Auditing Catastrophe Overflow Assignment

Research question and scope

This protocol asks whether each notice in a catastrophe surge queue reached a named owner, whether possible duplicates were linked rather than lost, and how often assignment changed before the first substantive contact. The unit of analysis is a notice record inside a published event window, not a claim outcome.

The review is administrative. It does not assess coverage, damage cause, reserve adequacy, or whether an adjuster made a good decision. Those questions require claim authority and are outside the evidence a queue export can support.

Population and fixed export

Define the population from a dated export filtered to the carrier event identifier and the published date-of-loss window. Record the export time, source system, included channels, status filters, and the deduplication rule. Include withdrawn notices and candidate duplicates so the denominator is not limited to clean assignments.

Preserve excluded records by reason, such as out-of-window, wrong line of business, or test record. A population that quietly drops the messiest notices will make the queue look faster than it was. The exclusion log is part of the result.

Measures and definitions

Capture the notice timestamp, the first assignment timestamp, the first owner, the number of owner changes, the duplicate-link flag, and the age of records that remain unassigned. Define the time zone, the rounding rule, and the treatment of missing timestamps before looking at the distribution.

A notice with no assignment time is not a zero. It is missing evidence, and it should appear in a separate count. If the system records an assignment to a queue rather than a person, say so; a queue is not a named owner for the purpose of this audit.

Sampling and reviewer agreement

Stratify the sample by channel, event day, geographic zone, duplicate flag, and assignment status. Include records that were reassigned and records that never reached an owner. Record the random seed and every replacement so another reviewer can rebuild the draw.

Two reviewers should classify a small calibration set before the main pass. Preserve first-pass judgments and adjudicate differences with the written definitions. Questions about claim handling are routed to the claims owner rather than resolved to complete a classification.

Analysis and limitations

Report counts, medians, aging bands, missingness, and the share of notices with more than one owner. Inspect the narratives behind long gaps before assigning a cause. A slow first assignment may reflect a system outage, a channel batch, or a queue with no available adjuster, not a failure by one person.

The audit cannot prove that assignment speed improved an outcome. Event conditions, notice quality, staffing, licensing, and system configuration all change during a surge. Use the measures to locate queue friction and to test a process change on a later event, not to rank individuals.

Privacy, reporting, and replication

Use the minimum notice-level fields needed, restrict access to the working file, and replace claim and policy identifiers with a controlled crosswalk. Report aggregate counts and ranges rather than small cells that could identify a policyholder or an employee.

Retain the query, exclusion log, sample seed, codebook, reviewer notes, and calculations. A qualified reviewer should be able to rebuild the sample and arrive at the same counts without access to the original claim system.

Method note

We translated public guidance on disaster assistance, internal control, and privacy into a queue-measurement protocol. No claim files, carrier event data, or adjuster performance records were used.