US Provisional Patent Pending (USPTO)
For banks, exchanges & high-throughput signing infrastructure

Hardware-Backed Key Security.
Zero Trust Custody.

Signing keys that rotate themselves, in microseconds — validated on real AWS Nitro Enclave and real AWS CloudHSM hardware.

A stolen key shouldn't stay a liability

Signing keys inside a TEE/HSM setup rotate themselves automatically, in microseconds — so a leaked or stolen key is stale before it's worth using, without slowing down signing.

The exposure window

Today, a signing key lives until someone manually notices a breach and rotates it — hours, sometimes weeks. Every minute it stays valid, a stolen key is still a fully usable liability for an attacker.

Why rotating more often isn't done today

Interactive HSM key rotation means a real network round trip — our own measured, best-case number (same-VPC, real AWS CloudHSM) is ~120ms, and that's the fast case. Too slow to do continuously on high-throughput signing infrastructure. Teams are forced to choose between frequent rotation and performance.

Where this shows up

Hot/warm wallet signing infrastructure, custody platforms, exchange withdrawal signing, payment/settlement message signing, crypto trading bots that sign every order (API keys or on-chain transactions), and any infrastructure — crypto or traditional finance — where a single compromised key can move real funds.

How it works

Autonomous key ratcheting — the mechanism, in three steps.

1

Bootstrap

TEE ↔ HSM run one ECDHE handshake and mutual attestation. Agree on K₀. One-time cost: ~0.3-0.5ms.

2

Session-key ratchet

Each side independently derives session key Kn+1 = HKDF-SHA256(Kn, salt=hardware measurement, info=counter). No network call.

3

Verify & rotate

Sliding-window replay protection + signed batch proofs make rollback attacks detectable. Measured: ~3µs/step.

Analogy

A self-reloading keycard, good for a set number of uses — the signer works at full speed without checking back with the vault each time. If tampered with, it instantly deactivates. Tachpawl does this for cryptographic signing keys.

What doesn't change

We don't replace the HSM — it stays the root of trust, generating the master seed and never releasing it. We remove it from the hot path, not from the trust model. Patent pending (USPTO provisional application, priority date 2026-07-13).

Why it matters for you

Exchanges, custodians & payment processors

Hot/warm wallet withdrawal signing, or payment and settlement message signing, is the highest-value attack surface you operate. Automatic, hardware-backed key rotation shrinks what any single compromised signing key is worth — without adding latency.

Signing infrastructure providers

If you provide signing infrastructure for other companies' wallets or payment rails, key-rotation frequency is a direct sales argument to your own customers — and one you can currently only offer at a real performance cost.

Regulated / institutional custody

Auditors and institutional clients increasingly ask about key-rotation policy, not just key-storage policy. Continuous, hardware-enforced rotation is a stronger answer than a manual quarterly rotation runbook.

What we're not saying: not a claim that key rotation alone stops every attack — it shrinks the value of one specific, common failure mode (a leaked or stolen signing key) without a performance cost. Not a replacement for your HSM or your existing key-custody policy. In scope: network/software key theft — the overwhelming majority of real incidents. Out of scope: an attacker who physically defeats a device's own tamper response and extracts key material directly — no HSM or TEE vendor claims protection against that, and neither do we.

Throughput at scale

A real 1,000,000-transaction run on real AWS Nitro + real AWS CloudHSM — measured end to end.

Wall clock
31 sec
1,000,000 real transactions, start to finish, real AWS Nitro Enclave
Windows authorized
1000/1000 100%
approved by a real, separate AWS CloudHSM cluster
Key rotation
3.12 µs/step (mean)
1,000/1,000 unique rotating signing addresses — zero repeats
vs. Visa/VisaNet's own published numbers: real average 1,700-3,000 tx/sec, real peak 4,000-7,000 tx/sec, theoretical maximum >65,000 tx/sec. This run's per-window throughput averaged 36,588 tx/sec (p10-p90: 20,523-48,609 tx/sec) — 5-12x Visa's real-world average and peak load. This is not a claim against their theoretical maximum.
What this run actually measured: a real AWS Nitro Enclave (non-debug mode — genuine hardware attestation, not the placeholder value debug mode produces) consumed a real Kafka/Redpanda topic, and every one of the 1,000 authorization windows was independently re-verified and approved by a real, physically separate AWS CloudHSM cluster — not an in-process HSM stand-in, not a synthetic loop.
Also measured: a separate run on a single developer laptop (no real HSM in the loop) previously reached 187,989 tx/sec steady-state — useful as evidence the pipeline itself is never the bottleneck, not a production or HSM-authorized claim.

Real hardware, not slideware

Every number here was measured — none modeled or estimated.

Tachpawl ratchet step
~3 µs
real AWS Nitro Enclave, 200k-sample run
HSM key generation
~120 ms
real AWS CloudHSM, measured — the operation Tachpawl's ratchet replaces

Product stage: working implementation, not ideation

A real, running Rust implementation — HSM-side code validated against both SoftHSM2 and a real, rented AWS CloudHSM cluster; TEE-side code validated end-to-end on a real AWS Nitro Enclave with real NSM hardware attestation (PCR0). Real on-chain settlement is proven too, submitted from outside the enclave since it has no network of its own — 17/17 real broadcasts succeeded.

IP

USPTO provisional patent application filed, priority date 2026-07-13.

DeFi & EVM compatible

Transaction signing uses secp256k1 + Keccak256 — the same primitives EVM chains use, so every signature is natively verifiable on-chain via Ethereum's own ecrecover, no custom verifier needed.

Threshold-ECDSA — real and network-verified

The signing backend is already pluggable behind one Rust trait. Cluster mode (single CloudHSM/SoftHSM2) is the production default today.

Real multi-party signing, no single key ever exists in one place

Each party runs its own separate process, and a real 2-of-3 threshold-ECDSA ceremony has been run across three genuinely separate AWS machines, converging on one shared public key and one byte-identical signature.

Re-run across three continents to get a real number, not a guess

Stockholm, Virginia, and Singapore — signing between the two farthest parties took 4-6 seconds real wall clock, still trivial against a once-per-window amortization argument.

What's left is organizational, not technical: today's demo still has one operator behind every party — genuine trust separation needs each party under a genuinely different controlling organization, which no infrastructure test alone can demonstrate. An independent-key multisig alternative mixing HSM, YubiKey and software keys per signer remains a second, separately buildable option for a client who prefers it.

Not locked to one cloud vendor's enclave

The identical tee-service image also runs, unmodified, on a second real TEE technology.

Verified on real Intel TDX hardware, not just booted

Beyond AWS Nitro Enclaves, the same Docker image runs inside a real Intel TDX confidential VM (Phala Cloud). A real TDX attestation quote was fetched from live hardware and independently verified against Intel's own certificate chain — not the platform's own say-so.

Stated plainly: we also tested whether this platform detects a stealth enclave restart the way AWS Nitro's module_id does. It partially does — it reliably tells two different deployments apart, but a plain restart of the same deployment isn't yet distinguishable that way. We'd rather state that boundary exactly than round it up.

See the live verification dashboard →

How hard we tried to break this

Every number below is a real test run against this exact codebase.

Tests passing
98
kdf-core + tee-adapter + hsm-adapter — 100%
Fuzz targets
7
tens of millions of executions, zero crashes
Mutants missed
0
124 mutants tested across both crates' locally-testable code

Unit & integration tests, everywhere

kdf-core: 69 tests. tee-adapter: 11 always-on + 4 real-AWS-S3 + 6 real-Nitro-hardware tests. hsm-adapter: 8 tests against both SoftHSM2 and a real, rented AWS CloudHSM cluster. 98 total, 100% passing, 97% line coverage on locally-testable code.

Fuzzed, not just tested

7 libFuzzer targets across every security-critical boundary — replay protection, rollback detection, attestation verify, ECDHE bootstrap, authorization tokens, frame parsing. Found and fixed 2 real exploitable bugs this way: an integer-overflow DoS and an X25519 key-recovery vulnerability — before anyone else could find them.

Mutation testing

cargo-mutants: 124 mutants tested across kdf-core and tee-adapter's locally-testable scope — 0 missed. Every gap it found got a real test, not a suppression.

What we're offering

Stated plainly, so there's no guessing.

FREE FOR THE FIRST DESIGN PARTNER

We're not selling a license. We're looking for one real deployment.

Integration is free for the first real design partner — an existing regulated custodian, exchange, or high-throughput signing operation willing to run this in a real or near-real deployment. No license fee, no integration fee.

In exchange: the right to reference the deployment as a real, in-production case study (scoped to whatever disclosure level you're comfortable with), and a revenue-share negotiated after a successful go-live — not before, and not assumed in advance.

If your team manages real signing infrastructure and key-rotation latency or exposure-window risk is something you've actually run into, that's the conversation worth having — not a sales pitch.

Let's talk

Code & benchmarks

Private GitHub repo — full Rust source, benchmark reports, and the attestation artifact. Access granted under NDA on request.

Next step

A short technical scoping call — what your current key-rotation setup costs you today, and whether this is a fit.