NEXUS BY BRIER · Fixed-scope deliverable
Calibration evidence your validators can re-verify.
The Model-Risk Calibration Attestation turns a decision log you already have into tamper-evident calibration evidence: per-model Brier and reliability analysis, a Merkle-chained evidence ledger, and a regulatory cover memo mapped to SR 26-2 and EU AI Act Articles 12–13 — all independently recomputable by any party you hand it to.
Mail client not opening? Write to hello@aequara.ai with subject “Calibration Attestation”.
Fixed scope · $2,500–$7,500 per attestation cycle, by log size and model count · no SDK, no integration
Three steps. No integration.
It is a deliverable over data that already exists — a CSV or JSONL export of decisions your systems have logged. We never claim more than the rows you supply support.
You export a decision log
Any CSV or JSONL your systems already produce: a model name, a predicted probability, and the realized outcome per decision. Column names are mapped, not prescribed. No SDK, no integration.
We compute and chain
Per-model Brier, Brier-skill vs a base-rate forecaster, ECE, MCE, AUC, and binned reliability tables — and every evidence row is SHA-256 Merkle-chained in the production ledger convention.
Anyone can re-verify
The deliverable ships with the recipe: one command recomputes the chain and every metric and confirms they match what is recorded. Your validators do not have to trust us — they can check.
Three artifacts, one verification recipe.
Your (model, predicted-probability, observed-outcome) rows, SHA-256 Merkle-chained — any edit, reorder, or deletion of a logged decision breaks the chain and is detectable by an independent party.
View sample →Per-model metrics — Brier, Brier skill, ECE, MCE, AUC, reliability bins — plus the Merkle tip and the full re-verification recipe.
View sample →Human-readable: calibration table, per-model reliability detail, a regulatory cover memo mapping the evidence to SR 26-2 and EU AI Act Articles 12–13, the tamper-evidence recipe, and a calibrated disclaimer.
View sample →What this attests — and what it doesn’t.
It attests your models on your logged decisions. The metrics are computed over the rows you supply — your model names, your stated confidences, your realized outcomes. It is not a blanket certification of a live workload beyond that log.
The chain makes tampering detectable, not impossible. Any edit, reorder, or deletion after the ledger is cut breaks the Merkle chain and is visible to anyone who runs the verifier. Provenance of the log itself remains your representation.
Framework mapping is indicative. SR 26-2 is non-binding guidance and places generative/agentic AI outside its scope; the cover memo addresses quantitative and statistical models and says so. Confirm applicability with counsel.
One log. One cycle. Evidence that survives scrutiny.
Send a note with your organization, the system you want attested, and rough log size. The founder reads the inbox; you’ll get a scoping answer, not a sales sequence.
Request an attestation →Or write to hello@aequara.ai directly.
AEQUARA provides calibration measurement and documentation, not legal, audit, or regulatory advice. Regulatory-framework applicability is indicative and should be confirmed with counsel. Attestations cover supplied decision logs only.