Consortium benchmarks
How the shared benchmark is built, which thresholds it enforces, and when it last rebuilt.
Current posture
Contributors
73
Privacy thresholds
k≥5 · ℓ≥3 · t≤0.2
Last rebuild
2026-05-14 04:12 UTC
Methodology in force
v3
Methodology version history
Methodology v3
In force- Effective from
- 2026-05-01
- Changes
- Tightened outlier handling for fund-size bands above $500M; added vintage-year cohort dimension (previously grouped into 5-year buckets); updated t-closeness threshold from 0.25 to 0.20 to match ARX financial-data defaults.
- Adversarial test
- View CI run (opens in new tab)
Methodology v2
Superseded 2026-05-01- Effective from
- 2026-03-15
- Changes
- Added ℓ-diversity (l≥3) as a second privacy criterion alongside k-anonymity; switched anonymization library from pycanon-only to anjana+pycanon dual stack.
- Adversarial test
- View CI run (opens in new tab)
Methodology v1
Superseded 2026-03-15- Effective from
- 2026-02-01
- Changes
- Initial methodology: k-anonymity (k≥5) only; IRR and TVPI bands; three cohort dimensions (asset_class, geography_region, vintage_bucket).
- Adversarial test
- View CI run (opens in new tab)
Re-identification test log
| Run date | Result | k | ℓ | t | CI run |
|---|---|---|---|---|---|
| 2026-05-14 04:15 UTC | PASS | 5 | 3 | 0.18 | View CI run (opens in new tab) |
| 2026-05-01 04:10 UTC | PASS | 6 | 3 | 0.17 | View CI run (opens in new tab) |
| 2026-04-15 04:08 UTC | PASS | 5 | 4 | 0.19 | View CI run (opens in new tab) |
Why the contributor list is not published
Identifying individual contributors would reverse the anonymity guarantee. We publish the count, the thresholds, and the methodology — not the roster.