Documentation.

Start with Skua Strait. It is the same desk the product uses.

1. Start on CAP. Press Play. Watch four aircraft leave the deck. Jump to 09:00 when Kite 2 is lost.
2. Switch Stories to Strike. The same three packs reconstruct that sortie. One aircraft is lost. The surface group stays intact.
3. Pause. Drag Kite 1 to a new station. Red ships update from that mark. Coastal batteries stay. Ships stay on water.
4. Use the command bar to assign a path or target, draw a barrier, or stamp destroyed / shutdown / immobilized.
5. Press W to set sunny, cloudy, overcast, rain, or night. Press F to sit in the selected class. Esc returns to satellite.
6. Save a run and toggle Ghost to compare it at low opacity. On the public desk that run stays in the tab only.
7. Open Red readings. Switch readings; ghosts show the other interpretation. Open “Why this number” to see the observations and the formula behind the percentage.
8. Open Next plans. Toggle Point and Distribution. COA 1 is the preferred stationing. COA 2 assumes pairing is blocked. COA 3 is later work and is scored as no change to the next launch.
9. Approve to stop the deck clock. Export or press E to download a commander brief and an A-GRA-shaped pack. Reject keeps the clock running.
10. Load the near-A-GRA log or the platform JSON. The same sortie reconstructs to the same mission hash. Click a shape under Field map, or the Normalize stage, for the full report.
11. Open the Select stage for the algorithm selection matrix and the works it cites.

Deep links

The desk reads its first state from the URL so a screen can be shared or captured: ?story=SKUA-STK-01&pack=platform&t=540&stage=select&tab=plans&view=mc&reading=H-B&coa=COA-2&evidence=1&paused=1&norm=near-agra.

Normalization report

Each ingest shape is mapped onto one canonical record of 19 fields. Every field is marked present (carried under its canonical meaning), mapped (a declared rename, vocabulary, or unit conversion), guessed (no ICD; the desk inferred the meaning and says so), or missing (filled from the frozen catalog). The report lists the source of each field, the unit conversions applied with their factors and counts, and how many timestamps were re-based. The score is (present + mapped + ½ guessed) / 19.

mission.id · mission.theater · mission.objective · mission.duration · mission.capturedAt · unit.id · unit.name · unit.side · unit.kind · unit.class · unit.status · unit.lostAt · fix.t · fix.x · fix.z · fix.alt · fix.hdg · route.tasked · events

Tracks are rounded to a canonical precision (t to 0.01 s, x and z to 1 m, altitude to 10 m, heading to 0.1°) so three shapes of the same sortie hash to the same mission hash, SHA-256 over the canonical record.

Red reading confidence

A reading lists observations on the pack: supporting and contrary. Each is a predicate over track ids, events, and times, and shows the numbers it was computed from. Confidence is the Beta-binomial posterior mean, (F + 2) / (F + A + 4), where F counts supporting observations that hold and A counts contrary observations that hold. A reading with no evidence sits at 0.50.

Monte Carlo parameters

Frozen in the pack: n = 2000 sampled worlds, seed = 20260911, mulberry32 generator, Gaussians as the sum of twelve uniforms minus six so every engine produces the same bits. Per world: the red reading is drawn by its normalized confidence; mobile red poses are jittered by 60 s of class speed; red attack ranges are scaled by 1 ± 0.15 (Gaussian, clipped to 0.6–1.4); blue stations are jittered by 0.15 km and blue weapon range by 1 ± 0.1 (clipped to 0.7–1.3). The baseline and every course are scored in the same world. Reported per course: p10 / p50 / p90 and mean of survivability, effect, and utility; P(best), the share of worlds in which the course has the highest utility (ties split); P(improve), the share in which it beats the baseline; and which other courses its p10–p90 intervals overlap. Utility is the pack's declared weighted sum of survivability and effect. The run hash, mcHash, is SHA-256 over the rounded summary; same seed, same hash.

Health

/api/plot/health returns the counts the benchmarks page is built from.

The public geometry is original. It is not an imported weapon model and it is not operational Navy data.

Open the demonstration · Measured checks · Packs and surfaces