the campaign · months ahead
To plan an installation, a decommissioning or a run of offloadings: how many days N operations take from a given month, which months open a window and how much DNV's α closes — counted over 32 years of past sea, nothing fitted.
area
Hs limit (OPLIM)
window (TR)
how many operations
starting on the 1st of
Every operation start, every 3 h, over 32 years of hindcast (1993–2024): a window of TR hours with Hs under the limit fits, or it does not. Exact fractions of integers, nothing fitted.
the scoreboard · the forecast against the satellite
Every band is committed before the sea happens and scored afterwards against what a satellite altimeter measured. A wrong forecast stays in the record for good.
what is already measured
coverage of the deciding band
96.5%
over 21,259 satellite comparisons it had not seen (it claims 90%)
CLEARED that the sea broke
0.33%
21 of 6,368 with Hs ≤ 2 m; ECMWF's band alone: 1.10%
Measured in hindsight: the bands calibrated before 2025-07 and judged only after it (certs/janela-providers-eval.json). Below, the forward record, which nobody can rewrite.
the admission rule
One trial per target day: a proposer's bands scored for the same day share the same satellite passes and are not independent, so one alone decides the day, drawn by sha256(proposer|day) — the draw does not look at the result. The exact binomial tail P[X ≤ k], X ~ Binomial(m, claim), is read only at m = 30, 60, 120, 240… days, against 1/40, 1/80, 1/160…: summed, 1/20 over the proposer's whole life, not 1/20 per day. The first reading under the bar prunes this version for good; only a recalibrated version, with a new record, comes back. Days are treated as independent: a declared convention, not a theorem. Covering more than it says does not prune; the Winkler score, exact, charges a band that is too wide. Rule fixed on 2026-10-06, before the first scoring. Admission is lost by record, never by opinion.
why these numbers stand as they do
- computed — a floating-point evaluation (certs/janela-providers-eval.json); the decisions are exact
why trust it
The verdict is exact arithmetic on the limit as printed; the band is a claim the scoreboard audits in public. The published band is rounded outward (Hs to 0.001 m, wind to 0.01 kn): a CLEARED or BLOCKED on it holds on the exact band; only UNDECIDED can grow. It is not an approval of an operation: it is evidence a surveyor can redo. The method ↗ · the deck (PDF)
Over 21,259 satellite comparisons no band had seen (since 2025-07): when the deciding band put Hs under 2 m, the sea went over the limit 0.33% of the time (21 of 6,368); with ECMWF's band alone, 1.10%.
what each verdict rests on
- validated
The Hs band, at the measured sites calibrated before 2025-07 and judged over 15 months it had not seen: the ECMWF ∪ NOAA union covered 96.5% of 21,259 satellite passes (it claims 9 in 10); of the CLEARED with Hs ≤ 2 m, the sea broke 0.33% (21 of 6,368).
- assumed
The wind band measured on the same 45,987 pairs (the altimeter's wind); it claims 9 in 10 by theorem, if the next error resembles the archive's. That was not tested outside the calibration: the scoreboard and the check of the decisions look only at Hs.
- assumed
The band carried to a unit or a terminal each unit uses the band of the nearest measured site, up to 350 km away; the forecast error there is assumed to hold here, and nothing measured it at the unit's node. On the card's chart, the halo.
- assumed
The α of Table 4-1 (DNV) adopted from DNV-OS-H101, calibrated in the North Sea; the 1-in-10,000 tail behind it is extrapolated: three years of satellite do not observe it.
- verified
The site α method reproduces Table 4-1 in 16 of 20 cells (±0.02) before looking at Brazil; the α estimated with it carries the same extrapolated tail.
The arithmetic is exact under every criterion; each decision is worth the rung of its weakest premise. By the band, at a measured site and with an Hs-only limit: validated. With a wind limit, at a unit, at a terminal or by α: assumed.
decided
exact arithmetic backs it
computed
float, or a number someone printed; nothing decided it
—
Forecast: ECMWF open data (CC BY 4.0) and NOAA WAVEWATCH III (public domain) · current: Generated using E.U. Copernicus Marine Service Information, doi:10.48670/moi-00016 · altimetry NOAA RADS · hindcast Ifremer WW3 (CC BY-SA 4.0) · units, fields and pre-salt: ANP — GeoMaps · coast: Natural Earth · code: apps/janela