Dataset Viewer
Auto-converted to Parquet Duplicate
synthetic
bool
1 class
id
stringclasses
10 values
given_name
stringclasses
10 values
family_name
stringclasses
8 values
date_of_birth
timestamp[s]date
1957-02-27 00:00:00
2004-03-30 00:00:00
emails
listlengths
1
1
usernames
listlengths
1
1
phones
listlengths
1
1
addresses
listlengths
1
1
employment
listlengths
1
1
education
listlengths
1
1
national_ids
listlengths
1
1
true
persona-0001
Joel
Fisher
2003-03-04T00:00:00
[ { "synthetic": true, "value": "synth_joel_fisher_0001@example.test", "kind": "primary" } ]
[ { "synthetic": true, "value": "synth_joel_fisher_0001" } ]
[ { "synthetic": true, "value": "+1-200-555-0100" } ]
[ { "synthetic": true, "house_number": 100, "street_name": "1 Example Avenue", "city": "Testville", "postal_code": "00000", "country_code": "ZZ" } ]
[ { "synthetic": true, "organization": "Example Works 0001", "role": "Synthetic Researcher" } ]
[ { "synthetic": true, "institution": "Test University 0001", "graduation_year": 2025 } ]
[ { "synthetic": true, "value": "SYN-202607199", "checksum_valid": false } ]
true
persona-0002
Leslie
Fisher
1975-04-24T00:00:00
[ { "synthetic": true, "value": "synth_leslie_begum_0002@example.test", "kind": "primary" } ]
[ { "synthetic": true, "value": "synth_leslie_begum_0002" } ]
[ { "synthetic": true, "value": "+1-200-555-0101" } ]
[ { "synthetic": true, "house_number": 100, "street_name": "1 Example Avenue", "city": "Testville", "postal_code": "00000", "country_code": "ZZ" } ]
[ { "synthetic": true, "organization": "Example Works 0002", "role": "Synthetic Researcher" } ]
[ { "synthetic": true, "institution": "Test University 0002", "graduation_year": 1997 } ]
[ { "synthetic": true, "value": "SYN-202607207", "checksum_valid": false } ]
true
persona-0003
Bradley
Fletcher
2004-03-30T00:00:00
[ { "synthetic": true, "value": "synth_bradley_fletcher_0003@example.test", "kind": "primary" } ]
[ { "synthetic": true, "value": "synth_bradley_fletcher_0003" } ]
[ { "synthetic": true, "value": "+1-200-555-0102" } ]
[ { "synthetic": true, "house_number": 102, "street_name": "3 Example Avenue", "city": "Testville", "postal_code": "00000", "country_code": "ZZ" } ]
[ { "synthetic": true, "organization": "Example Works 0002", "role": "Synthetic Researcher" } ]
[ { "synthetic": true, "institution": "Test University 0003", "graduation_year": 2026 } ]
[ { "synthetic": true, "value": "SYN-202607215", "checksum_valid": false } ]
true
persona-0004
Jacob
Sullivan
1981-01-13T00:00:00
[ { "synthetic": true, "value": "synth_jacob_sullivan_0004@example.test", "kind": "primary" } ]
[ { "synthetic": true, "value": "synth_jacob_sullivan_0004" } ]
[ { "synthetic": true, "value": "+1-200-555-0103" } ]
[ { "synthetic": true, "house_number": 103, "street_name": "4 Example Avenue", "city": "Testville", "postal_code": "00000", "country_code": "ZZ" } ]
[ { "synthetic": true, "organization": "Example Works 0004", "role": "Fixture Designer" } ]
[ { "synthetic": true, "institution": "Test University 0003", "graduation_year": 2026 } ]
[ { "synthetic": true, "value": "SYN-202607223", "checksum_valid": false } ]
true
persona-0005
Denis
Lawson
1965-06-07T00:00:00
[ { "synthetic": true, "value": "synth_denis_lawson_0005@example.test", "kind": "primary" } ]
[ { "synthetic": true, "value": "synth_denis_lawson_0005" } ]
[ { "synthetic": true, "value": "+1-200-555-0104" } ]
[ { "synthetic": true, "house_number": 104, "street_name": "4 Example Avenue", "city": "Testville", "postal_code": "00000", "country_code": "ZZ" } ]
[ { "synthetic": true, "organization": "Example Works 0005", "role": "Synthetic Researcher" } ]
[ { "synthetic": true, "institution": "Test University 0005", "graduation_year": 1987 } ]
[ { "synthetic": true, "value": "SYN-202607231", "checksum_valid": false } ]
true
persona-0006
Samuel
Fuller
1958-08-15T00:00:00
[ { "synthetic": true, "value": "synth_samuel_fuller_0006@example.test", "kind": "primary" } ]
[ { "synthetic": true, "value": "synth_samuel_fuller_0006" } ]
[ { "synthetic": true, "value": "+1-200-555-0105" } ]
[ { "synthetic": true, "house_number": 105, "street_name": "6 Example Avenue", "city": "Testville", "postal_code": "00000", "country_code": "ZZ" } ]
[ { "synthetic": true, "organization": "Example Works 0006", "role": "Test Engineer" } ]
[ { "synthetic": true, "institution": "Test University 0006", "graduation_year": 1980 } ]
[ { "synthetic": true, "value": "SYN-202607249", "checksum_valid": false } ]
true
persona-0007
Susan
Fuller
1973-09-26T00:00:00
[ { "synthetic": true, "value": "synth_susan_smith_0007@example.test", "kind": "primary" } ]
[ { "synthetic": true, "value": "synth_susan_smith_0007" } ]
[ { "synthetic": true, "value": "+1-200-555-0106" } ]
[ { "synthetic": true, "house_number": 105, "street_name": "6 Example Avenue", "city": "Testville", "postal_code": "00000", "country_code": "ZZ" } ]
[ { "synthetic": true, "organization": "Example Works 0007", "role": "Fixture Designer" } ]
[ { "synthetic": true, "institution": "Test University 0007", "graduation_year": 1995 } ]
[ { "synthetic": true, "value": "SYN-202607256", "checksum_valid": false } ]
true
persona-0008
Annette
Jones
1965-01-16T00:00:00
[ { "synthetic": true, "value": "synth_annette_jones_0008@example.test", "kind": "primary" } ]
[ { "synthetic": true, "value": "synth_annette_jones_0008" } ]
[ { "synthetic": true, "value": "+1-200-555-0107" } ]
[ { "synthetic": true, "house_number": 107, "street_name": "8 Example Avenue", "city": "Testville", "postal_code": "00000", "country_code": "ZZ" } ]
[ { "synthetic": true, "organization": "Example Works 0007", "role": "Fixture Designer" } ]
[ { "synthetic": true, "institution": "Test University 0008", "graduation_year": 1987 } ]
[ { "synthetic": true, "value": "SYN-202607264", "checksum_valid": false } ]
true
persona-0009
Paula
Robson
1991-10-14T00:00:00
[ { "synthetic": true, "value": "synth_paula_robson_0009@example.test", "kind": "primary" } ]
[ { "synthetic": true, "value": "synth_paula_robson_0009" } ]
[ { "synthetic": true, "value": "+1-200-555-0108" } ]
[ { "synthetic": true, "house_number": 108, "street_name": "9 Example Avenue", "city": "Testville", "postal_code": "00000", "country_code": "ZZ" } ]
[ { "synthetic": true, "organization": "Example Works 0009", "role": "Synthetic Researcher" } ]
[ { "synthetic": true, "institution": "Test University 0008", "graduation_year": 1987 } ]
[ { "synthetic": true, "value": "SYN-202607272", "checksum_valid": false } ]
true
persona-0010
Kathryn
Rhodes
1957-02-27T00:00:00
[ { "synthetic": true, "value": "synth_kathryn_rhodes_0010@example.test", "kind": "primary" } ]
[ { "synthetic": true, "value": "synth_kathryn_rhodes_0010" } ]
[ { "synthetic": true, "value": "+1-200-555-0109" } ]
[ { "synthetic": true, "house_number": 109, "street_name": "9 Example Avenue", "city": "Testville", "postal_code": "00000", "country_code": "ZZ" } ]
[ { "synthetic": true, "organization": "Example Works 0010", "role": "Test Engineer" } ]
[ { "synthetic": true, "institution": "Test University 0010", "graduation_year": 1979 } ]
[ { "synthetic": true, "value": "SYN-202607280", "checksum_valid": false } ]

SynthWorld Frozen Benchmarks

Deterministic, connected synthetic identity graphs with ground-truth answer keys, for testing privacy, PII-extraction, entity-resolution, and exposure-analysis systems — without collecting or fabricating data about real people.

Unlike sentence-level PII corpora, SynthWorld records are linked: personas form a connected society with an explicit relationship graph (family / colleague / classmate / neighbour / social), and every relationship edge carries the exact evidence that supports it. Each benchmark family keeps the product-safe public input physically separate from the evaluator-only answer key.

These artifacts are the frozen golden benchmarks of the SynthWorld generator (pip install idcognito-synthworld, v0.8.0+), generated from seed 20260719 at ten personas and authenticated by SHA-256 manifests. The generator's CI regenerates them byte-for-byte on every commit and fails on any drift, so scores measured against these files are comparable across time and systems.

Every record is unmistakably fake

Safety is mechanical, not procedural — enforced by the generator's type system and test suite:

  • every object carries synthetic: true;
  • emails use the reserved example.test domain;
  • phone numbers sit in the fictional North American 555-01xx block;
  • addresses are on example-named streets in Testville, postal code 00000, country ZZ;
  • national identifiers carry a SYN- prefix with deliberately invalid checksums.

This dataset must never be used to impersonate, target, or investigate any person. There is nothing to de-anonymise: no real person's data was used at any stage.

What's in the repository

frozen/ holds the authoritative artifacts, byte-identical to the ones shipped inside the Python package, plus their SHA-256 manifests:

File SHA-256
frozen/golden-v1.json 8b75fcd932dbbe2d0ea94d034f8c546c6c3857d3c99669180222f807cf48755d
frozen/extraction-golden-v1.json 69bf567bf122ed7831f0963883b0524c3b9d991b5ef0f7a5b6b7ce69ee234e57
frozen/extraction-public-golden-v1.json 10632f000f8aeb8ccd8557476b18b940cfd35b91f7cb38dcf209269de987160e
frozen/extraction-answer-golden-v1.json ffc6503df8cbb9d8f99161ee29324e8d0a0187901118e8eeaa590b49e7598f78
frozen/connection-golden-v1.json 044b52650039059b5841e0af9c512e2bbc7dbb089d43e465d43fda06889a8fe4
frozen/connection-public-golden-v1.json fa896ae417f75d6fc4ac650ec26683a39b3994bc004c743fc1fcc71f605ff17e
frozen/risk-public-golden-v1.json 690c2fb081826f72970af1e729651819c3563d9aa590190d566af24424238b33
frozen/risk-answer-golden-v1.json 32479aa077887a63d31a4de3dfbc822f01f6622f09ea6dd6d2a87e3af3cb319e
  • golden-v1.json — the exposure corpus: the world (10 personas, 9 evidence-backed relationship edges) plus one scripted exposure history per persona (breach, broker, search, and social records).
  • extraction-public-golden-v1.json / extraction-answer-golden-v1.json — the exact-span PII extraction benchmark: 62 product-safe pages, and a physically separate answer key with the exact character span of every planted identifier. Feed a system only the public pages; join the answer key afterwards to score. (extraction-golden-v1.json is the equivalent annotated bundle for offline evaluators — page and answer key together, so it is not a product-safe input.)
  • connection-golden-v1.json / connection-public-golden-v1.json — the adversarial entity-resolution benchmark and the product-safe public half (18 raw identity records). Public inputs and oracle truth are separate files by design.
  • risk-public-golden-v1.json / risk-answer-golden-v1.json — risk-calibration cases: opaque public observations, physically separated from the score-and-factor answer key.

viewer/ holds JSONL projections of the frozen artifacts so the Dataset Viewer can render them as tables (personas, relationships, public_extraction_pages, extraction_answers, public_identity_records). They are derived conveniences; only frozen/ is authoritative and checksummed.

Score a system against these benchmarks

Since v0.8.0 the package ships a unified evaluation SDK. A system consumes only the public input, emits predictions in an oracle-free schema, and the evaluator loads the answer key itself:

pip install idcognito-synthworld
synthworld evaluate extraction --predictions my_predictions.json --seed 20260719 --summary

Tasks: extraction, entity-resolution, relationship, risk. The report gives per-task metrics (exact/relaxed span F1, pairwise + B-cubed, edge F1 with evidence coverage, risk band accuracy + Brier) and failure slices, tied to the scored bytes by content digest.

The public/oracle boundary

Answer keys exist so evaluators can score systems — a pipeline that reads its own answer key is measuring nothing. The public artifacts (extraction-public-*, connection-public-*, risk-public-*) carry only what a system may see; the answer keys are separate files. The generator enforces this split in the product constructors; keep it when you build on these files. Because the golden benchmark ships in this repository the answer keys are public, so this separation is an API-hygiene guarantee, not an anti-cheating measure — adversarial or leaderboard use needs held-out private seeds.

Quick start

Browse tables with 🤗 Datasets:

from datasets import load_dataset

pages = load_dataset("Bluntmachetti7/synthworld-benchmarks", "public_extraction_pages")
answers = load_dataset("Bluntmachetti7/synthworld-benchmarks", "extraction_answers")
print(pages["golden"][0]["content"])          # product-safe input
print(answers["golden"][0]["spans"])           # evaluator-only exact spans

Score against the authoritative frozen bytes:

from huggingface_hub import hf_hub_download
import hashlib, json

path = hf_hub_download(
    "Bluntmachetti7/synthworld-benchmarks",
    "frozen/extraction-answer-golden-v1.json",
    repo_type="dataset",
)
blob = open(path, "rb").read()
assert hashlib.sha256(blob).hexdigest() == (
    "ffc6503df8cbb9d8f99161ee29324e8d0a0187901118e8eeaa590b49e7598f78"
)
answer_key = json.loads(blob)

Regenerate everything from source, at any scale, with the generator:

synthworld generate --seed 20260719 --persona-count 100 --output world.json

Links

Licensed under Apache-2.0. Copyright 2026 Idcognito contributors.

Downloads last month
46