Reproducibility standard for self-storage research

A result is reproducible only when another investigator can identify the same estimand, release, observations, transformations, exclusions, weights, and uncertainty procedure. A screenshot or a current facility-page price is not a substitute for a time-stamped observation because asking rents and availability change.

The current program publishes its design before publishing a number. No numeric series is available to reproduce yet. This standard describes the evidence package required when a qualified release exists.

• Canonical release URL, release identifier, version, timestamps, status, and file checksums.
• A precise estimand and separate definitions for recurring rent, promotions, fees, and fixed-horizon effective rent.
• Observation window, source classes, access method, geographic frame, and collection coverage.
• Parser and normalization versions, unit matching keys, deduplication logic, and all exclusion reasons.
• Facility, operator, state, and matched-panel counts plus missingness and concentration diagnostics.
• Index formula, weight construction, composition adjustment, bootstrap design, confidence interval, and sensitivity results.
• Suppressed estimates and failed publication gates, not only the geographies that passed.
• Rights statement explaining which aggregate files can be shared and which source records cannot.

Each retained observation receives a stable identifier derived from normalized identifying fields and a record hash. Collection time is stored in UTC. The audit trail links a published aggregate back to its release, method version, parser version, source URL class, inclusion decision, and transformation history without implying that restricted raw third-party material can be redistributed.

Random procedures, including facility-clustered bootstrap intervals, must declare their pseudorandom seed and iteration count. Time zones, rounding rules, missing-value handling, software runtime, and executable version are fixed in the release manifest.

A published release is never silently overwritten. A correction produces a new version, new checksums, an explanation of the change, and a link to the prior release. Prior files remain identifiable and are marked as current, superseded, or withdrawn. Method changes that alter the estimand or historical comparability receive a new methodology version.

Read the version and revision policy, data dictionary, and publication gates for the operational definitions.

• Public web offers may expire or change after their recorded observation time.
• Robots instructions, source terms, authentication requirements, and page structures can change.
• Third-party raw records may be unavailable for redistribution even when aggregate results can be published.
• An independently reacquired sample can test the method but may not recreate an earlier market snapshot exactly.
• Passing a statistical gate supports publication; it does not eliminate sampling, measurement, or coverage error.