Texas coastal flood days and hourly persistence
HomeRiskIQ research package, 24 October 2026

Purpose
A selected-gauge extension of the published Annual Flood Day Totals study. This public package provides the article, methods, original graphics and reproducible evidence. It is not a property-risk ranking or a claim of continuous street flooding. Open public-report.html for the standalone report. The public-only document builder can generate an editable copy.

Reproduce
Run: python3 reproduce.py
The numerical script uses only the Python standard library, does not access the network and validates the 48 frozen source files before calculations. It writes results.json and publication-table.csv. Run charts.py with matplotlib to rebuild the charts, and build_document.py with python-docx to rebuild the editable report. Render the report and inspect every page after any edits. The graphics are also provided as editable SVG files.

Data provenance
The evidence directory contains raw NOAA responses, a file/hash/size manifest and full request URLs. Source completion times were reconstructed from file modification times and are explicitly labeled as such. They are not independently logged HTTP timestamps. Data window: May 1, 2025 through April 30, 2026, UTC. Metric units; station datum STND. Current NOS minor-threshold snapshot, not proven historical threshold versions.

Quality policy
Keep official flood-day totals separate from the custom strict greater-than threshold calculation. Report all-value and zero-flag counts separately. A clean run requires all-zero hourly flags and exact one-hour timestamp adjacency. Missing or flagged readings break it. Do not interpret below-threshold inferred values as directly measured. All selected above-threshold hourly readings are clean. One above-threshold six-minute value at Rockport carries a flat-tolerance flag despite verified status.

Selection
Initial 12 stations: Charleston SC, Fort Pulaski GA, Mayport FL, Key West FL, Wilmington NC, The Battery NY, Boston MA, San Francisco CA, Charleston OR, Toke Point WA, Galveston Pier21 TX and Grand Isle LA.
Additional 10 after noticing a Texas contrast: Eagle Point TX, Freeport TX, Rockport TX, Port Isabel TX, Panama City FL, Cedar Key FL, St Petersburg FL, Naples FL, Trident Pier FL and Virginia Key FL.
Nineteen succeeded. Freeport and Port Isabel threshold requests returned 404; Naples returned a no-data hourly response. Failed stations are not zeros. Original discovery error strings are retained in discovery-results files. Complete HTTP failure bodies were not preserved for every failed request, so those exclusions cannot be fully reconstructed from the raw archive. Do not interpret the screen as a complete national or Texas sample.
Boston was selected after exploration as an illustrative contrast. The validated shortlist also includes two Florida gauges for transparent replication of the original six-station validation. They are not new Texas findings.

Important boundaries
15 Rockport hourly readings span 14 hours between endpoints, not 15 continuous hours. Gauge values do not establish inundation of any property. Toke Point and Grand Isle show equality/rounded-threshold discrepancies between the custom screen and official totals; do not substitute those custom results for official counts. Plus/minus 1 cm sensitivity is a diagnostic, not a confidence interval. A single year cannot establish a trend, annual expectation or cause.

Package manifest
package-manifest.json contains hashes and sizes for all included files except itself. The archive excludes render QA intermediates and temporary files. Any later edit requires a new manifest/version.
