Grand Isle historical comparison v1.0.0
Source records: NOAA CO-OPS; analysis and chart: Louisiana Deep Sea Fishing.
No NOAA endorsement. NOAA source material is public domain; see
https://tidesandcurrents.noaa.gov/disclaimers.html

Question: how did hourly measured height differ from hourly tide predictions
at station 8761724 in calendar year 2025? This is a descriptive historical station
comparison, not a fishing, causal weather, clearance or forecast-skill study.
Predictions are as retrieved, not archived issue-time forecast vintages.

Period: 2025-01-01 00:00 UTC through 2025-12-31 23:00 UTC.
Products: hourly_height (verified observations), predictions (interval=h).
Both: English height units (ft), datum=MLLW, time_zone=gmt.
Station and month fixed before retrieval. Data retrieved October 21, 2026.
Exact URLs, retrieval times and SHA-256 hashes are in provenance.json.
Original source files: observations.json, predictions.json, station.json.

Pair exactly on a 8760-hour UTC grid. Accept numeric pairs only when hourly
observation flags I,L are both 0. Missing/invalid/flagged hours are excluded
with reasons, never filled. Duplicates, malformed times/flags, wrong station
or incompatible request settings fail validation.

For each admitted pair: difference_ft = observed_ft - predicted_ft.
Mean difference = sum of signed differences / number of admitted pairs.
Mean absolute difference = sum of absolute differences / admitted pairs.
Median = middle sorted difference (mean of middle two for even count).
Daily mean = signed differences averaged within that UTC date's valid pairs.
Extremes refer to hourly samples, not unobserved between-hour extremes.
Thousandth-foot formatting is numeric precision, not measurement accuracy.

Standalone reproduction (Node 24, no network needed):
Download reproduce.mjs, protocol.json, observations.json and predictions.json
to one folder. Run node reproduce.mjs there to print the summary. No files are
modified by this standalone calculator. Compare the raw-file hashes with
provenance.json before running.

Reproduction in the repository (Node 24, no network needed):
  node site/scripts/build-noaa-year.mjs
  cd site
  npm test
Chart generation (Python 3, Matplotlib; delivery used Matplotlib 3.11.2):
  python site/scripts/plot-noaa-year.py
The source protocol and hashes are checked before rebuilding outputs.
Raw inputs are retained in research/noaa/grand-isle-2025/.
Source acquisition is separate and refuses to overwrite an existing snapshot.

Results use 8,758 eligible pairs out of 8,760 expected hours (two inferred observations excluded). No catch, fishing-effort, marsh-current,
bottom-depth or meteorological variables were acquired. No confidence interval
or representativeness claim is made. Independent scientific review is pending.
Corrections require a versioned output and recorded explanation; historical
collection dates do not become current conditions when the page is updated.

Monthly and seasonal means use admitted hourly pairs for BOTH observed and
predicted heights. They are hour-weighted, not means of monthly means. Means
are independently rounded, so subtracting printed means may differ by 0.001 ft.
Winter combines January, February and December 2025; it is not a contiguous
December-February winter. Spring = Mar-May; summer = Jun-Aug; fall = Sep-Nov.
These are within-year descriptive groupings, not multiyear seasonal normals.
monthly-summary.csv contains all 12 calendar-month summaries.