A machine decided.Did you tell them?

It grades whether each of your automated decision systems meets six transparency and recourse practices drawn from several automated-decision regimes — system by system. For each one you mark six controls 0/1/2: whether the use of automation is disclosed to the person, whether there’s a meaningful-logic and adverse-reason explanation, whether there’s a human review/appeal with authority to overturn, whether an opt-out or data-correction path is offered, whether the decision’s scope and jurisdictions are mapped, and whether notices and decisions are retained. The six are weighted to a 0–100 score banded DISCLOSED, GAPS, or MATERIAL TRANSPARENCY GAPS; UNDISCLOSED is reserved for the separate no-notice condition, and the portfolio rolls up to ALL TRANSPARENT, CONDITIONS OUTSTANDING, or REVIEW BEFORE CONTINUED USE, a prompt to check whether the missing controls apply and whether continued use is appropriate, not a direction to suspend a system. It grades the deployer’s process, never a person.

Based on the user’s marks, multiple transparency or recourse controls are missing or incomplete. This result does not mean the person received no notice; UNDISCLOSED is reserved for the separate no-notice condition. A score below 50 can reflect weak explanation, records, appeal, opt-out, or correction controls even where notice was given; only the separate disclosure gate returns UNDISCLOSED.

Because of the disclosure gate. A system is forced to UNDISCLOSED regardless of score when the person was never told an automated system made the decision. In the worked example, an insurance premium-tier model scores 74 — its explanation, appeal, opt-out, and scoping are in place — and still reads UNDISCLOSED because the person is never told. The gate worsens only — it never promotes a verdict. A system marked with no human appeal AND no opt-out is flagged separately as a recourse gap: the flag identifies a gap in recourse and does not change the verdict, so the tenant-screening example reads GAPS with the flag. For this scorecard’s operational recourse flag, either a functioning human-review path or an applicable opt-out path counts as evidence that some recourse exists. That is a scoring convention, not a legal rule that one right always substitutes for the other. Applicable law may require a particular path, impose conditions on substitution, or require additional rights.

No — they’re siblings covering different lanes. The AEDT Deployer Compliance Dossier is locked to automated employment decisions and the controls specific to that lane (a qualifying independent bias audit where an applicable rule requires one, candidate notice, NYC/Illinois/California employment rules). This scorecard covers automated decisions everywhere else — credit, insurance, housing, pricing, benefits, education, health care — and grades them against a composite set of transparency and recourse controls drawn from multiple regimes. Run the AEDT dossier for hiring tools and this scorecard for the rest; together they cover your automated decisions about people.

Several regimes address transparency and recourse when an automated system makes a significant decision about a person, on different timetables and with different triggers: the CCPA’s ADMT rules in California (the prescribed pre-use notice, applicable ADMT access and opt-out rights, and a conditional human-appeal mechanism within the opt-out framework), Colorado’s SB 26-189 (notice at the point of interaction, specified plain-language information within 30 days after an adverse consequential decision, correction, and meaningful human review and reconsideration), the GDPR’s Article 22 right around solely-automated decisions, and US adverse-action duties under ECOA/FCRA. Those laws move constantly — Colorado delayed its AI law, then replaced it outright before the original ever took effect. So the engine grades selected practices that recur across those regimes rather than any single statute’s checklist — the CCPA’s ADMT rules and Colorado’s SB 26-189 attach on 1 January 2027, while GDPR Article 22 and US ECOA/FCRA adverse-action duties apply now, to different decisions. A control may therefore read low where nothing in your jurisdiction requires it. The specific regimes and dates are named in the playbooks as planning references. Confirm which apply to you, and every deadline, with counsel.

No. It is a readiness aid that grades your transparency process from your own marks. It connects to nothing, files nothing, makes no decision, renders no compliance ruling, and cannot tell you whether a specific law applies to a specific decision. It is deliberately people-blind — it grades a decision system, never the person the decision is about, and ranks no one. It is not legal advice, a certification, an opinion of counsel, or a safe harbor. Use it to find and close the transparency gaps in your automated decisions, then confirm scope and obligations with qualified counsel.

A runnable zero-dependency Python engine, a workbook that reproduces it exactly (Start Here, Dashboard, and a Disclosure Scorecard with the mark definitions built into each column), a six-system worked example, and two playbooks: a Disclosure Mapping Playbook for inventorying your decision systems and marking honestly, and a Transparency-Remediation Runbook that sets the controls beside several regimes as planning references and walks the disclose / explain / recourse / opt-out steps. Pick the automated decisions that affect people — credit, insurance, housing, pricing, benefits — grade each one, and, where a control is short, close the one the tool names first. Deterministic and offline; one-time purchase, lifetime access, 12 months of updates.