input_text stringlengths 25 167 ⌀ | replaced_output stringlengths 25 167 ⌀ | replacement_lookup stringlengths 2 537 | group float64 4 4 ⌀ | conversation_hash stringlengths 36 36 ⌀ | predicted_category stringclasses 1 value | user_query stringlengths 13 1.12k ⌀ | target_response stringclasses 1 value | pii_units stringlengths 0 108 ⌀ | redacted_query stringlengths 13 1.12k ⌀ | original_id float64 108k 515k ⌀ | language stringclasses 1 value | original_dataset stringclasses 1 value | soft_replaced_query stringlengths 13 1.12k ⌀ | llm_model stringclasses 1 value | llm_score float64 9 10 ⌀ | llm_analysis stringlengths 203 1.05k ⌀ | scoring_success bool 1 class |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Turtles bask on sunlit rocks before slipping quietly into the water. | Turtles bask on sunlit rocks before slipping quietly into the water. | {} | 4 | null | null | null | null | null | null | null | null | null | null | null | null | null | null |
Mismatched socks sometimes end up in laundry baskets after washing. | Mismatched socks sometimes end up in laundry baskets after washing. | {} | 4 | null | null | null | null | null | null | null | null | null | null | null | null | null | null |
null | null | {"balashkevych7": "ivankoval12", "WR2": "XM9", "110 626 5076": "423 291 4685"} | null | 8fcb7ad0-0632-43a5-be56-9fae6f73cb21 | pii_masking | <html><body><p>balashkevych7, assess for benefits on 04/03/2048 at 10 AM. Living at ZIP WR2, Occupation: Artist. Confirm SSN 110 626 5076, enter OTP 199548.</p></body></html> | balashkevych7||WR2||110 626 5076 | <html><body><p>[REDACTED], assess for benefits on 04/03/2048 at 10 AM. Living at ZIP [REDACTED], Occupation: Artist. Confirm SSN [REDACTED], enter OTP 199548.</p></body></html> | 194,206 | en | 1en.jsonl | <html><body><p>ivankoval12, assess for benefits on 04/07/2048 at 10 AM. Living at ZIP XM9, Occupation: Artist. Confirm SSN 423 291 4685, enter OTP 199548.</p></body></html> | gpt-4.1 | 9 | The replacements are well executed: usernames and ZIP codes are swapped with plausible alternatives, and the SSN is replaced with a validly formatted number. The occupation and other non-PII details remain unchanged, preserving semantic meaning. The only minor issue is the unchanged OTP, which may or may not be considered PII depending on context, but typically OTPs are ephemeral and not reused. Overall, the replacement is highly effective and maintains cultural and semantic consistency. | true | |
The Parliament House in Canberra features a grass-covered roof and hosts Australia’s legislative sessions. | The Parliament House in Canberra features a grass-covered roof and hosts Australia’s legislative sessions. | {} | 4 | null | null | null | null | null | null | null | null | null | null | null | null | null | null |
An emergency fund provides a financial cushion for unexpected expenses. | An emergency fund provides a financial cushion for unexpected expenses. | {} | 4 | null | null | null | null | null | null | null | null | null | null | null | null | null | null |
null | null | {"X889856078343": "Z663715204932", "459-78-4613": "248-65-7921"} | null | ca65ca70-054c-482e-9836-8b8f259badf5 | pii_masking | Have any familial relationships been affected? If yes, include your X889856078343 and 459-78-4613. | X889856078343||459-78-4613 | Have any familial relationships been affected? If yes, include your [REDACTED] and [REDACTED]. | 173,147 | en | 1en.jsonl | Have any familial relationships been affected? If yes, include your Z663715204932 and 248-65-7921. | gpt-4.1 | 10 | The replacements maintain perfect semantic preservation, as the structure and intent of the query remain unchanged. The PII values (an ID and an SSN) are replaced with values of the same format, preserving format validity. There is no cultural inconsistency, as both the original and replacement values are generic and appropriate for the context. Anonymization is effective, with no leakage of original PII. This is an excellent replacement. | true | |
null | null | {"Colliers": "Silverton", "78758357502": "63219467158", "April/65": "March/65", "Hanson-Aguilar": "Hanson-Cortez"} | null | 267f53de-2237-4d00-995c-57373c339db4 | pii_masking | MBSR participation approved for Feminine, resident of Colliers, ID: 78758357502, DoB: April/65, managed by Hanson-Aguilar. | Colliers||78758357502||April/65 | MBSR participation approved for Feminine, resident of [REDACTED], ID: [REDACTED], DoB: [REDACTED], managed by Hanson-Aguilar. | 363,522 | en | 1en.jsonl | MBSR participation approved for Feminine, resident of Silverton, ID: 63219467158, DoB: March/65, managed by Hanson-Cortez. | gpt-4.1 | 9 | The soft replacements are well-executed: 'Colliers' to 'Silverton' maintains the locality structure, both plausible place names; the ID is replaced with another valid-looking number of the same length; 'April/65' to 'March/65' keeps the date format and similar age; 'Hanson-Aguilar' to 'Hanson-Cortez' preserves a similar cultural background and hyphenated surname structure. All PII is replaced, formats are valid, and semantic meaning is preserved. Minor issue: month change in DoB is a slight semantic drift, but overall, this is an excellent replacement. | true |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.