# Coastal-current research

Spacetime Research · FE-001 / FE-002  
Public derivative prepared 17 September 2026. FE-002 evidence snapshot: **17 September 2026, 15:01:19 UTC**.

Can a recent observation improve a tidal-current prediction—and can the answer be reconstructed from the evidence available when it was made?

These studies analyze public NOAA field observations at **Southampton Shoal Channel LB 6, station s08010, bin 4**, with a nominal depth of 6.71 m. They do not report a Spacetime sensor deployment. Physical depth at each observation time remains unverified. [NOAA station metadata](https://api.tidesandcurrents.noaa.gov/mdapi/prod/webapi/stations/s08010.json?expand=deployments,bins,harcon) · [NOAA observations and prediction API](https://api.tidesandcurrents.noaa.gov/api/prod/).

## FE-001 — A correction that did not improve the baseline

**Question.** Does carrying the latest observed difference from a harmonic prediction forward improve the one-hour current-vector estimate?

**Method.** Compare four fixed approaches: NOAA harmonic prediction, constant bias correction, residual persistence, and persistence. Convert speed and direction to east/north vectors in metres per second; match exact timestamps and the same nominal bin. The primary retrospective comparison uses 22–31 August 2026, a 12-minute assumed observation delay, and 1,654 paired cases. Those ten days were excluded from the local constant-bias fit. Missing observations were not interpolated.

**Result.** The proposed residual-persistence correction increased mean vector error from **0.13330 m/s to 0.14282 m/s: 7.14% worse** than the harmonic baseline. The hypothesis was not supported under this comparison. The retained numerical review independently reproduced the primary estimates from the archived source snapshots.

**Limits.** Only 1,654 of 2,390 scheduled primary comparisons had the required input and target observations. Results therefore describe the paired cases. Historical source availability, harmonic coefficient vintage and upstream training overlap are unknown; the 12-minute delay is an assumption. Row-level quality status and exact physical depth were unavailable. This is a retrospective comparison at one station/bin, not an as-issued forecast test or an operational safety result.

**Decision and reuse.** Retain the harmonic baseline; reject promotion of this correction on this evidence. Preserve the failed hypothesis, exact-time pairing, vector conversion, missingness accounting, archived source receipts and evaluator for subsequent tests.

## FE-002 — Preserve the prediction before its outcome

**Question.** Can a prospective instrument preserve acquired inputs and frozen predictions, then replay only exact later observations with missing evidence explicit?

**Method.** Retain the same four baselines without refitting. Record acquisition time, raw source bytes and checksums; freeze predictions before their target times. Admit the first eligible exact-timestamp observation acquired afterward. Keep subsequent revisions separate and leave missing target observations unresolved.

**Status at the stated snapshot.** **21 targets** had frozen predictions. **14** had eligible exact observations, **4** were awaiting an exact observation and **3** had not yet matured. The available one-, three- and six-hour groups contained **6, 3 and 5 cases** respectively. The pilot remains inconclusive about predictive performance.

**Limits.** Small, unequal samples and incomplete collection prevent method ranking. Acquisition time establishes when this instrument first saw a record, not when NOAA first made it public. Source issue time, coefficient vintage, row-level quality status and physical depth remain unresolved. Local checksums support change detection; they are not independent timestamp attestations.

**Reuse.** The reusable contribution is the receipt, version and exact-target replay contract—including pending, missing and revised evidence. These records support further evaluation; they do not establish an operational current model or a promoted Assure capability.

NOAA supplies the observations and model products. The experiment design, evaluation and conclusions are Spacetime's; the source links do not imply NOAA endorsement.