Pick a compliance node or a gated workflow already in the suite, sweep one or two of its numeric inputs across a range, and see exactly where the decision flips. This is calibration evidence: fair-lending threshold testing, pre-deployment testing, or just understanding how sensitive a policy is to its inputs, before an agent runs it for real.
Client-side sweepUp to 10,000 evaluationsSVG outcome surfaceNot legal advice
🔒 All inputs are processed locally in your browser. No data is transmitted. Do not enter real personal data — use synthetic or anonymised inputs only.
Fair-lending calibration example: the two nodes this page was built to cross-link are test_bifsg_bias_thresholds (Colorado SB 21-169 insurance-proxy bias test) and compute_disparity_metrics (ECOA / HMDA four-fifths adverse-impact test). Both are in the target list below — sweep either one to find the exact input value where a bias flag starts firing.
Read-only over existing kernels: this page re-runs the same deterministic compute() logic each node already ships, it does not change any node, add a kernel, or touch chaingraph.json. The target list below is a curated subset (numeric-threshold nodes and one gated workflow), not the full catalog.
Want to run a node's compute() step by step in a sandboxed, hermetic VM instead of sweeping it? See the kernel VM: it runs a kernel inside an in-browser QuickJS-ng WebAssembly sandbox under the ocg-deterministic-compute@1 profile and compares the execution_hash against a native run in this same tab.
1. Choose a target
2. Set inputs and pick what to sweep
0≤ 10,000 cap
Outcome surface
Flip points
3. Build and sign the calibration artifact
Execution hash (SHA-256)
—
Signature (§16 eddsa-jcs-2022, ephemeral did:key)
Not signed yet. Signing generates a fresh Ed25519 keypair in this tab only — it is not persisted anywhere and proves "this browser produced this calibration," not a durable identity.
Anchoring links out to anchor.ainumbers.co (a separate site, opened by your click, not fetched by this page) to timestamp the execution hash. If a sweep found several flip points, export the artifact and paste its flip_points hashes into anchor.ainumbers.co's batch tool (anchor_batch) to anchor all of them in one Merkle-batched call.