v0.12.1 — the water drop you could never see

Download for macOS · Download for Windows — assets below. Neither build is
code-signed, so your OS will complain the first time; see the README for the two clicks.

The bucket drop had no visual at all

Releasing 900 kg of water used to change the simulation and nothing else. The fire went
out, the score updated, and the only evidence the pilot got was a line of text. The shipped
screenshots have shown a bucket hanging over a flame front with no water anywhere for three
releases, because the game could not draw it.

There are now three events, because a drop is three events and they happen at different
times: the column leaving the bucket, the sheet bursting on the ground, and steam coming off
wet fuel for several seconds after. The delay between the first two is the real fall time,
t = (√(v₀² + 2gh) − v₀)/g — about 1.2 s from 12 m and 2.7 s from 45 m — so the height you
released from reads in the picture. Steam appears only where the simulation says fuel was
actually wetted, so what you see and what you scored cannot disagree.

Fixed

• A slung load hung from a slack, curved rope while you were in the air. agl() is
  altitude − ground_alt, and set_spawn plants the airframe at ground + 0.6 — so it
  returns 0.6 whatever you pass it. The sling believed the ground was 0.6 m under the
  helicopter and rested the bucket on that phantom surface. It now asks the terrain for the
  ground under the load, which is also correct on a slope or a shoreline.
• The game said "NO WATER HERE" while you hovered over a river. Refusing the fill is
  right — rivers here run 0.30 m deep at the median against a 1.6 m requirement — but
  refusing and lying are different things. It now refuses with the true depth, and a
  genuinely deep pool fills the bucket on its merits.
• A rattle over the rivers that grew as you descended. The ambience bed modulated 70%
  deep at 8–20 Hz, which is the band where hearing reads periodicity as roughness. Moving
  water has no periodic component; the envelope is now a slow random walk.
• Rock on the near banks rendered 1.2–1.3× brighter than the same stone beside it: the
  triplanar half of the terrain shader was the only albedo not paying the macro-variation
  darkening.
• A camera told to point at "the fire" pointed at burnt-out scar — the fire's head POI was
  publishing the ignition point.

Known and open

The tree LOD hand-off at ~820 m is measured but not fixed: the near tier and the far tier
draw visibly different forests and which one is wrong is still undecided.
ROTORWASH_NOIMPOSTORS=1 shows the comparison.

Full detail in CHANGELOG.md.