Skip to content
Dark Cosmos — Dev Docs

Overnight swarm report: 2026-10-14 (1.1 submission candidate)

Overnight swarm report: 2026-10-14 → 15

Section titled “Overnight swarm report: 2026-10-14 → 15”

Runbook: docs/overnight-swarm-runbook-2026-10-14.md. Ran 22:05 → ~00:45 BST, 10-min tick; the queue emptied early, so the handoff ran before the 02:10 horizon. Power: started on battery at 40%, on AC from ~23:00 (80%, not charging). No sleep gaps (every tick landed on time).

Main at handoff: 576e7cd8, full gate PASS 6292/6292 (561 scripts, count guard + boot smoke included), measured on exactly that tree.

0. The uncomfortable part first

Section titled “0. The uncomfortable part first”

The code-side queue is done, but docs/RELEASE-1.1.md §1 (release verification) and §2 (in-flight work) are still almost entirely open, and most of them need you or your devices: HDR/P3 on device, fresh-install stall check, the A17 over-the-top update path, A17 baker check, oldest ATV on the 4.8 fork, your playtest of Supernova / Warden AI / death sequence / Rift Bolt, the ASC privacy tick, and in §2 the alpha-fix batch, HP bar, SFX cull and Supernova bake. A mid-morning submission means either doing the §1 checks this morning (several hours of device time on their own) or consciously submitting without them, and deferring the §2 items to a 1.1.x. The freeze is 10-17, so submitting on the 15th gives up two days you had planned for exactly this.

My recommendation: do the §1 device checks in the order below (the fresh-install stall first; it is the one most likely to fail review), defer §2 explicitly, and submit when §1 is green.

1. Morning submission checklist (in order)

Section titled “1. Morning submission checklist (in order)”

Blockers: the §1 device checks (step 2). Nothing on the code side blocks.

  1. Merges: none left. Everything reviewed is on main (see §3). The only unmerged branch from the run is balance-1.1-proposals, which is a DECISION and remote-config, not part of the binary.
  2. Device checks on a Dev install of main (scripts/dev/install-dev.sh atv, then iphone, ipad; each bumps BUILD on main, fine):
    • Fresh-install first-launch stall (skill bh-fresh-install-stall-check, iPhone + ATV): BUILD 319 showed a 23-25 s first-launch wait on a fresh iPhone vs 0.3 s on store 255. A reviewer on a fresh device sees exactly this. Uninstall wipes the Dev save.
    • A17 iPhone over-the-top update (TestFlight 255 → candidate) and the baked-build draw check (RELEASE-1.1 §1).
    • iPhone haptics: boots, haptics fire, background + return mid-run with haptics on (the CoreHaptics @try fix; native code compiled clean tonight for iOS + macOS, never run on device).
    • Ratings prompt: Dev console review_prompt_test (MCP send_command; the console UI button needs the Worker redeploy in step 5) shows the system review sheet on iPhone.
    • ATV: clean boot, one full run, quit cleanly; oldest ATV if you have it to hand.
    • Settings panel (Droid’s, merged 33cd6b8e): volume, reduce shake & flashes, drone glow, from both the start and pause menus.
  3. Xcode Organizer → Crashes: 1.0.1 (255) iPad around 04:21 UTC 10-14 (the one unexplained real first-play exit); and the 264/307 CoreHaptics crash: if the stack shows __cxa_throw rather than objc_exception_throw, the @try fix does NOT cover it.
  4. Store build: scripts/release/release.sh from a clean main. It does the BUILD bump itself (+ manifest dump, commit, iOS + tvOS build, upload, store-N tag, push); asc.py check-version pre-flights that 1.1 is still open. Short version is already 1.1 in all 5 Apple presets. Mac: scripts/release/build-macos.sh. Skill bh-appstore-release.
  5. Console Worker redeploy from main (control/ changed tonight: review_prompt_test). Needs your OK.
  6. Site deploy: cd ~/Claude/bullet-heaven-site && git merge --ff-only ios-live-copy && scripts/deploy-site.sh (W6, 9bde032, local branch only). It removes the press-kit “Platforms: Mac” row and macOS from the JSON-LD gamePlatform, which are false today. Put them back once Mac 1.1 is approved. Needs your OK.
  7. Submit iOS + tvOS (+ macOS) 1.1 in App Store Connect, What’s New from the three whats_new_1.1_*.txt files (the Settings line is now true).

2. Decisions waiting on you

Section titled “2. Decisions waiting on you”
  1. Submit date (see §0): mid-morning with §1 done, or use the days to 10-17.
  2. Early-run difficulty (branch balance-1.1-proposals 0eb0c01a, doc docs/research/2026-10-14-balance-1.1-proposals.md, patches in remote-config/proposals/, all unpublished). New-player proxy (median bot) survival: shipped 59 s → A1 86 s → A1c 108 s → A2 132 s. A1 is early-only by construction; A1c and A2 also lift the strong bot 108 s → 258 / 278 s, so they ease minutes 2-5 for good players too (Charger + shooter damage are what kill good players there). The better long-term lever is an unbuilt BIOMASS_TARGET run-time ramp (code, two new tunables). These are remote-config patches, so they can ship after the binary.
  3. Orbit/Nova patch B1: Orbit damage share 0.1% → 11.8% (ring radius 120 → 260, hit radius 18 → 40, base damage 13 → 100), Nova 4.8% → 12.3% (area 180 → 280, damage 5.2 → 7). The Orbit ring gets twice as wide (a look change). Not measured: evolved Event Horizon, B1’s effect on survival. Code issues found (not fixed): the Orbit hit test ignores enemy size; armor cuts all DoT to 25%; Orbit/Nova get no damage upgrade in Survival.
  4. First-run Survival hints (Gauntlet II style, teaching Toby’s strategy): paused at your request until you have watched more of his run. The replay viewer is now fixed (mouse no longer takes over playback): tools/replay_viewer, replay .superpowers/replays/toby-2026-10-14-1837537114.json.

3. What shipped to main overnight

Section titled “3. What shipped to main overnight”
Item Commit Notes
W1 first-play crashes 5646d0f3 The “4/9 first plays crashed” audit figure was wrong (1 real unexplained exit). Fixed a CoreHaptics exception abort new in the 1.1 lineage; MetricKit re-poll. Doc docs/research/2026-10-14-first-play-crashes.md.
W2/W5/W8 ratings prompt a67b0d36 DcStoreReview (StoreKit) in the DcHaptics GDExtension. Asks once per build on return to the start menu after a first boss kill or a new Survival best ≥120 s; never on death or a quit run. iOS only for now: Mac is a no-op until the addon is in the macOS preset. Two reviews (R5 MERGE + round-2 LOWs fixed).
W4/W7 replay picks 133d4d2a, 576e7cd8 Human replays now record and replay level-up + dimension picks; playback never hands over to a human; tools/replay_viewer; Toby-vs-autopilot doc docs/research/2026-10-14-toby-replay-vs-autopilot.md. W7 found the post-merge test failure was a seed-dependent flake on the branch, not main’s changes.
Droid: qol-settings 33cd6b8e Merged by Droid at ~00:30. R7 had reviewed the conflict resolution read-only beforehand: MERGE, gate 6246/6246, nothing lost from main.
Droid: skip-tutorial, shop-levels, store-copy (before the run) Already on main at dispatch.
RELEASE-1.1 ticks this commit Ratings DONE; crash item points at the merge; difficulty + Orbit/Nova point at the DECISION branch.

Not on main: balance-1.1-proposals (DECISION), site ios-live-copy (deploy needs OK).

4. Findings filed

Section titled “4. Findings filed”
  • Real store install base for 1.0.1 (255) is ~6 devices, zero day-2 returns, zero tvOS store installs (the audit’s n = 9 included TestFlight / review / our own devices).
  • Fork shutdown crash on 315/322: still open (research doc above).
  • Two LOWs on Droid’s Settings panel: a stale “no reduced-motion setting” comment in render/death_sequence.gd:35; _sim_freeze_reason never says “settings” (harmless).
  • replay-picks follow-ups (LOW/NIT): a human click on a panel during playback still picks; ReplayDriver’s dimension fallback differs from main’s; sim/constants.gd comment placement; viewer not yet in docs/reference/running-and-testing.md.

5. State of main (measured)

Section titled “5. State of main (measured)”

576e7cd8: scripts/gate.sh PASS, 6292/6292 tests, 561 scripts, determinism baselines unchanged, boot smoke clean. DcHaptics native addon (iOS + macOS) compiled clean from main tonight.

6. Process notes

Section titled “6. Process notes”
  • No permission stalls, no sleep gaps; power moved to AC early.
  • Main moved under two merges (ratings-prompt, then Droid’s qol-settings); each was re-merged and re-gated, not pushed on a stale gate.
  • Droid merged its own branch mid-run, which made the planned morning decision moot. Two orchestrators merging to main overnight worked here only because both re-gated; worth one owner next time.