Verifiable, auditable, correctable.

AI research studio specializing in security & invariants. · caliperforge.com

CaliperForge researches autonomous systems you can verify, audit, and correct. Every shipped harness carries a clean reference where the invariant holds, a planted-bug twin where it fires, and a CI-asserted record of what our own gates caught. The receipts are the proof. See what we caught.

AI involvement is disclosed at point of use; full policy at caliperforge.com/ai-disclosure.

One operator (Michael Moffett, named below) orchestrates a small team of specialized AI agents: a Rust / Anchor specialist, an audit engineer, a content reviewer, and others as the work calls for them. Each agent is scoped, has a defined output discipline, and cross-checks the others before anything leaves the workspace. The operator reviews every load-bearing decision, runs cold-environment reproductions, and ships under his own name as operator-of-record.

Specialization velocity is the working pattern: a new chain, a new VM, a new specialist and a CI-verified tool with it, in days rather than months. Solana / Anchor on Asymmetric Research's Crucible is the current lane: four third-party programs ported onto shared rails, each on its own repo with its own CI, the fourth CI-green within 48 hours of the first. The current portfolio is protocol and smart-contract security.

Pinned below, Solana first. Everything else is on the repository list.

• cf-invariants — the AI invariant-author for Crucible and Trident on Solana / Anchor. A sidecar to the ecosystem's fuzzers, not a new one.
• solana-invariant-atlas — recurring Solana bug classes as clean / planted twin pairs on real programs, multi-seed reachability certified in CI.
• x402-regression-base — planted-twin regression harness for the x402 agentic-payments settlement threat model. Reference facilitator deployed and exercised on Base mainnet and Sepolia; receipts in docs/basescan/.
• uniswap-v4-invariants — defender-side invariant harness for Uniswap v4 hooks, on real v4-core.
• invariant-atlas — six historical DeFi exploits across four VMs, each reconstructed as a same-source clean / planted twin. Defender-side, pre-deploy.
• cf-modeleval — planted-twin discrimination-power harness for AI safety properties, across three providers.

The rest of the public listing. Same disclosure, same CI discipline.

• cf-invariants-starknet — snforge sidecar adding stateful invariant testing and AI-suggested invariants to Cairo 2.x. Twelve reference contracts, Voyager-verified on Starknet Sepolia.
• bsc-invariants — invariant testing and planted-twin CI for BSC DeFi protocols. PancakeSwap v3 harness live; Venus and Stargate planned.
• taiko-equivalence — differential equivalence tests for Taiko's Type-1 Ethereum-equivalence guarantee, encoded as clean / planted twin Foundry projects.
• agentic-payments-regression — CI-runnable planted-twin reproductions for the x402 agentic-payments threat model. Clean / planted pairs, INVARIANT VIOLATED marker convention, 16-seed reachability certification.
• solana-property-benchmark — pre-registered benchmark measuring how well automatically generated security invariants cover known vulnerability classes in Solana Anchor programs. The evaluation protocol was published before any result existed.
• apart-global-south-lost-in-translation — language-conditioned detection-rate eval harness for AI code auditors. EN / ES / PT / CS variant corpus built on Atlas planted-bug twins. Apart Global South AI Safety Hackathon 2026.
• taler-auditor-fault-injection — mutation-based fault-injection coverage for the GNU Taler exchange auditor (#9828 recoup-of-refresh, post-#9900 refresh paths).
• chimera-template-pack — reusable Foundry and Recon Chimera scaffold for EVM build-to-win contest entries, CI-verified.

• 5 Solana programs CI-green: Jito tip-distribution, tip-payment, priority-fee-distribution, the Pyth Solana Receiver, and an Anchor reference
• 4 VMs covered by the cross-VM exploit atlas
• 1 operator of record

What our own gates caught, published with provenance, at caliperforge.com/what-we-caught.

• Active chains: Solana, Ethereum / EVM, Base.
• Languages: Rust / Anchor, Solidity, Cairo.
• What we author: stateful-invariant suites, AI-suggested invariants tagged in source, CI-verified clean / planted-bug reference examples, security tooling on top of incumbent fuzzers (Crucible, Trident, snforge).
• What we don't claim: coverage parity or superiority with the underlying fuzzing engines. We author the layer above them.

Operated by Michael Moffett, accountable for every commit, PR, grant application, and bounty claim made under this org.

• Direct: michael@caliperforge.com
• Web: caliperforge.com
• X: @caliperforge · Farcaster: @caliperforge · Telegram: @caliperforge
• ENS: caliperforge.eth

For grant collaboration, hired-gun engagements, security tooling inquiries, or contribution questions: michael@caliperforge.com.