NOT an equivalence-class k computed over QI values; honest alias: qid_type_count_risk_label; null where quasi-identifiers are absent.", "estimated_reid_recall": 0.085, "k_anonymity_estimate": null, "qid_type_count_ris...
{"generation_seed": 2837584003964843217, "lattice_cell_id": "LxE:entity_type=HEALTH_INSURANCE_ID|language=it", "lattice_seed": 91237, "lattice_tier": "standard", "lattice_version": "1.0.0", "license": "CC0-1.0", "source_type": "synthetic_lattice_enrichment", "v1_3_0_record_id": "be6d9e71-e5c2-4fd3-8b92-ac8e1cae9565", "...
null
2e0c0133-1eb3-5818-9fcb-9fc3187313f7
out_of_scope_of_dataset
in_scope
in_scope
in_scope
out_of_scope_of_dataset
high
2.2.0
Latn
Record for Laura Ricci, Health Insurance Id: INS-563272960, contact laura.ricci86@mail.com. [#27611f9e10a7b4d1-749]
{"_caveat": "re_identification_resistance_score is a HEURISTIC PRIOR (closed-form over behavioral_signal_density), NOT a measured attack; exposure_index_prior is its honest alias. Exposure index is a DETERMINISTIC pre-screen PRIOR computed from behavioral-signal annotations. It is NOT measured RRS and MUST NOT be repor...
11
2.2.0
{"difficulty": "clean", "techniques": [], "type": null}
[{"anonymization_strategy": "pseudonymize", "category": "identity_demographics", "cluster_id": null, "context_dependency": "low", "end": 15, "entity_id": "e0", "entity_type": "PERSON_NAME", "information_anchor_score": 0.89, "mention_variant": "full_name", "nested_entities": null, "sensitivity_class": "direct_identifier...
[]
{"behavioral_signal_density": 0.1, "interest_topics": {"indicators": [], "present": false, "uniqueness": "none"}, "location_signals": {"indicators": [], "present": false, "uniqueness": "none"}, "personal_anecdote": {"indicators": [], "present": false, "uniqueness": "none"}, "professional_domain": {"indicators": [], "pr...
short
{"_caveat": "No variant is demonstrated anonymised under GDPR Recital 26. anonymized_pseudonymized and anonymized_llm_sanitized remain PERSONAL DATA (Art.4(5)). residual_quasi_identifier flags a partial identifier (e.g. ***-**-NNNN SSN tail) retained in a variant — a quasi-identifier, not a safe endpoint. Synthetic-onl...
unstructured_text
easy
["context_preservation", "diverse_pii_types", "multilingual"]
null
general
{"coreference_chains": [], "num_distinct_persons": 1, "num_repeated_entities": 0, "tracking_difficulty": "none"}
ko
Koreanic
diverse_pii_types
{"_caveat": "k_anonymity_estimate is a quasi-identifier-TYPE-count proxy (1->100, 2->20, 3->5, 4+->2), NOT an equivalence-class k computed over QI values; honest alias: qid_type_count_risk_label; null where quasi-identifiers are absent.", "estimated_reid_recall": 0.085, "k_anonymity_estimate": 100, "qid_type_count_risk...
{"generation_seed": 3157151977607983933, "lattice_cell_id": "LxE:entity_type=BANK_ROUTING_NUMBER|language=ko", "lattice_seed": 91237, "lattice_tier": "critical", "lattice_version": "1.0.0", "license": "CC0-1.0", "source_type": "synthetic_lattice_enrichment", "v1_3_0_record_id": "7780c25e-a29d-4fb0-8b5e-a11afd368b9a", "...
null
a240e872-d936-521d-9fcb-f298be482f6a
out_of_scope_of_dataset
in_scope
out_of_scope_of_dataset
out_of_scope_of_dataset
out_of_scope_of_dataset
high
2.2.0
Kore
Record for 예진 강, Bank Routing Number: 875523766, contact 예진.강72@mail.com. [#2bd074ff0f3ef33d-1334]
{"_caveat": "re_identification_resistance_score is a HEURISTIC PRIOR (closed-form over behavioral_signal_density), NOT a measured attack; exposure_index_prior is its honest alias. Exposure index is a DETERMINISTIC pre-screen PRIOR computed from behavioral-signal annotations. It is NOT measured RRS and MUST NOT be repor...
11
2.2.0
{"difficulty": "clean", "techniques": [], "type": null}
[{"anonymization_strategy": "pseudonymize", "category": "government_legal", "cluster_id": null, "context_dependency": "low", "end": 38, "entity_id": "e0", "entity_type": "COURT_CASE_NUMBER", "information_anchor_score": 0.87, "mention_variant": null, "nested_entities": null, "sensitivity_class": "direct_identifier", "st...
[]
{"behavioral_signal_density": 0.5233, "interest_topics": {"indicators": ["topic:fitness:1"], "present": true, "uniqueness": "moderate"}, "location_signals": {"indicators": ["local_reference:chicago:the_l"], "present": true, "uniqueness": "high"}, "personal_anecdote": {"indicators": [], "present": false, "uniqueness": "...
medium
{"_caveat": "No variant is demonstrated anonymised under GDPR Recital 26. anonymized_pseudonymized and anonymized_llm_sanitized remain PERSONAL DATA (Art.4(5)). residual_quasi_identifier flags a partial identifier (e.g. ***-**-NNNN SSN tail) retained in a variant — a quasi-identifier, not a safe endpoint. Synthetic-onl...
unstructured_text
hard
["context_preservation", "diverse_pii_types"]
witness_statement
legal
{"coreference_chains": [], "num_distinct_persons": 5, "num_repeated_entities": 0, "tracking_difficulty": "none"}
en
Germanic
diverse_pii_types
{"_caveat": "k_anonymity_estimate is a quasi-identifier-TYPE-count proxy (1->100, 2->20, 3->5, 4+->2), NOT an equivalence-class k computed over QI values; honest alias: qid_type_count_risk_label; null where quasi-identifiers are absent.", "estimated_reid_recall": 0.4448, "k_anonymity_estimate": 2, "qid_type_count_risk_...
{"generation_seed": 42, "license": "CC0-1.0", "source_type": "synthetic", "v1_3_0_record_id": "20480eb4-ae91-4478-bfa2-8e6c7f81fb9e", "v1_record_id": null}
null
38ec4556-fbdf-5fee-a81f-92d8324d8143
in_scope
in_scope
out_of_scope_of_dataset
out_of_scope_of_dataset
out_of_scope_of_dataset
high
2.2.0
Latn
WITNESS STATEMENT Case: 2022-CRIM-3632 Date: 2024-07-04T22:13:00Z I, Michelle Gonzalez, of 4999 Birch Way, Portland, PA 49478, make this statement voluntarily. My date of birth is 1966-09-24. I can be contacted at +1 (665) 509-7933. On the date in question, I was employed at Pied Piper as a VP of Engineering. I pers...
{"_caveat": "re_identification_resistance_score is a HEURISTIC PRIOR (closed-form over behavioral_signal_density), NOT a measured attack; exposure_index_prior is its honest alias. Exposure index is a DETERMINISTIC pre-screen PRIOR computed from behavioral-signal annotations. It is NOT measured RRS and MUST NOT be repor...
94
2.2.0
{"difficulty": "clean", "techniques": [], "type": null}
[{"anonymization_strategy": "pseudonymize", "category": "identity_demographics", "cluster_id": null, "context_dependency": "low", "end": 22, "entity_id": "e0", "entity_type": "PERSON_NAME", "information_anchor_score": 0.84, "mention_variant": "full_name", "nested_entities": null, "sensitivity_class": "direct_identifier...
[]
{"behavioral_signal_density": 0.1, "interest_topics": {"indicators": [], "present": false, "uniqueness": "none"}, "location_signals": {"indicators": [], "present": false, "uniqueness": "none"}, "personal_anecdote": {"indicators": [], "present": false, "uniqueness": "none"}, "professional_domain": {"indicators": [], "pr...
short
{"_caveat": "No variant is demonstrated anonymised under GDPR Recital 26. anonymized_pseudonymized and anonymized_llm_sanitized remain PERSONAL DATA (Art.4(5)). residual_quasi_identifier flags a partial identifier (e.g. ***-**-NNNN SSN tail) retained in a variant — a quasi-identifier, not a safe endpoint. Synthetic-onl...
unstructured_text
challenging
["context_preservation", "diverse_pii_types", "multilingual"]
null
general
{"coreference_chains": [], "num_distinct_persons": 1, "num_repeated_entities": 0, "tracking_difficulty": "none"}
pt
Romance
diverse_pii_types
{"_caveat": "k_anonymity_estimate is a quasi-identifier-TYPE-count proxy (1->100, 2->20, 3->5, 4+->2), NOT an equivalence-class k computed over QI values; honest alias: qid_type_count_risk_label; null where quasi-identifiers are absent.", "estimated_reid_recall": 0.085, "k_anonymity_estimate": null, "qid_type_count_ris...
{"generation_seed": 9137981067563857035, "lattice_cell_id": "LxE:entity_type=PASSPORT_NUMBER|language=pt", "lattice_seed": 91237, "lattice_tier": "critical", "lattice_version": "1.0.0", "license": "CC0-1.0", "source_type": "synthetic_lattice_enrichment", "v1_3_0_record_id": "079bacf9-4f8a-4178-b727-c6a24aaa8b00", "v1_r...
null
07935927-b1a4-56e6-992b-15496e0d9356
out_of_scope_of_dataset
in_scope
out_of_scope_of_dataset
out_of_scope_of_dataset
out_of_scope_of_dataset
high
2.2.0
Latn
Record for Pedro Alves, Passport Number: R63090986, contact pedro.alves91@zoho.com. [#7ed0a15da863748b-1400]
{"_caveat": "re_identification_resistance_score is a HEURISTIC PRIOR (closed-form over behavioral_signal_density), NOT a measured attack; exposure_index_prior is its honest alias. Exposure index is a DETERMINISTIC pre-screen PRIOR computed from behavioral-signal annotations. It is NOT measured RRS and MUST NOT be repor...
10
2.2.0
{"difficulty": "clean", "techniques": [], "type": null}
[{"anonymization_strategy": "pseudonymize", "category": "identity_demographics", "cluster_id": null, "context_dependency": "low", "end": 22, "entity_id": "e0", "entity_type": "PERSON_NAME", "information_anchor_score": 0.88, "mention_variant": "full_name", "nested_entities": null, "sensitivity_class": "direct_identifier...
[]
{"behavioral_signal_density": 0.1, "interest_topics": {"indicators": [], "present": false, "uniqueness": "none"}, "location_signals": {"indicators": [], "present": false, "uniqueness": "none"}, "personal_anecdote": {"indicators": [], "present": false, "uniqueness": "none"}, "professional_domain": {"indicators": [], "pr...
short
{"_caveat": "No variant is demonstrated anonymised under GDPR Recital 26. anonymized_pseudonymized and anonymized_llm_sanitized remain PERSONAL DATA (Art.4(5)). residual_quasi_identifier flags a partial identifier (e.g. ***-**-NNNN SSN tail) retained in a variant — a quasi-identifier, not a safe endpoint. Synthetic-onl...
unstructured_text
challenging
["context_preservation", "diverse_pii_types", "multilingual"]
null
general
{"coreference_chains": [], "num_distinct_persons": 1, "num_repeated_entities": 0, "tracking_difficulty": "none"}
hi
Indo-Aryan
diverse_pii_types
{"_caveat": "k_anonymity_estimate is a quasi-identifier-TYPE-count proxy (1->100, 2->20, 3->5, 4+->2), NOT an equivalence-class k computed over QI values; honest alias: qid_type_count_risk_label; null where quasi-identifiers are absent.", "estimated_reid_recall": 0.085, "k_anonymity_estimate": null, "qid_type_count_ris...
{"generation_seed": 10834775599406328466, "lattice_cell_id": "LxE:entity_type=API_KEY|language=hi", "lattice_seed": 91237, "lattice_tier": "critical", "lattice_version": "1.0.0", "license": "CC0-1.0", "source_type": "synthetic_lattice_enrichment", "v1_3_0_record_id": "80f427b8-31fc-4de8-a8fd-726eddd2a273", "v1_record_i...
null
a6bc1ed5-a9f0-5324-acd1-289f45a5e3ac
out_of_scope_of_dataset
in_scope
out_of_scope_of_dataset
out_of_scope_of_dataset
out_of_scope_of_dataset
high
2.2.0
Deva
Record for Pooja Verma, API key: sk-c1Q6RIHRTk6xpjDGH6000qUhN3fpUBSK, contact pooja.verma42@outlook.com. [#965cdb00a7af2a92-342]
{"_caveat": "re_identification_resistance_score is a HEURISTIC PRIOR (closed-form over behavioral_signal_density), NOT a measured attack; exposure_index_prior is its honest alias. Exposure index is a DETERMINISTIC pre-screen PRIOR computed from behavioral-signal annotations. It is NOT measured RRS and MUST NOT be repor...
10
2.2.0
{"difficulty": "clean", "techniques": [], "type": null}
[{"anonymization_strategy": "pseudonymize", "category": "identity_demographics", "cluster_id": null, "context_dependency": "low", "end": 21, "entity_id": "e0", "entity_type": "PERSON_NAME", "information_anchor_score": 0.76, "mention_variant": "full_name", "nested_entities": null, "sensitivity_class": "direct_identifier...
[]
{"behavioral_signal_density": 0.2667, "interest_topics": {"indicators": ["topic:fitness:1"], "present": true, "uniqueness": "moderate"}, "location_signals": {"indicators": [], "present": false, "uniqueness": "none"}, "personal_anecdote": {"indicators": [], "present": false, "uniqueness": "none"}, "professional_domain":...
short
{"_caveat": "No variant is demonstrated anonymised under GDPR Recital 26. anonymized_pseudonymized and anonymized_llm_sanitized remain PERSONAL DATA (Art.4(5)). residual_quasi_identifier flags a partial identifier (e.g. ***-**-NNNN SSN tail) retained in a variant — a quasi-identifier, not a safe endpoint. Synthetic-onl...
unstructured_text
hard
["context_preservation", "diverse_pii_types", "multilingual"]
null
general
{"coreference_chains": [], "num_distinct_persons": 1, "num_repeated_entities": 0, "tracking_difficulty": "none"}
es
Romance
diverse_pii_types
{"_caveat": "k_anonymity_estimate is a quasi-identifier-TYPE-count proxy (1->100, 2->20, 3->5, 4+->2), NOT an equivalence-class k computed over QI values; honest alias: qid_type_count_risk_label; null where quasi-identifiers are absent.", "estimated_reid_recall": 0.2267, "k_anonymity_estimate": null, "qid_type_count_ri...
{"generation_seed": 1481173571265173557, "lattice_cell_id": "LxE:entity_type=PASSPORT_NUMBER|language=es", "lattice_seed": 91237, "lattice_tier": "critical", "lattice_version": "1.0.0", "license": "CC0-1.0", "source_type": "synthetic_lattice_enrichment", "v1_3_0_record_id": "7768cce6-4fa7-47c0-a479-509943caa854", "v1_r...
null
3a85ee3f-2bae-5321-885b-9945d0c64c90
out_of_scope_of_dataset
in_scope
out_of_scope_of_dataset
out_of_scope_of_dataset
out_of_scope_of_dataset
high
2.2.0
Latn
Record for José López, Passport Number: F82046266, contact josé.lópez18@protonmail.com. [#148e2f83f5f51835-460]
{"_caveat": "re_identification_resistance_score is a HEURISTIC PRIOR (closed-form over behavioral_signal_density), NOT a measured attack; exposure_index_prior is its honest alias. Exposure index is a DETERMINISTIC pre-screen PRIOR computed from behavioral-signal annotations. It is NOT measured RRS and MUST NOT be repor...
10
2.2.0
{"difficulty": "clean", "techniques": [], "type": null}
[{"anonymization_strategy": "pseudonymize", "category": "identity_demographics", "cluster_id": null, "context_dependency": "low", "end": 19, "entity_id": "e0", "entity_type": "PERSON_NAME", "information_anchor_score": 0.9, "mention_variant": "full_name", "sensitivity_class": "direct_identifier", "start": 11, "text": "א...
[]
{"behavioral_signal_density": 0.1, "interest_topics": {"indicators": [], "present": false, "uniqueness": "none"}, "location_signals": {"indicators": [], "present": false, "uniqueness": "none"}, "personal_anecdote": {"indicators": [], "present": false, "uniqueness": "none"}, "professional_domain": {"indicators": [], "pr...
short
{"_caveat": "No variant is demonstrated anonymised under GDPR Recital 26. anonymized_pseudonymized and anonymized_llm_sanitized remain PERSONAL DATA (Art.4(5)). residual_quasi_identifier flags a partial identifier (e.g. ***-**-NNNN SSN tail) retained in a variant — a quasi-identifier, not a safe endpoint. Synthetic-onl...
unstructured_text
easy
["context_preservation", "diverse_pii_types", "multilingual"]
null
general
{"coreference_chains": [], "num_distinct_persons": 1, "num_repeated_entities": 0, "tracking_difficulty": "none"}
he
Semitic
diverse_pii_types
{"_caveat": "k_anonymity_estimate is a quasi-identifier-TYPE-count proxy (1->100, 2->20, 3->5, 4+->2), NOT an equivalence-class k computed over QI values; honest alias: qid_type_count_risk_label; null where quasi-identifiers are absent.", "estimated_reid_recall": 0.085, "k_anonymity_estimate": null, "qid_type_count_ris...
{"generation_seed": 10925129879724747877, "lattice_cell_id": "LxE:entity_type=BAR_NUMBER|language=he", "lattice_seed": 91237, "lattice_tier": "standard", "lattice_version": "1.0.0", "license": "CC0-1.0", "source_type": "synthetic_lattice_enrichment", "v1_3_0_record_id": "d7e2de2c-20fd-4198-9c13-7c00f74d4980", "v1_recor...
null
9a506a4a-1a06-54e7-a4d5-af50c8eb6f0e
out_of_scope_of_dataset
in_scope
out_of_scope_of_dataset
out_of_scope_of_dataset
out_of_scope_of_dataset
medium
2.2.0
Hebr
Record for איתן לוי, Bar Number: BAR-135280, contact איתן.לוי63@icloud.com. [#979ddbbde5ffec65-301]
{"_caveat": "re_identification_resistance_score is a HEURISTIC PRIOR (closed-form over behavioral_signal_density), NOT a measured attack; exposure_index_prior is its honest alias. Exposure index is a DETERMINISTIC pre-screen PRIOR computed from behavioral-signal annotations. It is NOT measured RRS and MUST NOT be repor...
10
2.2.0
{"difficulty": "clean", "techniques": [], "type": null}
[{"anonymization_strategy": "pseudonymize", "category": "identity_demographics", "cluster_id": null, "context_dependency": "low", "end": 26, "entity_id": "e0", "entity_type": "PERSON_NAME", "information_anchor_score": 0.89, "mention_variant": "full_name", "nested_entities": null, "sensitivity_class": "direct_identifier...
[]
{"behavioral_signal_density": 0.2433, "interest_topics": {"indicators": [], "present": false, "uniqueness": "none"}, "location_signals": {"indicators": [], "present": false, "uniqueness": "none"}, "personal_anecdote": {"indicators": [], "present": false, "uniqueness": "none"}, "professional_domain": {"indicators": ["in...
short
{"_caveat": "No variant is demonstrated anonymised under GDPR Recital 26. anonymized_pseudonymized and anonymized_llm_sanitized remain PERSONAL DATA (Art.4(5)). residual_quasi_identifier flags a partial identifier (e.g. ***-**-NNNN SSN tail) retained in a variant — a quasi-identifier, not a safe endpoint. Synthetic-onl...
unstructured_text
hard
["context_preservation", "diverse_pii_types", "multilingual"]
null
general
{"coreference_chains": [], "num_distinct_persons": 1, "num_repeated_entities": 0, "tracking_difficulty": "none"}
pt
Romance
diverse_pii_types
{"_caveat": "k_anonymity_estimate is a quasi-identifier-TYPE-count proxy (1->100, 2->20, 3->5, 4+->2), NOT an equivalence-class k computed over QI values; honest alias: qid_type_count_risk_label; null where quasi-identifiers are absent.", "estimated_reid_recall": 0.2068, "k_anonymity_estimate": null, "qid_type_count_ri...
{"generation_seed": 5296417550988081404, "lattice_cell_id": "LxE:entity_type=NPI_NUMBER|language=pt", "lattice_seed": 91237, "lattice_tier": "standard", "lattice_version": "1.0.0", "license": "CC0-1.0", "source_type": "synthetic_lattice_enrichment", "v1_3_0_record_id": "815294ab-3e97-47c5-bce3-60e196fc31ea", "v1_record...
null
0ecb3986-93d6-5ea8-995a-6d2e078c30ee
out_of_scope_of_dataset
in_scope
out_of_scope_of_dataset
out_of_scope_of_dataset
out_of_scope_of_dataset
high
2.2.0
Latn
Record for Beatriz Pereira, NPI: 1110981945, contact beatriz.pereira62@gmail.com. [#4980a7b374c9f8fc-719]
{"_caveat": "re_identification_resistance_score is a HEURISTIC PRIOR (closed-form over behavioral_signal_density), NOT a measured attack; exposure_index_prior is its honest alias. Exposure index is a DETERMINISTIC pre-screen PRIOR computed from behavioral-signal annotations. It is NOT measured RRS and MUST NOT be repor...
9
2.2.0
{"difficulty": "clean", "techniques": [], "type": null}
[{"anonymization_strategy": "pseudonymize", "category": "identity_demographics", "cluster_id": null, "context_dependency": "low", "end": 25, "entity_id": "e0", "entity_type": "PERSON_NAME", "information_anchor_score": 1.0, "mention_variant": null, "nested_entities": null, "sensitivity_class": "direct_identifier", "star...
[]
{"behavioral_signal_density": 0.2667, "interest_topics": {"indicators": ["topic:fitness:1"], "present": true, "uniqueness": "moderate"}, "location_signals": {"indicators": [], "present": false, "uniqueness": "none"}, "personal_anecdote": {"indicators": [], "present": false, "uniqueness": "none"}, "professional_domain":...
short
{"_caveat": "No variant is demonstrated anonymised under GDPR Recital 26. anonymized_pseudonymized and anonymized_llm_sanitized remain PERSONAL DATA (Art.4(5)). residual_quasi_identifier flags a partial identifier (e.g. ***-**-NNNN SSN tail) retained in a variant — a quasi-identifier, not a safe endpoint. Synthetic-onl...
semi_structured
easy
["context_preservation", "diverse_pii_types"]
null
general
{"coreference_chains": [], "num_distinct_persons": 0, "num_repeated_entities": 0, "tracking_difficulty": "none"}
en
Germanic
diverse_pii_types
{"_caveat": "k_anonymity_estimate is a quasi-identifier-TYPE-count proxy (1->100, 2->20, 3->5, 4+->2), NOT an equivalence-class k computed over QI values; honest alias: qid_type_count_risk_label; null where quasi-identifiers are absent.", "estimated_reid_recall": 0.2267, "k_anonymity_estimate": null, "qid_type_count_ri...
{"generation_seed": 42, "license": "CC0-1.0", "source_type": "synthetic", "v1_3_0_record_id": "1553785f-8024-5650-9af8-887bb5127712", "v1_record_id": "EVAL-012254"}
{"query": "What is the email address on file?", "relevant_entity_ids": ["e0", "e1"]}
f810946a-9376-5567-b50a-31f085cbd568
in_scope
in_scope
out_of_scope_of_dataset
out_of_scope_of_dataset
out_of_scope_of_dataset
high
2.2.0
Latn
{{"name": "Mary Rodriguez", "email": "mary.rodriguez36@protonmail.com", "phone": "+1 (816) 992-2656", "ssn": "851-27-5539", "ip": "95.83.178.64"}}
{"_caveat": "re_identification_resistance_score is a HEURISTIC PRIOR (closed-form over behavioral_signal_density), NOT a measured attack; exposure_index_prior is its honest alias. Exposure index is a DETERMINISTIC pre-screen PRIOR computed from behavioral-signal annotations. It is NOT measured RRS and MUST NOT be repor...
13
2.2.0
{"difficulty": "clean", "techniques": [], "type": null}
[{"anonymization_strategy": "pseudonymize", "category": "identity_demographics", "cluster_id": null, "context_dependency": "low", "end": 24, "entity_id": "e0", "entity_type": "PERSON_NAME", "information_anchor_score": 0.79, "mention_variant": "full_name", "nested_entities": null, "sensitivity_class": "direct_identifier...
[]
{"behavioral_signal_density": 0.1, "interest_topics": {"indicators": [], "present": false, "uniqueness": "none"}, "location_signals": {"indicators": [], "present": false, "uniqueness": "none"}, "personal_anecdote": {"indicators": [], "present": false, "uniqueness": "none"}, "professional_domain": {"indicators": [], "pr...
short
{"_caveat": "No variant is demonstrated anonymised under GDPR Recital 26. anonymized_pseudonymized and anonymized_llm_sanitized remain PERSONAL DATA (Art.4(5)). residual_quasi_identifier flags a partial identifier (e.g. ***-**-NNNN SSN tail) retained in a variant — a quasi-identifier, not a safe endpoint. Synthetic-onl...
unstructured_text
easy
["context_preservation", "diverse_pii_types", "multilingual"]
null
general
{"coreference_chains": [], "num_distinct_persons": 1, "num_repeated_entities": 0, "tracking_difficulty": "none"}
pt
Romance
diverse_pii_types
{"_caveat": "k_anonymity_estimate is a quasi-identifier-TYPE-count proxy (1->100, 2->20, 3->5, 4+->2), NOT an equivalence-class k computed over QI values; honest alias: qid_type_count_risk_label; null where quasi-identifiers are absent.", "estimated_reid_recall": 0.085, "k_anonymity_estimate": null, "qid_type_count_ris...
{"generation_seed": 13830939834591839103, "lattice_cell_id": "LxE:entity_type=TAX_ID|language=pt", "lattice_seed": 91237, "lattice_tier": "critical", "lattice_version": "1.0.0", "license": "CC0-1.0", "source_type": "synthetic_lattice_enrichment", "v1_3_0_record_id": "3e63ee6b-9dc4-4eb6-a02c-626f66fbf802", "v1_record_id...
null
17b79e08-d1ae-5797-9be7-b579005fcdac
out_of_scope_of_dataset
in_scope
out_of_scope_of_dataset
out_of_scope_of_dataset
out_of_scope_of_dataset
high
2.2.0
Latn
Record for Gabriel Souza, Tax Id: 67-5345375, contact gabriel.souza22@yahoo.com. [#bff15e8006a5837f-801]
{"_caveat": "re_identification_resistance_score is a HEURISTIC PRIOR (closed-form over behavioral_signal_density), NOT a measured attack; exposure_index_prior is its honest alias. Exposure index is a DETERMINISTIC pre-screen PRIOR computed from behavioral-signal annotations. It is NOT measured RRS and MUST NOT be repor...
10
2.2.0
{"difficulty": "clean", "techniques": [], "type": null}
[{"anonymization_strategy": "pseudonymize", "category": "identity_demographics", "cluster_id": null, "context_dependency": "low", "end": 39, "entity_id": "e0", "entity_type": "PERSON_NAME", "information_anchor_score": 0.86, "mention_variant": "full_name", "nested_entities": null, "sensitivity_class": "direct_identifier...
["HEALTH_CONDITION", "MEDICAL_RECORD_NUMBER", "MEDICATION_NAME"]
{"behavioral_signal_density": 0.4567, "interest_topics": {"indicators": [], "present": false, "uniqueness": "none"}, "location_signals": {"indicators": [], "present": false, "uniqueness": "none"}, "personal_anecdote": {"indicators": [], "present": false, "uniqueness": "none"}, "professional_domain": {"indicators": ["in...
short
{"_caveat": "No variant is demonstrated anonymised under GDPR Recital 26. anonymized_pseudonymized and anonymized_llm_sanitized remain PERSONAL DATA (Art.4(5)). residual_quasi_identifier flags a partial identifier (e.g. ***-**-NNNN SSN tail) retained in a variant — a quasi-identifier, not a safe endpoint. Synthetic-onl...
unstructured_text
moderate
["context_preservation", "diverse_pii_types", "multilingual"]
discharge_summary
clinical
{"coreference_chains": [], "num_distinct_persons": 2, "num_repeated_entities": 0, "tracking_difficulty": "none"}
fr
Romance
diverse_pii_types
{"_caveat": "k_anonymity_estimate is a quasi-identifier-TYPE-count proxy (1->100, 2->20, 3->5, 4+->2), NOT an equivalence-class k computed over QI values; honest alias: qid_type_count_risk_label; null where quasi-identifiers are absent.", "estimated_reid_recall": 0.3882, "k_anonymity_estimate": 20, "qid_type_count_risk...
{"generation_seed": 42, "license": "CC0-1.0", "source_type": "synthetic", "v1_3_0_record_id": "7e337b63-f7bf-47bd-a2fe-6d62e9261a38", "v1_record_id": null}
null
cf3613ae-1774-57bf-a3bb-2cb024a0bced
out_of_scope_of_dataset
in_scope
in_scope
in_scope
out_of_scope_of_dataset
high
2.2.0
Latn
Referral Letter From: Dr. Marie Bernard, Nordic Analytics To: Specialist Re: Marie Richard (DOB: 1981-10-25, MRN: MRN-2337518) Diagnosis: Hypothyroidism Current Medications: Losartan 50mg Contact: +1 (740) 241-6204, marie.richard28@mail.com
{"_caveat": "re_identification_resistance_score is a HEURISTIC PRIOR (closed-form over behavioral_signal_density), NOT a measured attack; exposure_index_prior is its honest alias. Exposure index is a DETERMINISTIC pre-screen PRIOR computed from behavioral-signal annotations. It is NOT measured RRS and MUST NOT be repor...
28
2.2.0
{"difficulty": "clean", "techniques": [], "type": null}
[{"anonymization_strategy": "pseudonymize", "category": "identity_demographics", "cluster_id": null, "context_dependency": "low", "end": 29, "entity_id": "e0", "entity_type": "PERSON_NAME", "information_anchor_score": 0.79, "mention_variant": "full_name", "nested_entities": null, "sensitivity_class": "direct_identifier...
[]
{"behavioral_signal_density": 0.1, "interest_topics": {"indicators": [], "present": false, "uniqueness": "none"}, "location_signals": {"indicators": [], "present": false, "uniqueness": "none"}, "personal_anecdote": {"indicators": [], "present": false, "uniqueness": "none"}, "professional_domain": {"indicators": [], "pr...
short
{"_caveat": "No variant is demonstrated anonymised under GDPR Recital 26. anonymized_pseudonymized and anonymized_llm_sanitized remain PERSONAL DATA (Art.4(5)). residual_quasi_identifier flags a partial identifier (e.g. ***-**-NNNN SSN tail) retained in a variant — a quasi-identifier, not a safe endpoint. Synthetic-onl...
unstructured_text
moderate
["context_preservation", "diverse_pii_types", "multilingual"]
null
general
{"coreference_chains": [], "num_distinct_persons": 1, "num_repeated_entities": 0, "tracking_difficulty": "none"}
it
Romance
diverse_pii_types
{"_caveat": "k_anonymity_estimate is a quasi-identifier-TYPE-count proxy (1->100, 2->20, 3->5, 4+->2), NOT an equivalence-class k computed over QI values; honest alias: qid_type_count_risk_label; null where quasi-identifiers are absent.", "estimated_reid_recall": 0.085, "k_anonymity_estimate": null, "qid_type_count_ris...
{"generation_seed": 6971621996025052282, "lattice_cell_id": "LxE:entity_type=NATIONAL_ID_NUMBER|language=it", "lattice_seed": 91237, "lattice_tier": "critical", "lattice_version": "1.0.0", "license": "CC0-1.0", "source_type": "synthetic_lattice_enrichment", "v1_3_0_record_id": "5a009595-8b2e-42d3-94f4-42e43700526e", "v...
null
760a332e-a5ac-5098-a7a0-dfe3be3b496e
out_of_scope_of_dataset
in_scope
out_of_scope_of_dataset
out_of_scope_of_dataset
out_of_scope_of_dataset
high
2.2.0
Latn
Record for Alessandro Colombo, National Id Number: NID-512664727, contact alessandro.colombo6@icloud.com. [#60c02d44a55f7c7a-575]
{"_caveat": "re_identification_resistance_score is a HEURISTIC PRIOR (closed-form over behavioral_signal_density), NOT a measured attack; exposure_index_prior is its honest alias. Exposure index is a DETERMINISTIC pre-screen PRIOR computed from behavioral-signal annotations. It is NOT measured RRS and MUST NOT be repor...
11
2.2.0
{"difficulty": "clean", "techniques": [], "type": null}
[{"anonymization_strategy": "pseudonymize", "category": "identity_demographics", "cluster_id": null, "context_dependency": "low", "end": 25, "entity_id": "e0", "entity_type": "PERSON_NAME", "information_anchor_score": 0.89, "mention_variant": "full_name", "sensitivity_class": "direct_identifier", "start": 11, "text": "...
["HEALTH_INSURANCE_ID"]
{"behavioral_signal_density": 0.1, "interest_topics": {"indicators": [], "present": false, "uniqueness": "none"}, "location_signals": {"indicators": [], "present": false, "uniqueness": "none"}, "personal_anecdote": {"indicators": [], "present": false, "uniqueness": "none"}, "professional_domain": {"indicators": [], "pr...
short
{"_caveat": "No variant is demonstrated anonymised under GDPR Recital 26. anonymized_pseudonymized and anonymized_llm_sanitized remain PERSONAL DATA (Art.4(5)). residual_quasi_identifier flags a partial identifier (e.g. ***-**-NNNN SSN tail) retained in a variant — a quasi-identifier, not a safe endpoint. Synthetic-onl...
unstructured_text
easy
["context_preservation", "diverse_pii_types", "multilingual"]
null
general
{"coreference_chains": [], "num_distinct_persons": 1, "num_repeated_entities": 0, "tracking_difficulty": "none"}
el
Hellenic
diverse_pii_types
{"_caveat": "k_anonymity_estimate is a quasi-identifier-TYPE-count proxy (1->100, 2->20, 3->5, 4+->2), NOT an equivalence-class k computed over QI values; honest alias: qid_type_count_risk_label; null where quasi-identifiers are absent.", "estimated_reid_recall": 0.085, "k_anonymity_estimate": null, "qid_type_count_ris...
{"generation_seed": 27458098849350495, "lattice_cell_id": "LxE:entity_type=HEALTH_INSURANCE_ID|language=el", "lattice_seed": 91237, "lattice_tier": "standard", "lattice_version": "1.0.0", "license": "CC0-1.0", "source_type": "synthetic_lattice_enrichment", "v1_3_0_record_id": "c0870e9b-c44e-4823-bc5c-1ea9b560d1f6", "v1...
null
ecf4465e-50b3-5feb-ac9d-d1f10eab9738
out_of_scope_of_dataset
in_scope
in_scope
in_scope
out_of_scope_of_dataset
medium
2.2.0
Grek
Record for Ιωάννης Μακρής, Health Insurance Id: INS-509868898, contact ιωάννης.μακρής32@hotmail.com. [#618cfed41f835f-642]
{"_caveat": "re_identification_resistance_score is a HEURISTIC PRIOR (closed-form over behavioral_signal_density), NOT a measured attack; exposure_index_prior is its honest alias. Exposure index is a DETERMINISTIC pre-screen PRIOR computed from behavioral-signal annotations. It is NOT measured RRS and MUST NOT be repor...
11
2.2.0
{"difficulty": "clean", "techniques": [], "type": null}
[{"anonymization_strategy": "pseudonymize", "category": "identity_demographics", "cluster_id": null, "context_dependency": "low", "end": 22, "entity_id": "e0", "entity_type": "PERSON_NAME", "information_anchor_score": 0.76, "mention_variant": "full_name", "sensitivity_class": "direct_identifier", "start": 11, "text": "...
[]
{"behavioral_signal_density": 0.1, "interest_topics": {"indicators": [], "present": false, "uniqueness": "none"}, "location_signals": {"indicators": [], "present": false, "uniqueness": "none"}, "personal_anecdote": {"indicators": [], "present": false, "uniqueness": "none"}, "professional_domain": {"indicators": [], "pr...
short
{"_caveat": "No variant is demonstrated anonymised under GDPR Recital 26. anonymized_pseudonymized and anonymized_llm_sanitized remain PERSONAL DATA (Art.4(5)). residual_quasi_identifier flags a partial identifier (e.g. ***-**-NNNN SSN tail) retained in a variant — a quasi-identifier, not a safe endpoint. Synthetic-onl...
unstructured_text
moderate
["context_preservation", "diverse_pii_types", "multilingual"]
null
general
{"coreference_chains": [], "num_distinct_persons": 1, "num_repeated_entities": 0, "tracking_difficulty": "none"}
ru
Slavic
diverse_pii_types
{"_caveat": "k_anonymity_estimate is a quasi-identifier-TYPE-count proxy (1->100, 2->20, 3->5, 4+->2), NOT an equivalence-class k computed over QI values; honest alias: qid_type_count_risk_label; null where quasi-identifiers are absent.", "estimated_reid_recall": 0.085, "k_anonymity_estimate": 100, "qid_type_count_risk...
{"generation_seed": 16478133470816008125, "lattice_cell_id": "LxE:entity_type=SALARY|language=ru", "lattice_seed": 91237, "lattice_tier": "standard", "lattice_version": "1.0.0", "license": "CC0-1.0", "source_type": "synthetic_lattice_enrichment", "v1_3_0_record_id": "75c10a03-7564-417f-8d7a-a0b1c57b5520", "v1_record_id...
null
db269119-dc1d-59d9-b169-e3ab5707425f
out_of_scope_of_dataset
in_scope
out_of_scope_of_dataset
out_of_scope_of_dataset
out_of_scope_of_dataset
high
2.2.0
Cyrl
Record for Анна Петров, Salary: $72,000, contact анна.петров69@hotmail.com. [#e4ae1716397677bd-144]
{"_caveat": "re_identification_resistance_score is a HEURISTIC PRIOR (closed-form over behavioral_signal_density), NOT a measured attack; exposure_index_prior is its honest alias. Exposure index is a DETERMINISTIC pre-screen PRIOR computed from behavioral-signal annotations. It is NOT measured RRS and MUST NOT be repor...
9
2.2.0
{"difficulty": "clean", "techniques": [], "type": null}
[{"anonymization_strategy": "pseudonymize", "category": "identity_demographics", "cluster_id": null, "context_dependency": "low", "end": 16, "entity_id": "e0", "entity_type": "PERSON_NAME", "information_anchor_score": 0.86, "mention_variant": "full_name", "nested_entities": null, "sensitivity_class": "direct_identifier...
["HEALTH_INSURANCE_ID"]
{"behavioral_signal_density": 0.1, "interest_topics": {"indicators": [], "present": false, "uniqueness": "none"}, "location_signals": {"indicators": [], "present": false, "uniqueness": "none"}, "personal_anecdote": {"indicators": [], "present": false, "uniqueness": "none"}, "professional_domain": {"indicators": [], "pr...
short
{"_caveat": "No variant is demonstrated anonymised under GDPR Recital 26. anonymized_pseudonymized and anonymized_llm_sanitized remain PERSONAL DATA (Art.4(5)). residual_quasi_identifier flags a partial identifier (e.g. ***-**-NNNN SSN tail) retained in a variant — a quasi-identifier, not a safe endpoint. Synthetic-onl...
unstructured_text
challenging
["context_preservation", "diverse_pii_types", "multilingual"]
null
general
{"coreference_chains": [], "num_distinct_persons": 1, "num_repeated_entities": 0, "tracking_difficulty": "none"}
ja
Japonic
diverse_pii_types
{"_caveat": "k_anonymity_estimate is a quasi-identifier-TYPE-count proxy (1->100, 2->20, 3->5, 4+->2), NOT an equivalence-class k computed over QI values; honest alias: qid_type_count_risk_label; null where quasi-identifiers are absent.", "estimated_reid_recall": 0.085, "k_anonymity_estimate": null, "qid_type_count_ris...
{"generation_seed": 8701800421124313966, "lattice_cell_id": "LxE:entity_type=HEALTH_INSURANCE_ID|language=ja", "lattice_seed": 91237, "lattice_tier": "standard", "lattice_version": "1.0.0", "license": "CC0-1.0", "source_type": "synthetic_lattice_enrichment", "v1_3_0_record_id": "336e5b7e-bcf0-4398-b502-0b6d49194cb4", "...
null
d8540689-ebc1-5bef-8021-f617ed051c56
out_of_scope_of_dataset
in_scope
in_scope
in_scope
out_of_scope_of_dataset
high
2.2.0
Jpan
Record for 愛子 鈴木, Health Insurance Id: INS-459845046, contact 愛子.鈴木26@fastmail.com. [#78c30160ff75936e-150]
{"_caveat": "re_identification_resistance_score is a HEURISTIC PRIOR (closed-form over behavioral_signal_density), NOT a measured attack; exposure_index_prior is its honest alias. Exposure index is a DETERMINISTIC pre-screen PRIOR computed from behavioral-signal annotations. It is NOT measured RRS and MUST NOT be repor...
11
2.2.0
End of preview. Expand in Data Studio

PII-Anon

A CC0 multilingual PII benchmark corpus of 782,677 records carrying 3,107,240 entity annotations across 66 entity types and 60 languages, spanning 7 evaluation dimensions. Each record exposes the five legally-distinct regulatory regime signals (gov-02 / FR-022) as separate reg_* columns — no merged compliance verdict.

Train vs. evaluation substrate

The 159,891 tier3_evaluation records are the EVALUATION substrate of the 782,677-record corpus (behavioral-signal / RRS scoring runs on this substrate), NOT the whole corpus.

Synthetic-enrichment disclosure

79.2% of records carry provenance.source_type='synthetic_lattice_enrichment' (the S-PWR power fill); synthetic power is not external validity.

Statistical power

PII-Anon v2 is powered for all single-factor marginal recall claims (95% Wilson CIs; credential/financial-critical types to ±0.5pp at recall 0.99, standard to ±1pp at 0.98) and for three pre-registered 2-way interactions (language×entity-type on a committed rectangle, domain×track, adversarial-type×entity-type). The corpus carries a committed evaluation lattice powering 17 languages across 11 writing systems (Latin, Han, Japanese, Hangul, Arabic, Devanagari, Cyrillic, Thai, Greek, Bengali, Hebrew) to statistically-calibrated positive-count targets (critical n≥1522, standard n≥753). It is not powered for the full multilingual×entity-type grid or any ≥3-way interaction; those are reported as exploratory. Synthetic-distribution power is not external validity — see the real-data correlation slice.

Baseline Detector Performance

How widely-used PII detectors score on this corpus (test split, language en; 31,048 records), ranked by F2 (recall-weighted — a missed PII is the costly error). Full per-type / per-domain / per-language tables, Wilson CIs, and provenance: baseline_results.json and BASELINES.md.

Power on a committed cell is statistical precision on the SYNTHETIC distribution, NOT external validity; not citable as a standalone recall claim absent the real-data correlation slice (FR-027). Synthetic-only (AX-001).

Rank Detector Precision Recall F2 Recall 95% CI Coverage
1 aws 0.769 0.728 0.736 [0.726, 0.730] 24/66
2 gliner 0.813 0.716 0.734 [0.714, 0.718] 23/66
3 gcp 0.722 0.700 0.704 [0.698, 0.702] 18/66
4 azure 0.730 0.688 0.696 [0.686, 0.690] 17/66
5 presidio 0.419 0.562 0.526 [0.560, 0.564] 20/66
6 regex 0.857 0.349 0.396 [0.347, 0.351] 9/66
7 piiranha 0.441 0.327 0.345 [0.325, 0.329] 16/66
8 stanza 0.583 0.308 0.340 [0.306, 0.310] 3/66
9 flair 0.565 0.295 0.326 [0.293, 0.297] 3/66
10 spacy 0.464 0.294 0.317 [0.292, 0.296] 3/66
11 scrubadub 0.818 0.169 0.201 [0.167, 0.170] 12/66

License

The data is released under CC0-1.0 (public-domain dedication); the accompanying code (loaders, scorers, exporters) is licensed Apache-2.0. The two licenses are distinct — using the data does not subject you to the code license, and vice versa.

Downloads last month
166