Skip to content

Tutorial 01 — Dry Run (No zk-SNARKs)

Goal: Get the framework running end-to-end in minutes without any circuit compilation or proof generation. This is the best starting point for understanding the data flow.

Prerequisites: Python dependencies installed (pip install -r requirements.txt).

Step 1 — Run the default dry run

bash ZKP_MODE=0 python driver.py

ZKP_MODE=0 disables all EZKL machinery. Leaf transforms and aggregations are computed using standard PyTorch forward() calls instead of generating zk-SNARK proofs.

Step 2 — Read the output

You should see output similar to:

``` Root Value: tensor([[ 9.2000, 2.9600, ...]]) Root Hash: 0x3a1f...

===== INCLUSION TEST: Picking raw data value that exists ===== ... Path walk complete. Leaf found at position 0x... Proof generation skipped (ZKP_MODE=0).

===== EXCLUSION TEST: Picking raw data value that does not exist ===== ... Leaf at position 0x... is a default leaf. Non-membership confirmed. Proof generation skipped (ZKP_MODE=0). ```

What happened

  1. driver.py loaded 12 user records from the s&p_loglikelihood_debug dataset.
  2. Each record's salted value was passed through the FlowThrough transformer (identity function).
  3. The Adder aggregated the transformed values into the Merkle tree bottom-up.
  4. The root value and hash were printed.
  5. An inclusion test was run on user_3 (a member of the dataset).
  6. An exclusion test was run on an absent record (not in the dataset).

Step 3 — Try a different transformer

bash ZKP_MODE=0 LTR_CHOICE=affine python driver.py

The Affine transformer applies m * x + c to each salted record before aggregation. The root value will be different but the membership/non-membership logic is identical.

Try each of:

bash ZKP_MODE=0 LTR_CHOICE=flow-through python driver.py ZKP_MODE=0 LTR_CHOICE=affine python driver.py ZKP_MODE=0 LTR_CHOICE=exponent python driver.py ZKP_MODE=0 LTR_CHOICE=length python driver.py

Step 4 — Understand the dataset

Open driver.py and look at the s&p_loglikelihood_debug branch (line ~53). Each entry in raw_data has:

  • name — a label (not used cryptographically).
  • value — a (1, 16) tensor: 6 binary features followed by 10 continuous values as the user salt.
  • transform_salt — a (1, 2) tensor appended after the transformer output.

The test_absent_data_record (labelled user_11 but with slightly different values) is the record used in the exclusion test.

What's next