Approve, deny, route, and enforce policy — without ever exposing the underlying data. No decryption. No plaintext. No exceptions.
The system evaluates encrypted values and produces a decision. The data is never exposed.
Traditional FHE schemes — BFV, CKKS — can add and multiply encrypted data. They are powerful arithmetic engines.
They cannot compare values. They cannot evaluate thresholds. They cannot branch. They cannot enforce rules.
That is why real systems still expose data to make decisions.
TFHE changes that.
Addition. Multiplication. Inner products. Polynomial operations on encrypted data.
Comparisons. Equality checks. Threshold decisions. Branching logic. All on encrypted data.
The server does not know the value. It does not know the identity. It does not know the input. It does not know the underlying data.
The system evaluates every possible path and cannot determine which one applies.
All numbers measured on AWS Graviton4 (ARM), sustained over 30 seconds.
| Operation | Precision | Throughput |
|---|---|---|
| Comparison | 8-bit | 768 TPS |
| Comparison | 16-bit | 372 TPS |
| Comparison | 32-bit | 182 TPS |
| Comparison | 64-bit | 91 TPS |
| Equality | 16-bit | 769 TPS |
Every doubling in precision doubles the cost. Performance is linear and predictable.
You do not choose the encryption scheme. The system does.
The FHE-IQ router selects the correct engine automatically.
Evaluate risk thresholds without seeing transactions. The system compares encrypted scores against encrypted cutoffs and returns a pass or fail.
Match records across institutions without exposing identities. Encrypted equality checks determine whether two encrypted values correspond — without revealing either one.
Approve or reject transactions without revealing amounts. Encrypted comparisons evaluate thresholds on ciphertext, not plaintext.
Check expirations and session validity without reading timestamps. Full 64-bit Unix timestamp comparison on encrypted values at 91 TPS.
What TFHE does not do efficiently.
TFHE is optimized for decision logic: comparisons, equality checks, threshold evaluations, and branching.
It is not efficient for large hash functions, high-depth boolean circuits, or full cryptographic primitives on encrypted data.
Example: SHA3-256 evaluation on encrypted data → approximately 0.30 TPS.
We publish this because performance limits matter.
TFHE is based on lattice cryptography (Learning With Errors). The hardness assumption that protects every encrypted decision is the same class of problem that underpins all NIST post-quantum standards.
Every decision is sealed with H33-74 — a 74-byte post-quantum proof. 32 bytes on-chain. 42 bytes off-chain.
Signed using three independent hardness assumptions: MLWE lattices, NTRU lattices, and hash-based signatures.
This creates a permanent, verifiable record of a decision made without exposing data.
A system that computes, decides, and proves — without ever exposing data.