Privacy release instrument

Noise Ledger

Target delta
1 × 10−5
Epsilon cap
3.0
Mechanism
Gaussian

Composition before publication

Compose the plan before the data speaks.

Each query looks small alone. Repetition accumulates privacy loss. Noise Ledger composes the full plan in Renyi space and tests it against one explicit release cap.
Composed estimateε 2.229
WITHIN CAP · ε 2.229

alpha 8 · composed RDP 1.0144 · converted epsilon 2.229

Planned releases

Mechanism ledger

Adjust a dose, then issue a fresh receipt.
Weekly active users
01Weekly active usersUnsampled Gaussian
1.0
24.6%
Retention cohort
02Retention cohortUnsampled Gaussian
1.0
43.8%
Support topic counts
03Support topic countsUnsampled Gaussian
1.0
31.5%
foundry-honeypot-ready

Inputs remain in this browser. This model does not inspect or process a dataset.

Release decision / bounded model

The plan fits the cap.

WITHIN CAP
Composed epsilon
2.229
Cap margin
+0.771
Best order
α 8
Basic composition
ε 5.384

Largest RDP contributor at α 8

Retention cohort

43.8% of composed RDP
  • PASSmore repetitions never reduce loss
  • PASSmore sensitivity never reduce loss
  • PASSmore noise never increase loss

Educational accountant for unsampled Gaussian mechanisms. Uses fixed Renyi orders and a documented RDP conversion bound. It does not model contribution bounding, subsampling amplification, PLD accounting, secure noise generation, production datasets, or legal compliance.

Google DP accounting reference ↗
foundry_page_view foundry_signup_submit foundry_core_action_submit foundry_core_action_complete