Four published surfaces that make every AEQUARA verdict inspectable, calibrated, and overridable — and one that is already live with measured data. The Constitution names what we commit to; Track Record proves it; Methodology shows the math; Appeal is your recourse when we miss.
The first AEQUARA Trust surface backed by measurement, not placeholders: a calibration leaderboard for 18 frontier models computed from 13,171 scored predictions — re-rankable by 9 axes (honest confidence, tail risk, cost…) and viewable across a 5-release generational ladder. It’s the empirical engine behind the model panel — and proves Commitment 01 (“calibrated, not theatrical”) with data that needs no user or human-panel input.
The published charter for what evidence we weigh, how calibrated confidence is computed, how panel disagreement is handled, what topics we decline, and what override rights every user holds. 12 sections + Five Principles. Inherits binding force from the ETHICS Charter.
Live · v1.0.0 codified 2026-06-02 Read the Constitution →Per-tool Brier calibration scores updated quarterly. When the model is right, we say so; when it's wrong, we publish the miss. Per-jurisdiction, per-language, per-counterparty-class breakdowns. Public-when-wrong is a binding commitment, not a marketing line.
Draft · 10 tools tracked · first scored cycle: Q3 2026 See the scoreboard →The math behind every verdict: the six-tier evidence ladder (T1 primary statutes through T6 synthetic), how source-confidence and per-claim recency multiply into a final per-source weight, and how panel-routing aggregates votes into a band-thresholded action label.
Draft · engineering substrate See the math →When AEQUARA is wrong and the wrong verdict caused you harm: file an appeal here. Reviewed within 7 business days. If sustained: full refund, public scoreboard publication of the miss (PII redacted), and a free 30-minute consultation with the editorial team. The override-anything button is permanent.
Draft · appeal form vibes-check File an appeal →Every confidence score is real. Computed from Brier methodology against a public scoreboard. "No data" never becomes a fake 50%.
Every verdict exposes the evidence chain: source list, model-panel votes, per-claim confidence. Default-open, not hidden behind a "Why?" click.
Every verdict overridable in one click. No "are you sure?" friction. Override is exactly as prominent as accept-this-verdict.
On topics where we lack substrate or the question is structurally ambiguous: we return NEEDS HUMAN REVIEW. Refusal is the calibrated answer.
When AEQUARA is wrong, we say so publicly. Brier degradation >0.05 = pause the tool, notify users, refund the quarter. Operational, not marketing.
The four surfaces above are the substrate of the Decision Constitution v1.0. If the substrate doesn't match the charter, the charter is theatrical — and theatrical trust signals violate the ETHICS Charter §3 (publication staged).