STR refrigerator outages: a food-safety handoff study

Methodology

1. Scope and preregistration: Review one occupied-stay report that a refrigerator is warm, without power, alarming, leaking, or not maintaining its expected condition. Freeze the property-local observation window, inclusion rules, cutoff, and codebook before reading outcomes. Include only traceable cases tied to a property and operational consequence. Exclude drills, demonstrations, duplicate notifications, unattributed recollections, and test-device events. This is a retrospective descriptive review, not an experiment, inspection, legal analysis, clinical assessment, or claim about STRVirtualAssistant performance. The research question is: Which timestamps and authority boundaries prevent a refrigerator outage report from becoming an unsupported statement that guest food is safe?
2. Evidence model: Capture guest wording, appliance, discovery time, last known normal state, power state, door-opening history if known, displayed and measured temperatures with source and time, food category as voluntarily reported, public guidance, troubleshooting boundary, maintenance contact, repair finding, decision owner, guest update, and closure. Preserve originals in approved systems and store controlled references in the study table. For each value, record source role, original time and timezone, retrieval time, correction history, and interpretation authority. Keep guest-reported, staff-observed, device-indicated, official-source-published, vendor-stated, manager-approved, contradicted, stale, and unknown as different codes. A sent message is not acknowledgement, a scheduled visit is not arrival, a vendor completion note is not independent verification, and silence is not evidence that a condition ended.
3. Source selection: Refrigeration and Food Safety, Food Safety During Power Outages, Food Safety in a Disaster or Emergency, Emergency Alerts provide the closest primary or authoritative hazard context. Federal Plain Language Guidelines, Records Management Guidance, Privacy Framework inform communication, privacy, security, or record control. Every listed source was checked on 2026-09-18. None studied this local population, assessed this company, or validated the workflow. A source may define a term or state general guidance; only traceable property artifacts support a statement about a reviewed case.
4. Sampling and comparison: Build a consecutive register for the frozen period rather than selecting memorable examples. Report counts for screened, included, excluded, duplicate, incomplete, conflicting, corrected, escalated, reopened, and unresolved cases. Compare complete event sequences with records where an initial label or final task status stands in for evidence. Stratify by property, occupancy, channel, reporter role, device involvement, after-hours status, and consequence before pooling. Show every denominator and do not combine categories with different decision rights.
5. Reliability: Two reviewers independently code an ordinary-and-exception sample without seeing each other's classifications. Log disagreement by field, identify whether wording, missing evidence, timestamp conversion, or authority caused it, and adjudicate against the codebook. After any rule change, recode a stable sample and retain both results. Agreement measures reproducibility only. It does not prove an observation true, an action safe, a source complete, or an outcome satisfactory.
6. Counterexample search: Look actively for a normal display after power returns treated as proof every stored item remained safe. Also look for a warm-food report dismissed because the appliance dashboard is online. These cases test whether proposed fields preserve a decision boundary or merely reward complete paperwork. Keep negative cases, late corrections, and outcomes that contradict the first interpretation. A later benign result cannot erase an earlier unsupported claim, and an adverse result cannot prove that every earlier action caused it.
7. Sensitivity checks: Repeat tables after excluding the largest property group, automation-originated alerts, records without final acknowledgement, and cases reconstructed later. Vary stale-information cutoffs and complete-handoff definitions. If a pattern changes, disclose that dependency and narrow the conclusion. Treat missingness as a result because records may be systematically absent during severe, overnight, or cross-channel events.
8. Authority boundary: The assistant may provide approved public guidance without tailoring medical advice, record unknown time and temperature, coordinate service, and route commercial decisions. It cannot inspect food remotely, decide whether an item is safe, diagnose the appliance, or approve reimbursement. When immediate danger is possible, the current emergency procedure takes priority over research collection. Researchers do not delay response to complete fields. The review records which approved path was used, what remained unknown, and who accepted the next decision. Any proposed change needs an owner, exception path, review date, and retirement condition rather than turning an association into an automatic rule.
9. Limitations: Records may not reveal initial temperature, outage duration, door openings, thermometer accuracy, food type, or preparation history; later cooling cannot reconstruct the warmest condition. Selection favors cases with retained digital artifacts. Properties, equipment, vendors, local rules, channel terms, guest behavior, staffing, connectivity, and severity differ. The study cannot estimate prevalence from a convenience sample, compare companies, calculate avoided harm, attribute causality, or establish a benchmark. Results apply only to included records, definitions, and cutoff and must change if later evidence changes a case.