Lab
138 explorable, interactive pieces — every concept we write about, made tangible.
Seven years of the Elliptic AML benchmark in four switchable charts: the 2019 table everyone forgot, the 2026 leakage-free re-ranking, one GraphSAGE trained on four different graphs, and the F1 cliff at time step 43. Hover, tap, or tab across the bars for exact precision and recall.
Semaphore's double-signal defense, explorable: send transactions to spend month-scoped nullifiers PBH-style, replay a spent proof to watch the gate revert with the real contract errors, and roll the month to see the external nullifier change reset the quota.
What a hijacked agent key costs under three custody models: a raw EOA key loses the whole wallet instantly, an expiry-only session key does too, and a per-period spend permission caps the bleed. Drag the sliders — or load a real 35.97 USDC/day permission pulled off Base — and watch the staircase.
Story Protocol's two royalty policies, runnable: tap any IP asset in a derivative chain to pay 100 WIP into it and watch LAP give every ancestor its absolute cut until the stack hits 100% and mints revert, while LRP compounds hop by hop and dilutes the original to dust.
A real Omen prediction market on Gnosis, replayed bet by bet: eight AI-agent trades walked P(YES) from 50% to 42% in 13 hours. Tap any bet to rewind the pools, then step in as the ninth bettor — set your estimate, bankroll, and stake, and compare it to the Kelly bet an agent would place.
Push the cost-of-corruption inequality for a restaking-secured AI oracle: set bonded stake, slash rate, audit probability, and value-at-stake, then add AVSs to watch the overloading attack flip the verdict from secure to exploitable.
The zkTLS trust boundary, explorable: switch between Plain TLS, MPC-TLS, Proxy-TLS and TEE-TLS to see where the session key lives and whom you must trust, then hit "forge the value" — plain TLS accepts the lie, the other three catch it at different costs.
Forward-pass gas for five neural nets run inside the EVM, on a log axis against the live Ethereum (60M) and Base (400M) block-gas ceilings. Toggle ML2SC's measured ~106k gas/edge vs an optimized ~1.5k floor, slide the gas price for a USD readout, and watch an MNIST net blow past 180 Ethereum blocks.
One hard question, a swarm of LLM nodes, two ways to agree. Run a round: majority voting picks the most common answer while peer-ranked Bradley-Terry consensus surfaces the best one, and the scoreboard converges on the gap Fortytwo measured. The evaluation-edge slider shows why.
Why memory injection beats prompt injection against on-chain AI agents, in three switchable charts over CrAIBench: 685 attacks skewed to trading, a 55.1%-vs-0% gap on the strongest model, and the fine-tuning defense that drops attack success 85.1% to 1.7%. Tap or arrow the bars for numbers.
Why re-executing an LLM doesn't reproduce it. One matmul reduces to the logit gap between 'Queens, New York' and 'New York City'; drag the production split-K and watch IEEE-754 rounding flip the token at split 5 and 6, breaking the verifier's digest check. Batch-invariant kernels turn it green.
Where reliability comes from in on-chain LLM trading agents, over DX Terminal Pro's production run: model upgrades move swap success 87% to 96%, but the operating layer takes the same weights to 99.9%. Switch panels for the prompt fixes that erased three failure modes. Tap or arrow the bars.