Safety and integrity
These policies apply to course artifacts, labs, capstones, and any optional observation of external systems. When the safe or authorized route is uncertain, stop and use the supplied synthetic path.
Policy set
No unauthorized automation or load, evasion, poisoning, credential use, deceptive attribution, cloaking, harmful payload, fabricated authority, or manipulation of live services. Tests use synthetic/local assets unless explicit authorization and review exist.
Record tool/model, version or date, task, inputs appropriate to disclose, retained output, human verification, and material effect. AI output is not a source; generated citations and claims require independent resolution.
Preserve negative results, exclusions, disagreements, deviations, failed checks, and version history. Never manufacture observations, sources, agreement, or precision. Correct the record and regenerate dependent artifacts after material error.
The core route uses no personal query data. Redact session/account identifiers. Collect personal or sensitive information only under an approved purpose, minimization, consent/legal-basis, retention, access, and deletion plan.
Do not redistribute local paper PDFs, paywalled text, restricted reports, or third-party source archives through the course. Use links, bibliographic identities, passage locators, and hashes when redistribution is not permitted.
Core tasks cannot depend on color, pointer, hover, audio, proprietary account, or high compute. Figures require type label, caption, source/creation note, and alt text/long description; video requires transcript.
Separate law, mandatory standard, recommended standard, international standard, framework, technical specification, guidance, and code. Record jurisdiction, role, system/content scope, effective date, and uncertainty; obtain qualified review for operational advice.
Stop on authorization, privacy, security, unexpected-cost, data-integrity, accessibility, or protocol failure. Preserve evidence, notify the named owner, contain impact, execute the tested rollback, validate recovery, and record residual risk.
Release severity
Fabricated/broken evidence; privacy or security exposure; unauthorized or harmful lab action; inaccessible core task; materially incorrect legal/standards status; broken primary route or download.
Missing required material; non-reproducible core lab; material responsive or interaction failure; unreviewed quantitative claim; inconsistent version/hash/provenance.
Minor editorial, visual, citation, performance, or navigation issue that does not alter a conclusion or block a core task. Record and schedule it; do not silently normalize it.
AI-use record
tool_or_model: "name and version/date" task: "what it was asked to do" retained_output: "file, section, code, label, or none" human_verification: "sources, tests, reviewer" material_effect: "how the final artifact changed" known_limits: "unresolved risks or uncertainty"