Every WHL algorithm shown on this site has working code behind it. Below are seven interactive demos that run the actual algorithm logic in your browser — no backend, no signup, no email gate. Open one, click around, see how the substrate behaves.
Each card below opens a working demo. The first four link out to the product or platform page that hosts them. The last four are embedded directly on this page. Nothing here is a mock — these are the same gates, verifiers, and curves that run inside CB-12, SDM, Codex, and the Governance SDK.
Move 10 sliders. Watch the AND conjunction evaluate live. When any gate drops below 0.5, see the weakest-link reporting fire.
Paste responses one at a time. Watch the verdict change from WARMUP to ALLOW to DRIFT_DETECTED to SPECTRAL_DIVERGE. SHA-256 fingerprint on every row.
Type a command. See it compile simultaneously to six target outputs: FPGA frame, JSON workflow, REST endpoint, Python AST, mesh frame, FSM transition.
Browse all 72 named adversarial attacks. See sample measured values, gate targets, and verdicts. Or hit Run Sweep to cycle through all 72 live.
Paste a JSON ledger. Verify the hash chain. Or load a sample tampered chain to see exactly which row breaks.
Slide the entropy delta. Cross the 1/φ threshold (≈ 0.618) and see the veto fire. The simplest of the gates, but a novel use of phi.
Feed the engine a stagnation history. Watch it generate 5 child-domain mutations, sandbox-test them, and promote the winner. The "break the ceiling" mechanism, live.
Self-prediction surprise dropped from 0.819 to 0.027 across 64,184 production cycles. See the actual learning curve as an SVG line plot.
Paste a JSON-lines ledger. The verifier checks every row against its predecessor's hash. Tampering at any row shows the break-point.
The 1/φ ratio (golden ratio's inverse) is the maximum allowable entropy delta for a single action. Crossing it indicates the action is too high-novelty to autonomously approve.
The meta-evolution mechanism. When the current optimization domain plateaus, the engine generates structurally diverse mutations, sandbox-tests them, and promotes the winner. Not parameter-evolution — domain-evolution.
Empirical Friston-style active inference, measured on disk. Each point on the curve is a moving-average self-prediction surprise. The line falls from 0.819 to 0.027 — a 96.8% reduction.
Each demo here corresponds to a shippable product or platform component. Scoped pilots run six to twelve weeks with a written engineering report and replayable receipt chain at the end.