Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
2026-11-15 12:50:06 -05:00
co-authored by Claude Opus 5.5
parent aa797580a7
commit 557a3e1f11
+59
View File
@@ -1807,3 +1807,62 @@ schedule is no longer behind, so the deferred priority/low issues are workable a
touches the simulation and the protocol, so it needs a soak before it is closed.
3. Then A3 character customization (#43), H10 bot follow-ups (#74), and H8 (#71),
which can only close with a real-GPU profile.
## 2026-11-15 (midday): #72, the packet ship and the lifeline mortar
**Process miss, found this run:** Eric answered #72 on 2026-11-12 ("B is good"), and
three days of runs passed it by. They treated `needs-eric` issues as untouchable until
the label changed, and never read the comments. **From now on, at the start of every
run, read new comments on every open `needs-eric` issue.** A decision from Eric is the
first work of that run.
**What was built (aa79758, closes #72):** the GDD's launch replaces the sky barge.
- `Drop.plan()`: the packet ship sails past offshore, on the first route (up to 32
headings from the "drop" stream, offsets 300 to 465 m from the middle) with no land
within `SHIP_COAST_CLEARANCE` (45 m, probed in eight directions). If none is clear,
it tries the four world sides; seed 38 is the only one in 60 that needed them. The
deck is at `SHIP_DECK_HEIGHT` (4 m).
- The mortar: jump fires along the look yaw; `charge_for_pitch()` maps -10° to 45° to a
charge from 0 to 1, giving 16 to 110 m/s over the ground. Every shot climbs at 75 m/s
(the apex is about 145 m). In `Movement._fall_from_ship()`, a FREEFALL player moving
up is climbing: no steering, no kite. There is no new drop state, so the protocol is
unchanged. `fire_at_end()` aims anyone left aboard at 80% of the way to the middle.
- `Drop.reach(charge, kite_at_top)` mirrors the fall tick by tick. A full charge
carries 782 m on a fall and 987 m with the kite opened at the top; land lies no more
than about 907 m from any route (60 seeds). `aim_charge()` picks a fall when one
reaches, since it is quicker.
- Bots: `NavBot._ride` fires when abeam of the goal, or in easy reach at its own
staggered moment, or just before the route ends (the forced shot aims at the middle,
not at the bot's goal). `_descend` opens the kite only when the goal is beyond the
rest of a fall. The first version, which glided from the top of every arc, took
everyone 45 to 63 s to land; this one takes 30 to 38 s, like the barge's 32 to 37 s.
- Art: `VesselModels.ship()` (two masts with tanned square sails, a salt-white boot-top,
a bowsprit, and the mortar on its bed; mirrored so the mortar faces the island),
`lifeline()`, the figure's `kite` part (figure.py; the `sail` part is gone), and
lifeline cords in `EntityView`. Barge was renamed to ship and wing-sail to gull-kite
everywhere, including the `match_ship` and `client_ship` capture scenes and a new
`match_launch` stage.
- `make match` and `test_scripted_player` moved from seed 29 back to seed 2. On 29 the
scripted player now falls in a landing fight.
**Tests and gates on aa79758:** `make check` 783 tests and smoke ok; `make match` ok
(seed 2, replay hash matches); `make tutorial-check` 12/12; `make client-check` ok (0
corrections); `make bot-matches` ok (20, 4.3% drowned); `make duo-matches` ok;
`make balance` ok (median 13.7 min, rivet repeater 30.7%, sea deaths 7.6%, up from
3.6%; noted on #74).
**Captures, looked at:** `vessels`, `match_ship`, `match_launch`, `match_glide`,
`client_ship`, `client_glide`, `figure_states`. The first `match_launch` showed the
cords running from the masthead and stopping short: `Basis.scaled()` scales world
axes. Fixed with `* Basis.from_scale()`, with a test on both ends of the cord.
**Release:** the v1.0.0 tag and draft predate this change. #64 now recommends cutting
v1.0.1 from `main`, which is Eric's call. Not soaked this run: the protocol didn't
change and `client-check` and the prediction link tests agree, but a `make soak`
before any v1.0.1 is due.
**Next run:**
1. Read new comments on `needs-eric` issues (#64, #65, #66). If Eric says to cut
v1.0.1, bump the version, run `make soak`, `dist-check` and `deb-check`, and prepare
the draft.
2. Otherwise B4 (#18), then A3 (#43) and H10 (#74), starting with the drownings.