feat(monetization): host the banner in tabViewBottomAccessory with a GameRoot-owned lease on an iOS 26.1 floor

[wei18]

What

The banner moves from per-screen inline slots into the shared tabViewBottomAccessory, on top of the #1062 banner session model. The shell now drives the accessory with tabViewBottomAccessory(isEnabled:), so no empty capsule is drawn when the ad gate denies. The iOS deployment floor rises to 26.1 for that API (owner decision on #1079, option 1).

The accessory path is what ships. #1029 B-6 PASSED: a real AdMob BannerView renders inside the accessory at non-zero size and survives placement changes. The tab-content-bottom fallback stays reachable and documented, but it is not wired.

Merge of main (post-#1062)

origin/main (728ef19) is merged into the #1024 branch following the #1080 rules.

• Inline slots stay removed. Every conflict hunk takes the feat(monetization): move banner to tabViewBottomAccessory, with tab-bottom fallback #1024 side. The inline Today, Practice and Settings slots do not come back. Main's non-conflicting rewrites of the unused fallback helpers are kept.
• The accessory reads the session. BannerAccessoryView is a plain BannerSlotView(isSuppressed: false, …) that reads \.bannerSession like every other slot. It takes no provider, gate or primer parameters. The ATT primer (C-33) stays on the session's onAdContext hook.
• No MonetizationBootSignal / bootSignal. Zero hits in Packages/.
• No board file edited by hand. Board slots auto-merged clean.
• Snapshots. The four snapshotIPhoneLightBannerVisible fixtures are dropped with the Today slot test that produced them.
• Second merge of main before opening (0a6739d). Main gained test(sudoku): freeze the session clock in the Ready-window test #1092 (Ready-window test clock) and chore(l10n): remove proven-orphan keys and relabel live stale keys in both catalogs #1093 (orphan l10n keys removed). The merge was clean; none of the five files main touched overlap with this branch. The branch adds no localized strings, and none of the 14 keys chore(l10n): remove proven-orphan keys and relabel live stale keys in both catalogs #1093 removed is referenced by branch code.
• Minesweeper Today-tab structure baselines restored (c1d72fa). The first merge combined feat(monetization): move banner to tabViewBottomAccessory, with tab-bottom fallback #1024's MinesweeperTodayTabViewTests.swift with main's view-structure .txt baselines. Main's test version still wrapped the host in a \.bannerSession modifier, so all five assertViewStructure checks failed on line 1 (the root type). The five PNG baselines are identical blobs at 7b423da, 728ef19 and HEAD. The actual structure output matches feat(monetization): move banner to tabViewBottomAccessory, with tab-bottom fallback #1024's .txt byte-for-byte, so those files were restored. This corrects a merge-side mix-up; nothing was re-recorded.

iOS floor 26.1 and isEnabled

• Project.swift: all four app targets iOS: "26.1". macOS stays 26.0.
• Every package .iOS(.v26) becomes .iOS("26.1"). The macOS-only ASCRegisterKit is untouched.
• RootShellView takes bottomAccessoryIsEnabled: Bool. The modifier is still attached unconditionally inside #if os(iOS). isEnabled is the SDK's own display switch, not a conditional attach.
• GameRoot passes bannerSession.isVisible, read from body, so @Observable re-renders the shell when the gate changes.
• Docs: design.md §2.4 as-built, foundations.md, SUBMISSION-CHECKLIST, both v2.6 review-information docs.

Built Debug simulator apps report:

App	MinimumOSVersion	UIDeviceFamily
Sudoku	26.1	1, 2
Minesweeper	26.1	1, 2

A workspace generated before the floor change fails to build with "compiling for iOS 26.0, but module … has a minimum deployment target of iOS 26.1". Run mise run tuist:generate after pulling this branch.

macOS ships no banner (design §2.4.1 option A)

The macOS build contains no tabViewBottomAccessory call, because the only call site is inside #if os(iOS). Checked on the committed tree:

$ git grep -n "tabViewBottomAccessory" 0a6739d3 -- '*.swift'
58 hits, classified:
- 52 are comments or doc comments
- 1 is the API call, inside #if os(iOS):
    Packages/GameShellKit/Sources/GameShellUI/RootShellView.swift
      185:  #if os(iOS)
      186:  .tabViewBottomAccessory(isEnabled: bottomAccessoryIsEnabled) { bottomAccessory() }
      187:  #endif
- 5 are string literals (suite name, Issue.record/#expect messages) in
  iOS-only test code:
    GameShellUITests/RootShellViewBottomAccessoryRenderTests.swift (whole file #if os(iOS))
    GameAppKitTests/BannerAccessoryPinTests.swift (the #if os(iOS) section)

On macOS, makeBottomAccessory() returns EmptyView. BannerAccessoryPinTests (c) pins this, and BannerAccessoryView itself is #if os(iOS). The §2.4.1 options B and C are not implemented.

Tests

Every pin below comes with a named mutation. Each mutation was run and turned the test red, then it was restored.

Pin	What it guarantees	Mutation that turns it red
BannerAccessoryPinTests (a), iOS, real GameRoot	An open gate renders a loaded accessory slot with non-zero height through Observation; dismiss() collapses the capsule again	(a1) BannerAccessoryView.body becomes EmptyView(); (d2) GameRoot passes bottomAccessoryIsEnabled: false
BannerAccessoryPinTests (b), iOS	A denied gate draws no capsule and registers no slot	(b) RootShellView uses isEnabled: true; (d1) GameRoot passes true
BannerAccessoryPinTests (c), macOS	makeBottomAccessory() is structurally EmptyView	Removing the #if os(iOS) fails the macOS build
BannerAccessoryReHostTests T2, iOS	A push+pop re-host on the real GameRoot keeps the same slot id and one request	BannerAccessoryView goes back to a self-owned lease (slot id changes, refresh 1→2)
BannerSlotExternalLeaseTests T1	Two sequential slots sharing one external lease give one register and one refresh	Registration ignores the external lease
BannerAccessoryMissingLeaseTests T4, iOS	No fallback. With no \.bannerAccessoryLease, over a started gate-open session, the hook fires, no slot registers or loads, and refreshCallCount == 0	The nil branch renders the self-owned BannerSlotView (a loaded slot appears and refresh becomes 1)
RootShellViewBottomAccessoryRenderTests, iOS	Sentinel content renders; with isEnabled: false nothing is drawn (container absent, or zero height and empty)	isEnabled: true

Test totals:

Suite	Passed
swift test AppMonetizationKit	181
swift test GameAppKit (macOS)	96
swift test GameShellKit	124
GameAppKitTests on the iOS Simulator	97

Teardown has no pin: it is hygiene, see Known limits.

The four totals above were run on a7ae575, and each run exited 0 with 0 ✘ or "recorded an issue" lines. The two later commits touch none of those packages, so the results carry over:

• c1d72fa restores five Minesweeper .txt baselines.
• 0a6739d merges main, touching the app catalogs, mise-tasks/scan/l10n, and SudokuKit tests only.

Both apps' packages were run on 0a6739d. SudokuKit was run in batches, as the repo requires, because its full suite segfaults on this machine.

Suite or batch	Result
swift test MinesweeperKit (full)	412 passed
SudokuKit --filter SudokuAppCompositionTests	69 passed
SudokuKit --filter SudokuPersistenceTests	24 passed
SudokuKit --filter SudokuUITests, skipping the six settings suites	376 passed
Settings half A: AudioSettingsSectionTests|ReminderSettingsSectionTests|SettingsIAPRowTests	14 passed
Settings half B, each suite on its own: SettingsViewAppStoreRowsTests, SettingsViewWindowSnapshotTests, SettingsViewTests (run 3 times, once with --no-parallel)	4 + 2 + 15 passed, no crash

Running the three half-B suites together in one invocation (--filter "SettingsViewAppStoreRowsTests|SettingsViewTests|SettingsViewWindowSnapshotTests") crashes swiftpm-testing-helper on this machine. The crash is already on main and does not come from this branch:

Tree	Combined runs	Helper crashes	Tests passed before the crash
This branch, 0a6739d	3	3	21, 21, 22
Clean main c2959d0 (throwaway worktree)	3	3	18, 22, 22

It is the known local SudokuKit settings segfault. Running each suite on its own, as in the table above, is the effective verification. Logs: build/evidence-1080/final-tests/merged/settingsB-{split,attribution}/.

One unattributed macOS failure, not reproduced. During the pin rewrite, one full swift test run of GameAppKit ended "96 tests in 22 suites failed after 2.019 seconds with 1 issue". That output was piped through tail -10, so the failing test's name and assertion were never captured. The ten surviving lines are all passing tests. SwiftPM writes no per-test log, so the detail cannot be recovered. It is not attributed to this branch or to main.

Check	Result
Consecutive full-suite reruns	15
Passed	15
Lines matching ✘, recorded an issue or failed	0

The rerun logs are local in build/evidence-1080/flake/gak-run-{1..15}.log.

Simulator evidence

Debug builds, Google's public test banner unit (forced by the LiveAdMobBridge DEBUG init), BGM muted, own simulators only. Screenshots and crops are local under build/evidence-1080/ and are not committed.

iPhone 17 Pro, both apps, on the lease-fix build f929c73 (app code identical at 0a6739d)

Evidence files are local, under build/evidence-1080/final/{sudoku,minesweeper}/. Each check was spot-verified by the Leader.

Check	Sudoku	Minesweeper
E1: loaded banner in the accessory (.expanded)	PASS: creative leading-aligned, 320 pt	PASS: slot (33,734,364×50) pt, ✕ (353,737,44×44)
E2: gate denied (sim has no iCloud, so the gate fails closed)	PASS: 12/12 polls found no slot; the tab bar has no capsule above it. Before #1079 an empty 48 pt capsule was drawn here.	PASS: same result
E3a: isEnabled flips to false with Settings pushed	PASS: ✕ tap, Settings stays pushed, capsule gone in under 2 s	PASS
E3b: isEnabled flips to false with a board presented	PASS: board ✕ tap, fullScreenCover board still presented (pause toggle and grid present), banner gone	PASS
E4: board banner geometry	PASS: slot (16,480,370×50) pt, creative 16→336, ✕ (336,483,44×44) in the gutter, themed ground, no black sliver outside the creative	PASS: slot (16,704,370×50) pt (sits lower, above the tap-mode row), ✕ (336,707,44×44), themed ground

Built apps report MinimumOSVersion 26.1.

Test-fill note. The ads shown are Google's rotating test creatives. Sudoku got the 468×60 letterboxed creative; Minesweeper got a different one with no letterbox.

Minesweeper mirror request probe (iPhone, f929c73)

The accessory is shared GameAppKit code, and the request-count probe had only run on Sudoku. So Minesweeper was run once more. Setup: a throwaway worktree at f929c73 with DEBUG os_log lines at lease attach, lease deinit and ad request (build/evidence-1080/ms-mirror/ms-mirror-instrumentation.patch), and a single continuous log (ms-iphone.log).

Step	Accessory lease-attach	Accessory lease-deinit	ad-request
Cold launch	5A35E8FF…	—	1
ATT "Not now"	—	—	0
Settings push	—	—	0
Pop to Today	—	—	0
Idle 30 s	—	—	0

• Result: 1 request for the whole run, and the accessory lease never deinits (PASS).
• Other deinits are expected: all 6 other lease-deinit ids in the log were never attached. They are the self-owned leases created and discarded on each re-host, as described in the lease note above.

iPad regular width (E5)

iPad Pro 13-inch (M5), iOS 26.5, portrait. iPadOS puts the tab bar at the top, so the accessory sits at the bottom of the tab content. That matches Apple's .expanded definition: "at the bottom of the tab's content view".

App	Gate	Banner slot frame (pt)	✕ frame (pt)	Capsule
Sudoku	open	46, 1306, 940×50	366, 1309, 44×44	drawn, about 33–994 pt wide
Minesweeper	open	46, 1306, 940×50	366, 1309, 44×44	drawn
Sudoku	denied	no slot element	none	none (full-res crop of bottom 180 px)
Minesweeper	denied	no slot element	none	none (full-res crop of bottom 180 px)

In the open case the creative is 320 pt wide, leading-aligned at x = 46, and the ✕ sits in the gutter past its trailing edge. The denied state is the sim's natural one: no iCloud account, so the gate store is unreadable and the gate fails closed.

Test-fill observation, not a defect. On both iPad and iPhone, Sudoku runs received Google's "This is a 468x60 test ad" creative, which shows black letterbox bars above and below inside the 320×50 frame. The iPhone run is build/evidence-1080/remount-probe/ec-head-iphone-reparent-att-1.png. The Minesweeper iPad run received a house creative that fills the frame cleanly. The bars are inside the creative, so the bridge's clear background cannot reach them, and a real fill is needed to judge production. Crops: build/evidence-1080/sudoku-ipad/E5-open-behindprimer-creative-fullres.png, build/evidence-1080/minesweeper-ipad/E5-open-creative-fullres.png.

.inline is not applicable

The shipped shell never enters .inline on iPhone, so there is no .inline evidence (PM ruling on #1080).

• Apple, tabViewBottomAccessory: "On iPhone, the placement of the bottom accessory depends on the tab bar size: when the tab bar is normal size, the accessory appears above it; when the tab bar is collapsed, the accessory displays inline."
• Apple, TabBarMinimizeBehavior.automatic: "On iOS, iPadOS, tvOS, and watchOS, the tab bar does not minimize."

$ git grep -n "tabBarMinimizeBehavior" 0a6739d3 -- Packages
(no hits)

Known limits

• fix(monetization): ATT primer does not gate the first ad request — maybePresentOnAdContext returns before the user decides #1091, the ATT primer does not gate the first ad request. maybePresentOnAdContext() presents the primer and returns without waiting for the user's choice, so the first load goes out while the primer is on screen. This is pre-existing from fix(monetization): restore cold-launch gate resolution + consent-before-request ordering #1062 and is tracked in fix(monetization): ATT primer does not gate the first ad request — maybePresentOnAdContext returns before the user decides #1091. Observation: build/evidence-1080/sudoku-ipad/remount-burst/stack-before.png (creative loaded behind the primer).
• The accessory re-host re-requested ads. Fixed here.

  • Cause. UIKit re-hosts tabViewBottomAccessory's content on sheet dismissal, push and pop. That reset the slot's @StateObject lease, disposed the loaded ad and sent a new request.

  • Measured on the same scripted run:

    Build	Accessory requests
    Unpatched branch	8 on each device
    Main's inline slots	2
    This PR (GameRoot-owned lease)	1, on both iPhone and iPad

  • Evidence. build/evidence-1080/remount-probe/ (E-a / E-c logs).

• fix(monetization): accessory banner creative unpainted ~1–2 s after tab-accessory re-host #1094: creative unpainted after a re-host. The same BannerView survives each re-host with no new request, but its creative area stays empty for about 0.6–2.5 s before painting, measured from simulator video.
  • Recording is ruled out: main's never-re-hosted slot, recorded the same way, stays painted.
  • One layout nudge was tried; its trigger never fired, and it was dropped.
  • This is cosmetic. It is tracked in fix(monetization): accessory banner creative unpainted ~1–2 s after tab-accessory re-host #1094 with per-device, per-transition numbers.
• Teardown hygiene. The app is single-scene (no UIApplicationSceneManifest), so GameRoot, and the accessory lease it owns, lives for the whole process. Scene teardown is process death.

  • Harness result. In a hosted-window diagnostic, releasing the hosting controller leaves the GameRoot-owned lease alive. session.slots still holds that same lease id 5 s later, and nothing is disposed.

  • Evidence. build/evidence-1080/remount-probe/test-ec-teardown-v2.log.

  • Open question. What retains the lease after the host is torn down is not identified. There is no production teardown path for it.

  • No accumulation per re-host. Across each whole E-c script (ec-head-{iphone,ipad}.log), self-owned leases that were never attached deinit in step with their inits. The one left alive on each device is the live host's own lease.

    Device	Never-attached inits	Never-attached deinits	Alive at end	All unique lease ids, init / deinit
    iPhone	23	22	1	25 / 23
    iPad	16	15	1	18 / 16

    Counting rule: the first three columns leave out the two attached leases on each device. One is the accessory's GameRoot-owned lease, which stays alive by design. The other is the board slot's lease, which deinits on leave. Adding them back gives the last column. The evidence index comment attributes each ad request to its slot and explains where each log ends.

    So old hosted content is released on every re-host. The retention seen in the harness is specific to harness teardown.

  • Other slots. The board, Practice and Settings slots keep self-owned leases and deinit normally. Examples: the board slot lease deinits on leave in E-c, and in main S3.

  • Correction. An earlier v1 run reported the lease as deallocated. That run never actually captured the lease, so it is superseded by v2.

Review so far

This is a pre-PR Leader-side code review of the lease-ownership commit only (6d549fe7..f929c734). It is not the PM gate. Two independent reviewers read committed content through git show.

Reviewer	Verdict	Findings
sonnet	APPROVE	1 NIT: BannerSlotBandLayout was moved to its own file. The move is byte-identical apart from private becoming internal, it stays under the 400-line limit, and it is documented in the file header. No change needed.
haiku	APPROVE	All contract items PASS: minimal public surface, existing init untouched, no fallback, dismantle comment only, GameShellKit untouched, no leftover probe code or production IDs.

Neither reviewer re-ran the T1/T2 red mutations. Both were run during implementation, and the evidence is recorded in impl-notes.

Gates

Run on 0a6739d, the final HEAD.

$ mise run scan:admob
WARN: canImport(GoogleMobileAds) outside the bridge (soft dependency):
Packages/AppMonetizationKit/Package.swift:47:            // `#if canImport(GoogleMobileAds)` / `canImport(UserMessagingPlatform)`
Packages/AppMonetizationKit/Sources/AdsAdMob/LiveAdMobAdProvider.swift:35:    #if canImport(GoogleMobileAds)
PASS: AdMob isolation intact (bridge: Packages/AppMonetizationKit/Sources/AdsAdMob/LiveAdMobBridge.swift).

$ mise run scan:hygiene   # prints only on failure
exit 0

$ mise run scan:l10n
L10n gate passed: all 2 catalogs 0 placeholders, 0 stale, all 7 locales complete.

The scan:admob warning is not from this PR. git grep -n "canImport(GoogleMobileAds)" shows the same two non-bridge hits on origin/main.

Cleanup note: stash 1f79ad3f

The tagged stash "1058-default-removal-moot-2026-09-11" only removes the default value of a bootSignal: MonetizationBootSignal parameter and adds that argument at call sites. That type no longer exists.

Check	Result
Stash changed lines	215
Lines naming bootSignal / MonetizationBootSignal	117
Remaining lines	old side of the same calls, plus import MonetizationCore for the type
MonetizationBootSignal|bootSignal in Packages/ at HEAD and origin/main	0 files

The drop is recorded in a comment on this PR.

Closes #1080
part of #1024
part of #1014

🤖 Generated with Claude Code

[wei18]

Stash 1f79ad3f moot check (#1080 step 8)

#1080 step 8 asks for the stash to be dropped only after confirming it is moot. The tagged stash is "1058-default-removal-moot-2026-09-11", made on fix/1058-bannerslot-task-lifecycle.

What it contains. Every change in it removes the default value of a bootSignal: MonetizationBootSignal parameter, or adds that argument at call sites. Its base commit is 85c65cb.

Check	Result
Changed lines in the stash (git stash show -p)	215
Lines naming bootSignal or MonetizationBootSignal	117
The other 98 lines	The old side of the same call sites, plus import MonetizationCore for the type
git grep -c -E 'MonetizationBootSignal|bootSignal' in Packages/ at this PR's HEAD	0 files
Same grep at origin/main	0 files

The type the stash edits was deleted by #1062, so nothing in it can apply. Moot. The stash is dropped right after this comment. The drop command and its result are in the Leader's report to the PM.

[wei18]

Evidence index (#1080)

These are the key numbers from the PR body, with where each measurement lives. Evidence files are local and uncommitted, under build/evidence-1080/ in the #1080 worktree. Every row below was checked by the Leader against the logs, CSVs or crops, not taken from agent reports.

Claim	Number	Evidence
Accessory ad requests before the lease fix, same scripted run	8 on iPhone, 8 on iPad (main's inline slots: 2)	remount-probe/ea-head-{iphone,ipad}.{log,marks}
Accessory ad requests with the GameRoot-owned lease	1 on iPhone, 1 on iPad (cold launch only). Each log's other request is the board slot's; see "E-c request attribution" below	remount-probe/ec-head-{iphone,ipad}.{log,marks}
Minesweeper mirror, same probe	1 request; the accessory lease 5A35E8FF… never deinits	ms-mirror/ms-iphone.log, ms-mirror-instrumentation.patch
Gate closed in-session (h2), 30 s idle	0 requests, external lease never deinits, both devices	remount-probe/h2-{iphone,ipad}.{log,marks}
Re-host churn, per-transition accumulation	Never-attached self-owned lease inits/deinits: iPhone 23/22, iPad 16/15 (1 = live host). All unique lease ids: iPhone 25/23, iPad 18/16. See the counting rule below	remount-probe/ec-head-*.log
Repaint gap after re-host (#1094)	iPad 644–1926 ms, iPhone 824–1341 ms; C1 control on main: 0 empty frames of 144	remount-probe/{ipad,iphone}-gap-*.csv, c1-ipad-main.*
E1–E4, both apps, iPhone	10/10 PASS	final/{sudoku,minesweeper}/E{1,2,3a,3b,4}-*
E5, iPad regular width	Capsule drawn when the gate is open; no capsule when denied; both apps	sudoku-ipad/, minesweeper-ipad/
Practice and Settings wrapper removal (9799f96): before 0a6739d vs after, iPhone, both apps, gate denied and gate open	0 differing pixels in every top, bottom and edge band outside the clock, home-indicator and ad-creative exclusions, except P-top in both apps: about 1,000 px on the Liquid Glass tab-selection pill still settling (P-bottom, the same screen after it settles, is 0). Tab bar and banner slot frames identical	wrapper-ab/{Sudoku,Minesweeper}/compare/*-{top,bottom,edges}.png; raw captures wrapper-ab/*/{before,after}/; residual zoom wrapper-ab/Sudoku/compare/P-top-residual-zoom.png
Settings half-B combined-run crash	3/3 on HEAD, 3/3 on clean main c2959d0, so it predates this PR; each suite alone passes	final-tests/merged/settingsB-{split,attribution}/
Teardown hygiene (single-scene)	Harness retains the lease 5 s after host release (v2)	remount-probe/test-ec-teardown-v2.log
Unattributed macOS failure during the pin rewrite	Not reproduced in 15/15 reruns	flake/gak-run-{1..15}.log

The throwaway probe patches are archived in remount-probe/archive/. That covers the host-owned lease prototype, the open-ad-gate port, the repaint-nudge attempt and the analysis scripts.

E-c request attribution

The probe logs do not tag a request with its slot. Each request below is attributed from three things: the lease-attach line just before it, the step in the .marks file, and whether that lease id later deinits.

iPhone, ec-head-iphone.log

Request (line, time)	Preceding lease-attach	Step	Slot	Lease deinits
20, 19:10:25.366	line 14, C732562E…	step1 cold launch	Accessory	Never
165, 19:14:05.815	line 164, B5811C37…	h1-open-board, tap 19:14:05.026	Board	Yes, line 199 at 19:14:56.142, after h1-leave

iPad, ec-head-ipad.log

Request (line, time)	Preceding lease-attach	Step	Slot	Lease deinits
24, 19:23:43.484	line 14, 812B3C8F…	step1 cold launch	Accessory	Never
104, 19:27:30.685	line 103, D861415F…	h1-open-board, tap 19:27:29.172	Board	Yes, line 131 at 19:28:15.745, after h1-leave

Each device therefore made 1 accessory request, at cold launch, and 1 board request.

Lease counting rule. The "Re-host churn" row counts only self-owned leases that were never attached: iPhone 23 inits / 22 deinits, iPad 16 / 15. The raw unique lease ids in the same files are iPhone 25 / 23 and iPad 18 / 16. The difference is exactly the two attached leases in the tables above. The accessory's lease never deinits, by design. The board's lease deinits once.

End of each log. Both logs end with "Child process terminated with signal 15". That line is the script stopping its own log stream: run-ec.sh kills the stream just before it writes done. The log files were last written at 19:16:00 (iPhone) and 19:29:17 (iPad), 1–2 s before done. The last probe event is at 19:14:56.183 on iPhone and 19:28:15.842 on iPad. These later steps have no log lines:

• h1-after-leave-idle15
• h2-dismiss-ad, marked NOT-FOUND, so nothing was tapped
• h2-idle30
• done

None of these steps navigates, so no probe events are expected. The files alone cannot rule out output lost from a buffer when the stream was killed. No claim in this index depends on those steps. The h2 results come from separate continuous per-device logs, remount-probe/h2-{iphone,ipad}.{log,marks}.

[wei18]

PM merge record (head 9799f960). Gate satisfied. Recording the T6 verdict here because a tooling outage on the PM side blocked posting it when it returned.

• Dual code review at 0a6739d3. Opus (merge resolution + production correctness): no production defect; the two merges forward from the pre-fix(monetization): restore cold-launch gate resolution + consent-before-request ordering #1062 branch resurrected nothing but one stale comment; fix(monetization): restore cold-launch gate resolution + consent-before-request ordering #1062 invariants intact; the lease shape matches the approved conditions; iOS/macOS cold launch, purchase and repoll traced. Findings: 1 BLOCKER (resurrected pre-fix(monetization): restore cold-launch gate resolution + consent-before-request ordering #1062 comment), 1 MAJOR (new_game/scaffold still emitted iOS 26.0), 3 MINOR. Sonnet (suites, mutations, snapshots, mirror, evidence): all 11 suite counts matched, T1 mutation red, only 2 tests deleted with no orphan baselines, the 5 Minesweeper .txt files byte-identical to 7b423daf and their PNGs sha256-identical to main, mirror symmetric, 6 evidence spot-checks matched. Both verdicts MERGE after fixes.
• Fixes 6d3b7401 + 9799f960 per the scope rulings recorded on feat(monetization): #1024 follow-through after #1062 merges — merge main, simulator evidence, open PR #1080: comment, scaffold, doc and visibility fixes; the obsolete fallback helpers and the banner: params they orphaned removed (both Practice hubs, SettingsView, SettingsScreen, PracticeHubShellView, SettingsShellView); the TodayTabHost row dropped from BannerSlotCollapsedHeightTests; both Today comments rewritten; the dead Daily hub chain deferred to chore(ui): remove dead banner ViewBuilder params from the Daily hubs (unused since #1020) #1096. The single-child wrapper removal was accepted only after a before/after simulator smoke of Practice and Settings on both apps came back equivalent (0 differing pixels in every band except a settling Liquid Glass pill; identical tab-bar and slot frames).
• Fresh-context T6 at 9799f960: ACCEPT. It reviewed the post-CR delta item by item (every ruled fix present, nothing else changed, the three Daily hub files untouched) and executed the three iOS-only pin mutations that neither CR could run under the no-simulator rule: T2 re-host (green then red then green; slot id changed + refresh 2), T4 no-fallback (red: slots non-empty + refresh 1), and the GameShellKit isEnabled pin (red: container non-nil, non-zero height). It created and deleted its own simulator, and verified the evidence-index line numbers against the logs byte for byte.
• One documentation gap, not blocking: the body's test-count table has no SettingsKit row, though the delta touches that package. T6 ran it: 51 tests, all pass.

Known limits carried forward: #1091 (the ATT primer does not gate the first request), #1094 (0.6-1.9 s repaint gap after an accessory re-host), and teardown retention seen only in the test harness (hygiene; no per-re-host accumulation: self-owned lease inits/deinits 23/22 iPhone, 16/15 iPad). Follow-up: #1096.