Solo playtest · 2026-08-15 · 03:17–04:41 UTC · 1 character · 4 sessions · 659 events, 659 well-formed · one server crash mid-run
1h 23m
PLAYED (4 SESSIONS)
26:10
INTRO CHAIN
5 / 15
CONTRACT KILLS
1
TIMES SUNK
39
SKILL LEVEL-UPS
80 ms
AVG PING
Session milestones
Time from character creation. Three acts again — Intro (light), outfitting (mid), combat (dark) — but this run
is 26:10 to Intro complete vs 6:50 last time, and it carries a red band: the server went silent at 04:06:31 and a new one cold-started at 04:15:45 —
9:14 of dark, mid-voyage to the pirate island. There was no session_end and no server_shutdown drain: it died, it did not stop.
The Navigation-2 level-up at 04:06:31 was earned again at 04:16:13, so ~2 minutes of progress since the last checkpoint went with it.
Sessions
#
START (UTC)
END
LENGTH
HOW IT ENDED
1
03:17:14
03:44:51
27:37
disconnect
2
03:45:04
—
30:41+
server crashed — no session_end, last event 04:06:31
3
04:15:45
04:28:49
13:04
disconnect
4
04:28:57
—
12:18+
still connected at pull
Session 2 has no session_end — the server died under it at 04:06:31 and its length is to-last-event ("+"). Session 4 was still live when this data was pulled (04:41:16), same convention.
Total time in world 1h 23m across a 1h 24m wall-clock span.
Tutorial step pacing
Gap since the previous step in the same chain; the first Battleship bar is measured from the start of session 2 (the chain was picked up after a relog).
Warehouse and first ship are the long poles again (4:46 and 10:30), as is the anvil→cannons stretch on the Battleship side (9:06).
Note the played server was one build behind the current step table: it emitted the older 12-step Intro / 12-step Battleship ids — the explicit anvil-scaffold steps that landed after the last report weren't on this server yet, so the anvil is once more folded into the cannons bar.
New this build: tut_retrieve_provisions closes the Intro chain, and tutbs_equip_hull_armor is now its own step.
The five-launch dock loop
Between 03:41:04 and 03:41:50 the ship was launched 5 times and docked 5 times in 45 seconds — launch, embark, dock, repeat, each round trip 3–10 s
(ship_launched → Embark → ship_docked → Dock). Every one is a real space transition (island → sea → island) with its own space_enter pair.
It reads as the player testing the embark/dock loop, not confusion — but it is also 10 space transitions in under a minute, which is a useful stress datapoint: nothing hitched (12 Hz held) and no event was dropped.
Combat
TIME (UTC)
OPPONENT
DURATION
VOLLEYS F/T
DMG DEALT/TAKEN
OUTCOME
04:16:34
trader T1
16.8s
5 / 5
100 / 72
sunk_them
04:16:46
trader T1
0.7s
0 / 1
0 / 3
sunk_us
04:20:00
pirate T2
24.3s
3 / 3
51 / 63
disengaged
04:23:33
pirate T1 ×2
35.9s
7 / 6
151 / 6
sunk_them
04:37:41
pirate T1
23.1s
5 / 5
125 / 12
sunk_them
04:37:57
pirate T1
12.4s
4 / 4
125 / 9
sunk_them
04:38:38
pirate T1
15.7s
1 / 1
26 / 0
disengaged
04:39:19
pirate T1
20.7s
6 / 5
125 / 39
sunk_them
Eight engagements, eight real fights, no ghosts. Last report's two 1.2 s "engagement-teardown artifacts" (dead ships firing a posthumous broadside — see the incident report) are gone; every row here has volleys on both sides.
✖ Fight 1–2: sunk by a trader. The first two rows are the run's story: the first ship engaged after the crash restart was an AI trader battleship (neutral, T1), not a pirate. The player won that fight
100–72 — leaving the hull on ~3 HP — and 11 s later a second trader battleship fired one volley for 3 damage and sank it. Home-dock rescue fired within 2 s (space_hosted island), the ship was
repaired (RepairShip at 04:17:28) and relaunched 42 s after sinking. That is the first player wreck the analytics have ever recorded, and the rescue path worked end-to-end.
Question for design: should a T1 trader battleship really be a fair fight for a fresh tutorial hull — and should attacking one aggro the whole convoy?
● Fights 4–8: the tutorial fight, properly. Fight 4 was a 2-on-1 against pirates ("×2" — a second merchant joined mid-fight and lapsed out) and still went 151–6.
Kills 3–5 came in a 100-second burst (04:37–04:39) with two dead-hull refusals in between. Then the same wall as last time: 19 FireCannon refusals, 16 of them one-per-second at 04:41:01–16 as the last events of the run — 195 cannonballs crafted, and still out.
5 of 15 contract kills.
The sea is filling up with pirates
Pirate hull count on the hosted sea, per 5-minute census. Traders hold flat at ~50 all run. Pirates climb ~148 every 5 minutes, without bound —
408 → 1591 in the 40 minutes before the crash — reset to 403 when the new server re-hosted the sea, then began climbing again at the same slope (up to 983 by the last census).
With islands_pirate constant at 19 and wreckage constant at 40, that is a spawner that keeps producing pirate hulls that never despawn. The evening playtest earlier today showed the same slope on its sea (358 → 577 in 10 min).
This is the strongest lead for the crash: it died at 312 MB with ~1,600 AI hulls on one sea.
Server memory
Resident memory per minute, both server incarnations. A steady ~1 MB/min climb on each — 243 → 312 MB in the 48:02 before the crash, 254 → 290 MB in the 24:01 after —
tracking the pirate count line above almost exactly. Tick rate held 11.99–12.00 fps throughout and CPU stayed under 16 %, so this is not a load stall; it is unbounded growth.
312 MB is not itself fatal, so the crash cause is not proven from telemetry alone — but the growth is real, and it needs the server log from 01A0035A.
Skill progression
39 level-ups across 1h 23m played. Land skills front-load through 04:00, sea skills take over from the first launch, and LandBattle makes its first appearance
(2 → 3) from two chickens on the resource island at 04:26 — the first creature kills the analytics have seen from a tutorial run. Navigation reached 11.
Output & friction
Crafted (243 items)
Harvested wood ×25, stone ×31 across 14 sessions (287 s swinging). 568 move orders, 87 interactions,
30 warehouse ops. Cannonballs: 195 crafted vs 30 last run — the "craft two sets" step change is visible, and the player went well beyond it.
Refused actions (68)
All server-side can_execute gates. OpenWreckage leads at 31 — clustered right after each kill (8 clicks in 14 s at 04:38:05, 12 in 40 s at 04:39:27) while the ship was still moving/out of range of the wreck;
the eventual OpenWreckage completions all landed 15–30 s later. That is a player mashing on a target that isn't ready yet: worth a clearer "not in range" cue.
Dock ×10 is 9 clicks in 2 s at 04:25:06 arriving at the resource island — same pattern.
Where the run ended
Battleship chain at step tutbs_destroy_pirate ✓ → tutbs_complete_contract pending, kills at 5 of 15, ammo empty, still at sea at pull time.
0 coins moved all run (no coins_delta at all — same as last time; nothing in the tutorial pays yet).
Two chickens, five pirates, one trader, one wreck of our own.
Carried over from last report: the 15-kill contract is still ~3× the ammo a diligent player crafts, even at 195 balls. New this run: a neutral trader battleship is a lethal first opponent for a fresh hull, and the sea is leaking pirate ships until the server falls over.
Final character snapshot
SKILL
LEVEL
Constitution
10
Navigation
8
Crafting
8
Woodcutting
6
Construction
6
Mining
5
LandBattle
3
SeaBattle
3
Smithing
2
Commerce
1
Fishing
1
Cooking
1
Snapshot at the last session_end (04:28:49); session 4's later level-ups (Navigation 11, SeaBattle 6, Crafting 9) are not yet in it. Session 2's snapshot never happened — the crash skipped it.
SERVER HEALTH
Servers
01A0035A (crashed 04:06) → 01A003A1
Tick rate
11.99–12.00 fps throughout
Memory
243 → 312 MB, then 254 → 290 MB (~1 MB/min)
Peak AI hulls
1591 pirate + ~55 trader on one sea
Avg ping
80.4 ms
Events ingested
659 (659 well-formed)
Join → in world
3.5 s cold / 0.24 s warm
Source: Athena atma_analytics.events, workgroup atma-analytics, dt ≥ 2026-08-14 · scanned 107 KB ·
playtest 2f52dd79 · character 6613a893 · 4 play sessions · pulled 04:41:16 UTC while session 4 was live
Same-day evening playtest fe4368fa (3 characters, older build, 21:15–23:52 UTC) is in the same pull but not in this report; its 4.1-min and 10-min sessions still show the pre-fix ghost engagements.
Chart colours are the AtmaSuite chart tokens. Durations derived from event timestamps per the telemetry "derive, don't emit" rule.