AEP Reporting

Artifacts

Keep AEP outputs separate from the UHQS scorecard:

Artifact	Role
SCORECARD.json / report	Normative UHQS (unchanged)
advanced-evidence.json	Machine-verifiable AEP addendum
ADVANCED-EVIDENCE.md	Analyst-readable methodology, results, limitations

Schemas:

• aep-experiment.schema.json
• aep-trial.schema.json
• advanced-evidence.schema.json

Interpretation wording

Preferred:

AEP evidence observed … under the declared controlled conditions.

Forbidden:

UHBS-certified attacker engagement / proven against all attackers / AEP grade replaces UHQS.

Always state uhqs_unchanged: true.

Status examples

Positive (valid)

Primary outcome computed with adequate samples and passing evaluator control. Example: DTDR ≈ 2.2 with n≥5 per arm and control_status=passed.

Null / weak effect (valid with interval crossing indifference)

Metric computed, but confidence interval includes no practical effect. Still valid if design requirements were met — report the interval honestly.

Inconclusive

Too few trials, undefined KM median under heavy censoring, or missing detector rows for FSV layers.

Control failed (control_failed)

Evaluator control trials mark evaluator_control_passed: false. Capability-linked metrics are downgraded; do not claim attacker-capability conclusions.

Banner (required in Markdown)

Every Markdown addendum includes an informative-only, lab/sandbox banner generated by uhbs aep report, plus a short academic-credit block (Zhu 2019; Collins et al. 2024; Ersok et al. 2022; Li et al. 2020) with the statement that citation does not imply endorsement. Full ledger: Research foundations & credits.