docs: choose rain that changes grip on a lap the seed picks

docs/adr/0103-rain-changes-grip-on-a-lap-the-seed-picks.md

@@ -0,0 +1,111 @@
+# ADR 0103: Rain changes grip on a lap the seed picks
+
+## Status
+
+Accepted (2026-11-08), a design with nothing built. The comparison and its
+evidence are in `docs/mockups/rain-on-a-seeded-lap.html`. Extends ADR 0083's
+picks, ADR 0082's rolls from the seed and ADR 0098's presentation zones. The
+build waits for the playtest ADR 0085 and ADR 0095 are pending, since rain sits
+on top of that tyre tune.
+
+## Context
+
+Nothing changes during a race. Card X6 of
+`docs/mockups/course-excitement.html` asks for rain on a lap the seed picks; pi
+objected that it tangles grip with tyre life and risks unreadable failures in
+an eight-kart pack. Three versions were compared: visual rain only, grip only,
+and grip with a wet tyre fitted in the pit lane.
+
+A throwaway trial of about 400 bot races scaled the grip factors mid-race, with
+bots, wear and the pit call unchanged. It compared levels and is too small to
+tune from:
+
+- Grip alone sends karts to the pit lane. Over five laps on courses with a lane,
+  Ace bots made 1.26 stops a kart after the rain's lap in the dry control and
+  1.56 at the mild level below.
+- A wet tyre would rarely be a choice. Packs already stop 1.2 times a kart over
+  three laps in the dry, so most karts would collect one at a stop they were
+  making anyway. An extra stop is a closer call: the mild level costs about 2.9
+  s a lap on new tyres, a noisy mean of -1.4 to +5.4 s across the courses, against
+  a stop's 8 s, so two wet laps barely repay it and four repay about 11.6 s before
+  traffic. It would also do nothing on the oval, which has no lane, or on Tyres
+  Off, which has no stops, and it would need a second tyre state and a bot call.
+- Heavier grip loss overruns the end cap. Over five laps, karts past it were 39
+  of 512 dry, 59 at the mild level and 104 at a heavy level (steering and
+  acceleration x 0.80, braking x 0.65, wash-out + 0.10 rad).
+
+## Decision
+
+- **Grip only.** There is no wet tyre. A stop still restores life and nothing
+  else (ADR 0082).
+- **Weather is a pick**, Dry or Rain, chosen by the creator with the other picks
+  (ADR 0083), carried with them as one value (ADR 0091). It starts on Dry until it has been played. A kept room or replay
+  with no pick is Dry and hashes as it does today.
+- **The seed picks the lap**, from a hash that draws nothing from the sim's
+  random stream: from lap 2 to the lap before the last, or lap 2 or 3 of a
+  3-lap race, which would otherwise have one lap to choose from.
+- **The whole field gets wet together.** The rain begins on the first tick any
+  kart starts the picked lap. Wetness climbs from 0 to 1 over 20 s and holds to
+  the flag; it never dries. The sim's state holds the tick it began, and the
+  snapshot carries that to the prediction. Learning it a tenth of a second late
+  is 0.5% of the wet, too little to feel.
+- **Wetness scales one row of factors** on top of the tyre tables, in
+  `crates/sim/src/tuning.rs` (ADR 0014), as a starting tune:
+
+  | Effect, fully wet | Factor |
+  | --- | --- |
+  | Steering rate | x 0.92 |
+  | Acceleration | x 0.92 |
+  | Braking | x 0.80 |
+  | Wash-out, added at any life | + 0.05 rad, to at most 0.18 rad |
+
+  The 0.18 rad is the most a hop's airtime lets go, an invariant of ADR 0087 the
+  tuning asserts. The top speed cap, wear rates and mini-turbo charge do not
+  change, so the speed readout and the tyre gauge keep meaning what they mean.
+- **Bots keep today's rules.** Their steering shares the rate, so they slow for
+  it, and the wear they read decides their pit call. Their table of lap costs
+  stays dry.
+- **The look follows the wetness, so every screen gets the same rain:**
+  - Clouds gather over the leader's last quarter of the lap before, changing
+    only the look.
+  - The sky greys through the View's light scale, with no lightning or flashes.
+  - The asphalt darkens with a soft sheen; kerbs, white lines and boost pads keep
+    their colours.
+  - Spray is a low, pale plume at each kart's rear wheels, no higher than their
+    tops and about two kart lengths long, so wings, helmets and liveries stay
+    visible through a pack. Item boxes, the runaway wheel, drones and worn-tyre
+    smoke stay readable.
+  - Streaks stop under roofs through presentation zones, which follow the own
+    kart (ADR 0098), and a viewer who asks for less motion gets fewer. Grip stays
+    wet under a roof, since the sim never reads a zone.
+  - A RAIN plate shows as it arrives, and a rain bed plays under the SFX toggle,
+    levelled by wetness and never pitched.
+
+## Consequences
+
+- Rain works on every course and with Tyres Off, though the oval barely feels it:
+  under 1 s a lap on new tyres in the trial.
+- Pi's objection is partly answered. One row of factors is simpler than a second
+  tyre, and spray has a budget, but worn tyres in the wet still stack. Flat out on
+  full lock a kart slides 0.05 rad on new tyres and 0.15 at the cliff, and the 0.18
+  cap holds near 5% life, under the drift's 0.30. At 0% it slides less than in the
+  dry, about 0.06 against 0.10, since the slower steering loads the tyres less.
+  Whether this reads as rain or as a broken kart on a keyboard is for a playtest.
+- The spray budget comes from a hand-drawn sketch. A staged eight-kart race in the
+  build checks it, and the budget gives if a pack does not read.
+- Five-lap races overrun the end cap more often, 59 karts of 512 against 39 in
+  the trial. The build's sweep reports it, and the maintainer decides whether the
+  row or `RACE_END_CAP_S` gives, as ADR 0085 left a wider cap to them.
+- The trial did not count missed pit lane forks, though a slide of 0.15 rad or
+  more has carried bots past one (ADR 0087) and worn wet tyres reach it. It raced
+  only lap 2 of 3 and lap 3 of 5, at full strength at once, so not the ramp, the
+  other seeded laps or Tyres Off packs. The build's sweep covers all of these.
+- Rain on lap 3 of 3 is a last-lap scramble where a stop cannot pay. On the oval,
+  whose laps take about 14 s, that race ends before the 20 s ramp does.
+- An Ace bot underrates a stop in the wet, since its lap costs are dry.
+- Dry races and existing replay fixtures hash as before; a new fixture covers
+  rain. The pick bumps the protocol version and regenerates the types.
+- The glossary and `docs/style-guide.html` gain their entries with the build, when
+  there is something drawn to describe.
+- Build issues follow this record: the sim and the pick; the look; then the
+  playtest and retune, before Rain can become a default.