MichaelScottsman--LiquidAss-Frame/docs/phase2/wp/C2a-review-R2.md

C2a Home, folders, What's New: independent review, round R2

Reviewer: an independent adversarial review agent (sessions 2 and 3 of this review; session 1 ran 10:49-12:28 and was cut off). Date: 2026-10-07. Steam build 11094443. Scope: package C2a as its log (docs/phase2/wp/C2a.md) reports it: "READY: wp.c2a" with wp.p3 and wp.c1a, M3 reached, native "written and runs". Nothing in the builder's log or in session 1's notes was taken on trust: every finding below was re-derived from the code and re-measured live in session 2 or 3, unless it says "static".

Status

Done (session 4, 16:20-16:30, a short re-check at the user's request to go faster; see §1.4). Nothing that C2a owns has changed since session 2. The tree's 41-home.js is 09:25, 41-home.css 08:34 and 41-home.json 08:25, and the Frame's copies have the same md5 (checked 16:22). 41-search-apps.js still does not exist. After 16:05 only P10's lab code (lab_gates.js, contracts/lab.md) and other agents' logs changed. A static spot-check shows M3, M4, M5, M6, M7, M8 and M12 still in the code. A live run with P10's newest gate code (CSS-only, --flags wp.c2a,wp.p3,wp.c1a) gives the same results: Home at rest passes all four gates in both modes, and M11 still fails. The verdict, findings and counts below are unchanged. Device: both steps undid their flags. At 16:27 the flags file held another agent's pad-bfs (pid 416535), not this review's. No room frame was taken and no native session was run.

Done (session 3, 15:15-16:10). Session 3 re-checked everything that changed after session 2 (§1.3): P3's attention fix for M2 (14:42), P10's pad-bfs recovery (14:15) and session-5 gate code (14:51), P8's answer to REQ #17, and PLAN R2-15's label rule. C2a's own files are unchanged since session 2 (41-home.js 09:25, 41-home.css 08:34, 41-home.json 08:25; the Frame's copies have the same md5, checked 15:16). Native code is unchanged since session 2's native runs (nothing under device/, native/ or the reporter changed after 13:19 except 05-attention.js), so no third native-session was needed. Every session-3 step was CSS-only (native=off on its step line; a step that overlapped another agent's native session was re-run) and used --flags wp.c2a,wp.p3,wp.c1a. No room frame was taken in session 3. Device: my steps undid their own options; at 16:09 another agent's step and native session were running (their wp.p3=false in the flags file is theirs, pid 408743, not this review's).

Done (session 2, 13:00-13:57). All planned CSS-only checks, two native-session runs (P7 is complete) and the static review are finished.

• Deployed files on the Frame are byte-identical to the tree (md5 41-home.js 8e0aae98…, 41-home.css 5d9990f3…, 41-home.json 8c8a0faa…, checked 13:01).
• Device left as found: theme on, CSS only (my sessions ended "back to CSS only: yes"), no flag of mine left on (every step used --flags; the session flags file holds other agents' c2bToggle, wp.c1a, wp.c1b, wp.c2b, wp.p3, which were there before this review), menus closed by the lab, pointer parked by the lab. All five hv --look frames were viewed and deleted at once on the PC; the lab deleted the Frame copies; no look folder of this review is left.
• Evidence files: shots/p2_c2a_r2_*.png, shots/p2_motion_c2a_r2_*, shots/p2_cmp_c2a_home_t1_r2*.png; raw outputs and scripts in the reviewer's scratch (…/scratchpad/c2arev/).

Verdict

fix. Home's skeleton is right and much of it is good: the 4-5-4 honeycomb with real art, the segmented control, the explicit neighbours with one-step memory, LB/RB and D-pad page turns, Steam's own tile menu, X = Play through the action logger, folders and What's New, the footer handling, removal, idle cost and the strings fallback all pass live, and the CSS-only gates pass in the states the builder tested. But the headline state (the open card) is visibly layered over what lies beneath it in both tiers, row-3 name plates are cut by the window edge in the shipping layout, two folders show broken-image icons, the native fragment fails G-DEPTH (R1 and R2) on routes the package owns, grid moves are silent, G-PAD (now measurable) fails on the folder route, and the card build list and the M3 evidence are incomplete. 0 blockers, 14 majors, 16 minors.

Session 3 changes: M2 is resolved by P3's platform fix (re-measured live in both modes, §1.3), so it no longer counts; M15 is new (G-PAD on the folder route, now that pad-bfs recovers after exits). What's New passes G-PAD by parity with Steam's own Home, B at the Home root goes back in history in both modes, Home's labels meet PLAN R2-15 (no glyph cut), and every gate re-run with P10's session-5 code gives the session-2 verdicts (M11 still fails).

1. What was run (session 2)

All steps through python glass.py … with --flags wp.c2a,wp.p3,wp.c1a and --mode laser|pad unless stated. "CSS-only" means the step line said native=off; a CSS step that overlapped another agent's native session was re-run (the batch runner waited for native off).

Id	Command	Result
S2-P1	js probe1.js --mode pad	Installs in 0.9 s; windowless, route home, low layout (C1a's capsule: data-lgs-search="capsule"); 13 cells (12 games + All Games); segments 140 / 172 / 140 × 80 contiguous at y 94-174, track (412, 102) 456 × 64; What's New (973, 24) 203 × 60; Back box (14, 14, 80, 80); dots centred on x 640 at y 700; 17 plates; mosaic bands 14-172, 170-302, 350-482, 530-662; footer hidden (no #Footer). Cards on cells 1, 4, 5, 6, 10 (rects in M9). Collections: Great On Frame, Ready To Play, Installed, Non-Steam, Soundtracks, VR, Empty Collections. Apps page 1 A-Z from Camera Switch
S2-A1	js probe2_pad.js --mode pad, js probe3_laser.js --mode laser	Re-entry bug (M2); card close 343 ms (pad) / 334 ms (laser) after the move or leave, never data-state="closing" (M10); peek hover: no .lgs-dwell, opacity .58 unchanged (m1); P-29 rects in laser mode (M9)
S2-V1	shot main p2_c2a_r2_{home_laser,card_pad,card_laser,coll_pad,apps_laser,folder_pad,wn_pad,allgames_plate} (CSS-only)	Viewed composited over room-lounge.jpg, room-studio.jpg and grey (the Read view ignores alpha, so every judgement was made on composites). Findings M1, M3, M4, m2; the rest of the look matches p2_home-apps_home-t1.png (see 1.1)
S2-G1	gates main --route /library/home --only size,type,outline,motion with the pre for rest, the card by pad focus, and the card by a real CDP laser hover (--hover), both modes	PASS all four in rest and card states, both modes
S2-G2	gates on /library/lgs/folder/<VR> (both modes) and What's New (pre: click What's New; both modes)	PASS all four
S2-G3	gates … --media reduce (card, pad); --media contrast (card, pad, shot p2_c2a_r2_contrast)	Reduce: PASS. Contrast: SIZE/TYPE/MOTION PASS; OUTLINE FAIL only on the edge profile of the discs' High-Contrast 2 px edge (lgs-home-disc::before ratios .66-.99), which is the AT-18 look by design; plates near-opaque (viewed). Gate defect, REQ to P10
S2-W1	js t_sweep.js (L.gates.size/type/outline/atRest per state), both modes	Rest, card on row 2, card on row 1, Collections, folder, What's New: all PASS. Apps with a row-1 program attended: SIZE FAIL in both modes (M11)
S2-F1	focus main --pairs pairs.json --keep p2_c2a_r2_focus --mode pad; focus … --pairs pairs_laser.json --keep p2_c2a_r2_hover --mode laser; luma recomputed offline over studio / lounge / dark L18 / bright L225 (601 weights, capsule masks, inset 6)	P-14 Collections segment +63.7 / +63.9 / +73.7 / +47.9 PASS; What's New +59.4 / +59.6 / +73.9 / +48.0 PASS; P-16 band 8-16 px outside the focused selected Recent +28.9 / +24.8 / +41.5 / +8.9 (m3); P-03 laser hover +21.2 / +21.2 / +24.3 / +15.9 (segment), +19.9 / +19.8 / +24.5 / +16.0 (What's New) PASS (inside +10..+25)
S2-D1	js probe7_pad.js, probe9_pad.js (D-pad, bumpers, triggers, X, ≡, B through DispatchVirtualButtonClick; PlayAudioURL spy that records and never plays)	Right ×3, page turn at the row end and back, Down/Up memory, RB → Collections → Apps, RB at the end stays, LB back, RT/LT pages, Up into the top row, along it to What's New, Down back to the remembered cell, Up back to What's New, A on a segment changes section with focus on its first cell: PASS. X logs primary(2377520643) mode: logged. ≡ opens Steam's tile menu (Play, Add to Favorites, Add to, Manage, Developer, Properties..., Cancel), card is-menu-source, B closes it. Sounds: section deck_ui_tab_transition_01, page deck_ui_navigation, menu show/hide; grid moves: no sound at all (M8)
S2-D2	js probe10_stock.js --stock --mode pad	Stock reference: every D-pad move on Steam's Home shelf requests deck_ui_misc_10; on the library, ≡ then B leaves no .gpfocus in any window, as ours does (so focus after the tile menu is parity, not a C2a defect); Left from the first item leaves main on stock too
S2-B1	pad-bfs on Home, folder, What's New (--mode pad), Home --mode laser, and stock Home (--stock)	Inconclusive (tool / device state): stock Home reached 4 of 46 targets (exit, focus back by root, then every node untakeable); ours 1 of 20 (first Down → exit:none, LOST), folder 4 nodes, What's New BLOCKED ("no gamepad focus after FocusApplicationRoot"). The manual sequences S2-D1 reach every section, page, top-row control and cell. REQ to P10. Superseded by S3-B1
S2-R1	js probe4_robust.js --mode pad (render failures injected by wrapping P2's actions.primaryInfo / data.recentGames for the step, restored in finally)	M7 confirmed live; HomeView-level failure falls back to Steam's Home in window-full (good); Home recovers after re-navigation
S2-R2	js probe5_install.js --flags wp.p3,wp.c1a (AT-22: P2's ready() made to throw a transient error, so P1 retries and the module is not left failed for other agents)	While failed: /library/home = Steam's Home in window-full PASS; the retry installed the module, Home back to windowless. A folder route while failed renders nothing on a windowless (glass-less) page (m16)
S2-R3	js probe6_misc.js --mode pad	AT-23 (home.test.lang('de')): Apps segment becomes the grid glyph, Empty Collections label omitted, Steam tokens kept PASS. Leaks over 3 cycles Home → folder → What's New: footer hide count 1 → 1, attention registrations 3 → 3, live 1 PASS. Idle: no running animation, P1 tick 0.5 Hz (0.5 ms / 4 s), listeners 9 → 9 PASS. G-REMOVE for the module (flag off on a folder route with Steam's tile menu open, 8 rows): route /library/home with Steam's Home in window-full, menu closed, 0 C2a nodes or attributes, footer count 1, registrations 3 → 2 PASS; flag back on reinstalls
S2-R4	js probe8_dom.js --mode pad	Row-3 plates outside the window (M3); broken minis (M4); over-fade (m2); Apps page 2 = Hide/Show Screens … VLC media player: 23 programs, no Liquid Glass PASS (AT-5b)
S2-R5	js probe11_aud.js with --stock and themed	AT-6 sanity (relative): Steam's Home on stock /library/home and on our What's New both show 39 visible focusables with identical paths under Steam's page root, 0 shrunk below 85 % PASS (visible part at scroll 0)
S2-M1	motion main --pre m_ramp.js / m_section.js / m_page.js --name c2a_r2_*, and the ramp with --media reduce	Every duration and easing a token, nothing at rest PASS; reduce: opacity only, 180 ms PASS. Filmstrips viewed: the ramp grows from the disc and shows glass before content (good); the section and page strips show a blank frame at f = 0 (M10). m_section "FAIL" is P-54 on the selection pill's designed 156 px travel (tool false positive, REQ to P10)
S2-C1	cmp docs/phase2/mockups/home-apps-home-t1.html --id C2a --live with searchCircle and without	Standard layout: 46 of 49 rects 0 px off; back / search +20 (the map points at the 80 px boxes, m13); card-play +65 (Steam's "Stream"/"Install" with the X badge in pad mode). Viewed side by side: same scene. Default (low) layout: rows +24..+40, segments +80 (the documented fallback, HA §3.1.1)
S2-N1	native-session 13:21-13:25 (16 steps: reporter snapshots, 8 sgcheck, 5 hv --look, conformance --only P-11,P-34,P-46..P-51)	See 1.2
S2-N2	native-session 13:27-13:30 (__LGS_LAYERS.debug for the name plate; shot of Steam's texture with acked plates)	See 1.2
S2-L1	ledger --out <scratch>	All 24 rows owned by C2a are open (M14)

1.1 Visual verdict against the mockups (CSS-only)

Composited over the mockups' rooms, rest Home reads as the visionOS app-selection view: centred honeycomb, one-line labels with the on-room shadow, a light segmented control with the white pill, a small What's New capsule, page dots, a blurred peek. Sizes match p2_home-apps_home-t1.png (S2-C1). Two differences come from C1a and are not C2a's: the default layout puts C1a's 536 × 80 search capsule above the section control (the searchCircle fallback), and Home's Back is a bare chevron that is hard to see over a bright curtain. The Collections and Apps views and the folder route look right apart from M3, M4 and m2. The open card is where it stops feeling native (M1).

1.2 Native (two sessions; P7 complete)

• Report: windowless, shapes: [], 17-18 plates (C1a's shell-back 90 × 90 tex = 60 CSS and shell-search 780 × 96 = the visible 520 × 64 capsule: session 1's REQ to C1a is resolved), 4 mosaic bands, errors: []. Card: one pop home-card 480 × 360 r 54 at 0.0407 u (15 mm at r 1.0), interactive: false, over its own occluder plate. Menu open (with wp.c1c): one pop, the menu at 10.0 mm, modal, card at 0: AT-10 PASS for the card and the menu.
• sgcheck: card pad PASS; card laser PASS; folder card PASS; menu PASS; All Games attended FAIL R1 (texture px (1443, 797)); Apps program attended FAIL R1 ((1275, 527)); laser "rest" FAIL R1 (the lab's parking point (1400, 900) tex lies on the All Games disc, so a disc was attended: REQ to P10); What's New FAIL R2 (tabs, tab-arrow, tab-arrow.1 at 4.43 mm). See M5, M6.
• conformance (native step): P-11, P-13, P-17, P-46, P-47, P-48, P-51 PASS. Its failures on /library/home are not C2a's: P-01 / P-89 examples come from other sheets (no .gpfocus selector of 41-home.css lacks a mode scope; its :hover rules are resets); P-02's one example is 41-home.css's Steam-ring reset (box-shadow: none), which is fine; P-43 is C1a's pointer-proxy ring; P-82 is C1a's Back text; P-08 is the absolute-path AUD artifact of the moved Home.
• hv (5 looks: card, laser lift, 30° off axis, ramp, What's New): the headset faced away from the dashboard, which sat at the bottom edge of every frame, partly out of view. Disc art, plates, the top row, the card with Play and What's New's window glass were visible; no grey slab (the REQ #17 symptom was not reproduced at 13:23). Doubling, plate glass beside the lifted disc and the ramp frame (AT-10c, PLAN-2a-1 native) could not be judged at that size. Frames deleted.
• Steam's texture with acks (p2_c2a_r2_native_card_tex.png): the acked card is transparent (rgba(0,0,0,0)), so its crop carries the lifted own disc, its cloud badge and the tops of the row-3 discs (M1).
• P8's answer to REQ #17 (C2a.md, 14:12, six native sessions 13:05-14:04): not reproducible with today's tree; two daemon/reporter defects that drew such a slab are fixed (mosaic bands next to a cover; main left out while SteamVR's own page shows). P8 independently saw M1's native half: "the card's rectangular crop carries the neighbour disc's corner to +15 mm". P8 also leaves one P7 item open that touches Home: at hv --offaxis 20 the art that the card's hole splits off goes behind the disc plates (REQ P8->P7).

1.3 Session 3 (15:15-16:10): what changed since session 2

Scripts in the reviewer's scratch (probe13_bounce.js … probe19_rest.js, list5.txt, list6.txt); outputs out_s3_*.txt, r3_bfs_*.json.

Id	Command	Result
S3-A1	js probe13_bounce.js --mode pad (card open on cell 6, Down, back Up after 60 / 200 / 280 / 450 ms; card sampled every 50 ms for 2.2 s)	60, 200, 280 ms: the card never leaves (g250820@800 throughout). 450 ms: card gone while away, back 0.8 s after the return (a fresh ramp, as designed). M2 fixed (pad)
S3-A2	js probe2_pad.js --mode pad, js probe3_laser.js --mode laser (session 2's M2 probes, unchanged)	Pad 120 ms bounce: card kept at +0.6 s and +2.2 s. Laser 120 ms excursion onto disc 7 and back onto the card: card kept at +0.7 s and +2.2 s. M2 fixed (both). Unchanged: card close 333 / 332 ms with no data-state="closing" (M10, m5); P-29 Play rects as in session 2 (M9); peek hover gives no dwell, opacity .58 (m1)
S3-A3	js probe13_bounce.js --mode laser (excursions 60 / 200 / 280 / 450 ms)	60, 200 ms: card kept. 280 ms (with L.hover's own overhead, past the 300 ms grace) and 450 ms: card closed and reopened 0.8 s after the return. As designed
S3-L1	js probe14_labels.js --mode pad (PLAN R2-15 E-GRID (labels) on every .lgs-home-label, the plate name, card title and status, segments, What's New; Recent pages 1-2 with and without a card, Collections, Apps pages 1-2 with the longest name's plate, folder VR)	0 issues on 91 label instances in 8 states: every label is one line, inside its cell, ellipsized by its own box where it overflows (text-overflow: ellipsis; 19 instances ellipsized); no word-break / overflow-wrap / hyphens override; the attended cell's plate shows the whole name ("Frametop Display Settings", 292 wide). PASS (row-3 plates cut by the window edge stay M3)
S3-B1	pad-bfs (P10's 14:15 recovery: opposite, tree, root, overlay, nav) on /library/home --mode pad and --mode laser, the VR folder --mode pad, What's New --mode pad, stock /library/home --stock	Home, pad: 26 nodes, 24 reached; "unreached": the open card's Play (158, 248, 170 × 80) and More (370, 248, 80 × 80), which are div role="button" under the cell's nav node, not Steam focusables (a pad-bfs universe artefact, M15); 5 exits (Left at column 1, at Recent, at C1a's capsule), all recovered (overlay), as stock; 1 irreversible (What's New -Left-> Recent, Recent -Right-> Collections: a take-order artefact, S3-B3); 1 node untakeable (S3-B2). Home, laser: 26 of 26 reached (no card opens on focus in laser mode), same irreversible, B → history back. Folder, pad: 29 nodes; "unreached": the card's Install and More (same artefact); 1 irreversible: C1a's search capsule -Down-> cell 3 of row 1, which -Up-> "View in Library" (a real path, M15); 4 row-end cells untakeable (S3-B2); B → /library/home. What's New, pad vs stock Home: 44 nodes each, the same 72-target universe, 36 unreached each (Steam's avatars, feed arrows, "Special Offers"), 15 irreversible each, all of Steam's own kinds; B → /library/home (HA §3.6). G-PAD: What's New PASS by parity; Home PASS once the artefacts are set aside (S3-B2, S3-B3); folder FAIL (M15)
S3-B2	js probe16_take.js --mode pad (direct takes of the row-end cells pad-bfs called untakeable; D-pad Right past row 2's end and back)	All six takes land and hold at 0.9 s; Right from cell 9 turns to page 2 (AMID EVIL VR), Left returns to cell 9 on page 1. The "untakeable" nodes are a pad-bfs artefact (All Games repeats on every Recent page with one key; nodes on later pages), not a C2a defect. REQ to P10
S3-B3	js probe17_top.js --mode pad (the top row on real paths)	Row-1 cell, Up → Recent (selected), Right × 3 → Collections, Apps, What's New, Left → Apps, Left → Collections, Right × 2 → What's New; Down → the remembered cell, Up → What's New (top memory), Left → Apps. Only an artificial order (Recent taken, then What's New taken directly) gives What's New -Left-> Recent: the segment group's own focus memory. Reversible on every user path; the pad-bfs irreversible is an artefact
S3-B4	js probe18_b.js --mode pad (history AllGames → Home; B on a cell, without and with the card open)	Both: /library/home → /library/tab/AllGames (stock SN-N3), session-3 retry with native=off. pad-bfs's "B: nothing" in pad mode is not reproduced on a defined history (most likely its own history state at the end of the sweep)
S3-G1	js t_sweep.js both modes (L.gates.size/type/outline/atRest per state, session-5 gate code)	Rest, card on rows 1 and 2, Collections, folder, What's New: SIZE/TYPE/OUTLINE PASS in both modes. Apps with a row-1 program attended: SIZE FAIL in both modes (M11, unchanged). Laser folder "rest": 6 running transitions at 1.1 s once; not reproduced (S3-G3)
S3-G2	gates main --only size,type,outline,motion on Home rest (both modes), card (pad; laser with a real --hover), folder (pad), What's New (pad)	PASS all four in every state (same verdicts as S2-G1/G2 under P10's session-5 SIZE, OUTLINE and E-SEG ≥ 140 rules)
S3-G3	js probe19_rest.js --mode laser (folder entry, sampled at 1.1, 2.5, 4 s)	No attention, no card or plate, no running animation at any sample: PASS

1.4 Session 4 (16:20-16:30): short re-check

Id	Command	Result
S4-0	md5 of device/rt/41-home.js, theme/41-home.css, theme/layers/41-home.json on the PC and on the Frame (frame_ssh.py run)	8e0aae98…, 5d9990f3…, 8c8a0faa… on both, the same as session 2. C2a's code is unchanged since the review's live runs
S4-S1	Static greps on 41-home.js and 41-home.json	Unchanged. M3: NamePlate puts the plate at y + 74 with no bottom clamp (538-547). M4: minis are bare <img> with no onError (423). M5: no "r" on home-disc. M6: no supersedes. M7: one inner ErrorBoundary with no fallback (817). M8: homeSound is called only for tab and page (676, 683). M12: HOME_SECTIONS has three entries, and nothing calls desktopWindow
S4-G1	gates main --route /library/home --pre @pre_wait.js --only size,type,outline,motion --flags wp.c2a,wp.p3,wp.c1a --mode pad|laser (16:23, native=off)	PASS all four in both modes (pad: SIZE skipped 3 as covered by the transient card)
S4-W1	js t_sweep.js --flags wp.c2a,wp.p3,wp.c1a --mode pad (P10's newest L.gates)	Rest, card on rows 1 and 2, Collections, folder and What's New pass SIZE, TYPE and OUTLINE. Apps with a row-1 program attended fails SIZE: "Apps" segment "P-08 hit 99 % own, 1 % other (lgs-home-cell … CMake)". M11 unchanged

2. Findings

Blocker

None.

Major

M1. The open card does not occlude what lies under it, in either tier. (session 1's M4, now also native)

• The card is mounted inside its cell (D-C2a-10) and the cell's disc stays at opacity 1 and ×1.10 under it (S2-P1: disc (350, 350) 132 × 132 under a card whose art ends at y 468, so the disc's lower edge, glow and 40 px cloud badge sit under the text area).
• CSS-only: the text plate is rgb(0 0 0 / .66) under a white .14 → .03 gradient with backdrop-filter: none (41-home.css 550-551, 573), so a third of everything below shows through: the own disc, and the neighbours the card overlaps (a row-2 card covers the row-3 discs by 20 px; a row-1 card in the low layout covers the row-2 discs by 62 px, so "Half-Life 2: VR Mod"'s art shows directly under "12.4 hrs played all time" in p2_c2a_r2_folder_pad.png). The lifted disc also pokes 8 px above a row-1 card's top edge.
• Native: the card's plate is acked, so its CSS fill becomes transparent (05-native.css 222-226) and the crop at +15 mm carries Steam's texture under the card at full strength: the own disc and slices of the neighbouring discs are lifted to +15 mm and cut at the crop's edge (p2_c2a_r2_native_card_tex.png, S2-N2).
• HA §3.4 says the disc morphs into the card; p2_home-apps_home-t1.png shows nothing under it. This is the package's headline state.
• Fix: hide the attended cell's disc (and its cloud badge) while its card is open (fade it on morph-open); give the card the in-page liquid blur PLAN §1.6 allows in CSS-only mode and a more opaque text plate; keep the card from covering other discs (or give the covered cells visibility: hidden art under it); in native mode the crop must not contain other discs (exclude them from the crop, or keep the card inside its own cell's band).

M2. The card never comes back after a quick bounce (both inputs). Resolved in session 3, not counted: P3 changed 05-attention.js at 14:42 (a step that comes due during the leave grace waits, in every input mode; new onReenter; contracts/interaction.md §2.2), and S3-A1 to S3-A3 show the card kept after 60-280 ms bounces in both modes. C2a needs no change; an onReenter handler would only add robustness. Session 2's text, for the record:

• Pad: card open on cell 6, Down then Up within 120 ms: at +0.6 s and +2.2 s card: null, att: null, focus on cell 6 (S2-A1). A 350 ms bounce keeps the card.
• Laser: card open on cell 6, a 120 ms excursion of the pointer onto disc 7 and back onto the card: the card closes and does not reopen while the pointer rests on the card (S2-A1).
• Cause: P3 keeps the old attention record alive during the 300 ms leave grace and does not refire its steps on re-entry (05-attention.js 216-224), while the neighbour's 400 ms step still fires during its leave grace (234-239: the laser check is skipped once a leave timer is set; the pad path has no check) and sets HS.att to the neighbour; the neighbour's onLeave then clears HS.att (41-home.js 150-165). Nothing ever sets the card's key again.
• AT-8 does not test it; a D-pad correction or ordinary laser jitter at the card's edge is enough.
• Fix (C2a): in onEnter, restore HS.att from the attention record's reached steps (HS.attend.active()), and ignore onStep from an element whose leave is pending. A P3 change (no step while a leave is pending) would also fix it; REQ to P3.

M3. Row-3 name plates are cut off by the window edge in the shipping layout. (new)

• In the low layout (the default while C1a's searchCircle is off) a row-3 plate's top is at y 596 + 74 = 670, so its 60 px capsule ends at y 730 on a 720 px window: All Games (906, 670) 139 × 60, Apps cell 11 (433, 670) 190 × 60 (S2-R4). p2_c2a_r2_allgames_plate.png shows the capsule's lower half missing and the page dots hidden under it.
• All Games is cell 13 of every Recent page, so this is seen on every Home visit with a gamepad (and every 13th cell of Collections and Apps).
• HA §3.4 asks for the plate to be "clamped 16 px inside the overlay"; NamePlate has no clamp (41-home.js 538-547).
• Native: the reporter sees the clipped 50 px plate (home-plate 210 × 75 tex) and drops its pop (rule 5).
• Fix: clamp the plate's bottom to y 704 (place it above the label line, or above the disc, for row 3).

M4. Broken-image icons in the Non-Steam and Soundtracks folders. (new)

• Home › Collections: Non-Steam 9 of 9 and Soundtracks 9 of 9 mini images fail to load (…/assets/<id>/library_600x900.jpg), so both discs show nine Chromium broken-image glyphs (p2_c2a_r2_coll_pad.png, S2-R4).
• The minis are plain <img> without onError (41-home.js 423); homeFolderItem takes homeArt(R, ov).portrait only (245), although HA §3.3 says "header crops for Soundtracks" and shortcuts need their custom art.
• Fix: try the same candidates as useArt (custom portrait, cached portrait, header, logo) and drop a failed mini to the empty tile.

M5. Native: the attended disc pops as a rounded square, and sgcheck R1 fails. (session 1's M9, now live)

• The home-disc rule sets no "r": "capsule"; the reporter reads the disc's border-radius: 50% with parseFloat (device/lgs_layers.js 1090) and reports home-disc 198 × 198 with r 75 (S2-N1), a squircle over the circular occluder plate (r 99).
• sgcheck: FAIL R1 "not covered at texture px (1443, 797)" with All Games attended, and at (1275, 527) with a program attended (S2-N1). Every non-game cell (programs, folders, All Games) keeps its disc popped while attended, and games do so for their first 0.8 s.
• The squircle liquid slab stands outside the round disc at the four diagonals: a halo around the lifted disc.
• G-DEPTH (AT-10b) fails on the package's own route. Fix: "r": "capsule" on home-disc.

M6. Native: the legacy tabs and tab-arrow pops on What's New fail sgcheck R2. (session 1's M8, now live)

• sgcheck on /library/lgs/steamhome: main 4 pops at 0, 4.5, 15 mm; FAIL R2 tabs, tab-arrow, tab-arrow.1 at 4.43 mm, outside PLAN §1.7's {0, 10, 15, 25} (S2-N1). tabs lifts C2a's own track (%{TabHeaderRowWrapper}::after, 41-home.css 247).
• 41-home.json has no supersedes; contract reporter §3.5 names C2a among the owners who retire tabs and card. tab-arrow is superseded only while wp.c2c is on.
• REQ C2c->C2a REQ-11 (C2c.md 55) and REQ P6->C2a (P6.md 36, the literal inset 0 1.5px 1.5px -1px on the Home tab arrows, 41-home.css 337, 352) are still open.
• Fix: "supersedes": [{"id": "tabs", "flag": "wp.c2a"}, {"id": "tab-arrow", "flag": "wp.c2a"}] (or pop the What's New track and arrows with an earlier rule at an allowed depth), answer both REQs.

M7. A render failure inside Home replaces the whole page with an error page floating on the room. (session 1's M7, now confirmed live)

• An exception while rendering the card (injected in actions.primaryInfo, S2-R1): the inner R.ui.ErrorBoundary of HomeView (41-home.js 817, no fallback) shows P2's "This view could not be shown / Back" page; route /library/home, glass still windowless (live 1, because useLive() runs outside that boundary), so the error page has no window glass; focus went to C1a's search field.
• HA §3.7 and P2's override contract promise Steam's own Home (the outer boundary's fallback: steamChildren); that path works only for an exception in HomeView itself (S2-R1 (b): Steam's Home in window-full).
• Fix: drop the inner boundary or give it fallback: HS.steam and an onError that turns live off (refreshGlass()), so the route falls back to Steam's Home in window-full; better, isolate the card in its own boundary that just closes the card.

M8. D-pad moves inside the honeycomb are silent (P-74, must). (new)

• Stock Home requests deck_ui_misc_10 on every D-pad move (S2-D2), and so does our top row (Steam's own flow navigation). Our grid moves (Right, Down, Up between cells) request no sound at all (S2-D1, twice). Page turns (deck_ui_navigation) and section changes (deck_ui_tab_transition_01) are right.
• Cause: the explicit neighbours move focus with BTakeFocus(3) from onMove* and return true (41-home.js 226, 719-775), which bypasses Steam's navigation sound. HA §3.5's "focus moves … already play Steam's own BasicNav" is not true live.
• Fix: play Steam's navigation sound through P3's rt.sound (the same ENavSound Steam uses for a focus move) whenever moveFrom moves focus itself; AT-21 should count it.

M9. The card's primary action leaves the comfort zone for edge cells (P-29, must). (session 1's m5, raised: a must with no recorded deviation, PLAN §4.4)

• P-29: Play within |x − 640| ≤ 400. Live (S2-P1, S2-A1): cell 5 (row 2, x 192) puts Play at x 56-176 (laser, centre 116, 524 off) / 56-206 (pad, centre 131, 509 off); cells 1 and 10 (x 304) put it at 168-288 in laser mode (centre 228, 412 off). 3 of 13 positions per page fail in laser mode, 1 in pad mode.
• Fix: mirror the action row (More left, Play right) on cards left of centre, or clamp the card inward; or record a deviation in HA and PLAN §4.4.

M10. Nothing on Home has an exit motion, so section and page changes blink. (session 1's M5, confirmed with filmstrips)

• The card and the name plate unmount on leave (pad 343 ms, laser 334 ms after the move, S2-A1); data-state="closing" is never rendered, so morph-close 441 ms and materialize-out 350 ms (HA §3.4, §11; 41-home.css 1103, 1136-1142, 1384) never run.
• Section and page changes replace the grid at once (it is keyed by section and page, 41-home.js 824): p2_motion_c2a_r2_section_strip.png and …_page_strip.png show an empty page at f = 0 before the new grid fades in. .is-leaving (the 150 ms page-out) is never set.
• The route-entry stagger reads --lgs-home-ring, which nothing sets (all discs start together); the card's morph uses --lgs-ease-b0 (41-home.css 1131) where HA §3.4 and the morph-open token say b .20.
• G-MOTION's "content before glass on exit" cannot hold; Liquid Glass motion is one of the quality bar's points.

M11. G-SIZE fails in a route-matrix state ("focus on a cell", Apps). (session 1's M6, reconfirmed)

• Low layout: the segments' hit band ends at y 174, row 1 starts at 176; an attended row-1 disc lifts ×1.10 to y 170 and takes the bottom of the band: "Apps" segment "P-08 hit 99 % own, 1 % other (lgs-home-cell … CMake)" in both modes (S2-W1). Same under "Collections" for cells at x 528 and 752.
• Fix: 6 more px between the segments' hits and row 1 in the low layout, or lift the disc from its bottom edge.

M12. Two card builds are missing: the search provider and the Windows section. (session 1's M1, reconfirmed statically)

• device/rt/41-search-apps.js does not exist (card T3: "the search provider (S-A, S-B, S-C)"; HA AT-5(c), AT-13). The log's reason ("C1b is at M0") is stale: C1b is READY with search.addProvider live (21-search.js 223, 487). C1b's own summary row (Q6) is not built either, so nothing would render the provider's cells yet; rank() and the "no Liquid Glass" check (AT-5(c)) are testable today through providers().
• HOME_SECTIONS = ['recent', 'collections', 'apps'] (41-home.js 43): no Windows section (card: "sections Recent, Collections, Apps, Windows"; HA §3.2; AT-4 desktopWindow; retention LA-P4's Home path). Nothing calls actions.desktopWindow.
• Nothing is lost against stock (Steam's "+" keeps both), but the card's scope and two acceptance tests are not met.

M13. Home's download and update states are not built. (session 1's M2, reconfirmed statically)

• 41-home.css styles .is-downloading, .lgs-home-ring-progress and .lgs-home-update; 41-home.js never sets or renders them. A downloading game or one with an update looks like any installed game on Home and on its card (HA §3.3; retention LA-T8 on Home). Steam's stock Home shelf shows that status; on our Home it survives only on the What's New route.

M14. The READY / M3 claim rests on tests that were not run or not recorded. (session 1's M3, reconfirmed)

• No evidence row for HA AT-2 (live), AT-9, AT-10b, AT-10c, AT-11, AT-16, AT-17, AT-18, AT-19, AT-20 (live), AT-21, AT-22, AT-23 or PLAN-2a-1 (live cmp); PLAN §2.1 M3 needs "the package's own tests pass in both input modes".
• glass.py ledger resolves test ids against evidence rows whose first cell starts with the id; C2a's rows are named M3-N1, M3-M1 …, so all 24 C2a functions are open (S2-L1).
• This review ran most of them (table above); C2a should record them under their AT ids after the fixes.

M15. G-PAD fails on the folder route: Down from the search capsule and Up do not return (P-20, must). (new in session 3: pad-bfs was unmeasurable in session 2)

• In the folder the top row holds only "View in Library" at the right (945-1176 × 14-94). Steam's geometric Down from C1a's search capsule (372-908) skips it and lands on row 1, cell 3 (652, 176); C2a's explicit Up from row 1 (focusTop, 41-home.js 723-733, 766) then goes to "View in Library", and only a second Up reaches the capsule (S3-B1, r3_bfs_folder_pad.json: node 7 -down-> 6, node 6 -up-> 3). It is a real path, not a take-order artefact: the user goes round a triangle (grid → View in Library → capsule → grid).
• HA §3.5's one-step memory ("the opposite direction always returns to the cell you came from") covers only moves C2a makes itself, not a move into the grid that Steam made. On Home the same capsule's Down lands on the Collections segment and Up returns to it, so Home is reversible.
• Fix: route Down from the capsule to "View in Library" on folder routes (the folder's top row is C2a's; the capsule is C1a's, so either an onMoveDown C1a exposes or a C2a onFocus on row 1 that records a .gpfocus arriving from outside the grid in the one-step memory). Re-run pad-bfs on the folder in both modes.
• Home passes G-PAD once pad-bfs's artefacts are set aside (all checked live in session 3): the card's "unreached" Play and More are div role="button" (41-home.js 526-531) with no nav node of their own, counted because pad-bfs's universe takes any [role=button] whose ancestor has a nav node (lab_bfs.js 83-95, L.navNode walks 12 fibers up); the gamepad uses them through X and ≡ (S2-D1). Left from What's New is reversible on every user path (S3-B3). The "untakeable" row-end cells take and hold focus directly (S3-B2). Exits Left from column 1 match stock. REQ to P10 below. What's New passes by parity with Steam's Home (S3-B1).

Minor

• m1. Peeks have no laser hover look and no press look. P3's dwell selector is .Panel, button, [role="button"], [role="tab"], a[href], [data-lgs-dwell] (05-attention.js 20); the peek is a bare div (41-home.js 805), so .lgs-home-peek:is(.lgs-dwell:hover, .lgs-attend-80) never matches (live: no dwell, opacity .58). They are the laser's only page-turn targets. Fix: data-lgs-dwell (or role="button" + aria-label).
• m2. Name plates fade labels they do not cover (D-C2a-5 says a label fades only when its text is covered). The plate rule fades every label in the same row within 260 px (41-home.js 710-713): the Apps plate (718, 490) 292 wide fades "Firewall" (text 604-676); the VR plate fades "Soundtracks" and "Empty Collections"; All Games fades "Rez Infinite" (S2-R4, shots). Use the plate's real rect, as the card's rule does.
• m3. P-16 glow band over a bright room: +8.9 L (studio +28.9, lounge +24.8, dark +41.5; S2-F1). P-16 is a "should"; record it.
• m4. The card's +15 mm has 2.4 % click-safe margin (static): its smallest focusables are the 80 px Play and More hits, cap 0.0417 u against 0.0407 u at r 1.0; any window scale below r 0.975 caps the card to +10 mm while the lifted disc stays at +15 (two depths for one item; AT-10 fails). Live r was 1.0.
• m5. HA-17 not built: in gamepad mode the old card stays for the 300 ms leave grace after a move (343 ms measured) instead of closing at once.
• m6. The name plate never pops in native mode: debug() says {"rule":"home-plate","skip":"covered"} (S2-N2): the plate is pointer-events: none, so the reporter's hit test sees the cell under it. Add "hitTest": false to home-plate (HA §10.4 wants +15).
• m7. Program discs lose their stable plate ids (p4, p5, p7 in debug()): 'home-disc-' + item.key contains characters outside [A-Za-z0-9-_.:] for most programs (the key comes from the command line), so the reporter falls back to positional ids and glassd's materialize and acks follow positions, not discs. Sanitize the key (hash it).
• m8. Native card ramp changes depth: home-card has no from, so the card rises from 0 while the disc's pop sinks from +15 when the card opens (HA §10.2 says the depth never changes during the ramp). "from": "cut" or fromMm: 15 on home-card (static).
• m9. Glyph badges are text letters ("X", "≡"; P-27, a should). Recorded as D-C2a-12.
• m10. Mosaic bands do not follow a flat card or name plate (native, static). Bands are fixed per row; a card that stays flat (its menu open, rule 6) loses what lies outside the bands (in the low layout a row-2 card spans y 316-556, outside the bands at 328-350 and 482-530). Give the card and the plate data-lgs-mosaic.
• m11. Unbuilt or dead parts: the two-line name plate (.two, .lgs-home-plate-sub: "27 games", "N collections with no games"), Recent's session MRU of launched programs (HA §3.2), data-lgs-plate-inset: -4 while popped (HA §10.2), .is-attended, .is-leaving, [data-state="closing"], part B of 41-home.css (C1a adopted the hook).
• m12. Peeks of an Apps page show a bare icon without a disc (the VLC cone floats at x 1782 in p2_c2a_r2_apps_laser.png), unlike the disc peeks of the other sections.
• m13. PLAN-2a-1 bookkeeping: C2a-cmp.json maps back and search to C1a's 80 px boxes (+20 px against the mockup's 60 px circles); no mockup shows the shipping low layout (search capsule above the segments), so cmp in the default state fails by design; no PLAN-2a-1 run is recorded in the log.
• m14. Stale log text: C2a's Status says 41-search-apps.js waits for "C1b at M0" and lists REQ #16 as open (P10 answered it). REQ #17 (native grey slab) was not reproduced at 13:23 (discs with art in the headset view); P8 closed it at 14:12 (§1.2).
• m15. What's New entered before Home has rendered in this install is blank (session 1's m2, static): SteamHomeRoute renders HS.steam, which only the Home override sets (41-home.js 141, 842). S2-R1 (c) entered it after Home had rendered and it worked. Render steamChildren captured at install, or redirect to /library/home first.
• m16. A folder route while the module is failed or off is a glass-less empty page: C1a's route map makes /library/lgs/folder/* windowless unconditionally, so a history entry into a folder after a failed install shows nothing on the room (S2-R2, #Footer present). Ask C1a to make folder routes windowless only while the home hook answers, or have P2 drop dead /library/lgs/ entries.

3. Function retention (HA §12, rows owned by C2a)

Row	Laser path	Gamepad path	Verified in sessions 2 and 3
H1 recent games, running first	dwell → card	D-pad, focus → card	Both work (S2-P1, S2-V1); the card survives quick bounces since P3's fix (S3-A1 to S3-A3)
H2 open a game page	click disc / card body	A	Static (cell onActivate → actions.navigate); session 1 logged both
H3 tile menu	card More	≡	≡ → Steam's 7-row menu, B closes (S2-D1); More: session 1
H4 library	All Games cell	D-pad + A	Cell 13 on every Recent page (S2-P1)
H5 feeds	What's New capsule	Up, Right to What's New, A	D-pad reaches What's New (S2-D1); route works; What's New gates PASS (S2-G2)
H6-H8 What's New, Friends, Recommended cards	click	focus + A	Steam's page: 39 = 39 controls (S2-R5); gamepad traversal = Steam's own Home (44 nodes, same 72-target universe, same unreached and irreversible kinds; S3-B1)
H9, T8, R3, R4 status	card art, dots	same	Not installed shows; downloading / update missing (M13); running dot untested (no running game)
T1, R1 primary action	card Play	X	X → primary logged (S2-D1)
C1-C3, N1, N3 collections and folders	click	D-pad + A, B back	Folders work (S2-V1, S2-G2); Non-Steam and Soundtracks discs show broken images (M4)
S2 search result (programs)	—	—	No provider (M12)
P2 launch a program	Apps cell click	A	Static (launchNonSteam); session 1 logged it; search path missing (M12)
P4 desktop window	—	—	No Home path (M12); Steam's "+" keeps it
SN-N2 tab bar from the left edge	—	Left at page 1	Focus leaves main exactly as on stock (S2-D1, S2-D2)
SN-N3 B at the root	—	B	History back (/library/home → /library/tab/AllGames), as stock, with and without the card open (S2-D1, S3-B4)

Every row that exists has both paths; the gaps are the missing builds of M12 and M13, not a lost Steam function.

4. Conformance (P-items for this area)

Item	Result
P-01, P-02, P-89	41-home.css clean: every .gpfocus selector is mode-scoped; laser looks key on :hover / .lgs-dwell / lgs-attend-*; no hover-only reveal (AT-8(d)). The route-level FAILs come from other sheets
P-03	PASS (+16 to +24.5 L, S2-F1)
P-06	PASS (session 1: no scale before .lgs-dwell)
P-08, P-80, P-83	PASS except M11
P-12	More circle tooltip above with Steam's string; no tooltips on text controls (static)
P-13, P-17, P-22	PASS (one lit cell; no ring; entry focus on cell 1)
P-14	PASS; P-16 should: bright room +8.9 (m3)
P-19, P-59	PASS (×1.10 lift, card ramp, page dots, LB/RB, edge paging)
P-20	Home: PASS (S3-B1 to S3-B3, artefacts set aside); folder: FAIL (M15); What's New: parity with stock
P-21	≡ then B leaves no .gpfocus on our Home and on the stock library alike (S2-D2): parity, not C2a's
P-22, P-24	Entry focus on cell 1 on Home and folder, on Steam's first capsule on What's New; B at the Home root = history back (S3-B4), in a folder and on What's New = back to Home (S3-B1)
P-26 / P-27	badges only in gamepad mode / text letters (m9)
P-29	FAIL (M9)
P-33	PASS (18-28 targets)
P-38, P-39, P-41	PASS (labels 20 px Medium with the on-room shadow; card status 18 px)
P-42, P-43, P-45	PASS in CSS-only (gates); High Contrast's 2 px edge flagged by the edge profile (gate defect)
P-46, P-47, P-48, P-51, P-11	PASS in native (S2-N1); G-DEPTH fails R1/R2 (M5, M6)
P-52, P-56, P-58	PASS (gates, motion, reduce)
P-53, P-54	PASS (pill travel flagged by the tool's P-54 heuristic: false positive)
P-61	PASS (Home lattice; empty collections dimmed and last)
P-71	Launch is logged; Home's layers have no dematerialize (M10)
P-74	FAIL (M8)
P-79	RB at the last section plays no bumper-end sound (should; minor, not listed above)

5. Requests

• REQ C2a-R2->C2a: fix M1 and M3-M15 and m1-m16 above (M2 needs no change since P3's fix), then record the HA tests under their AT ids so glass.py ledger resolves them (M14); answer REQ C2c->C2a REQ-11 and REQ P6->C2a (M6).
  • C2a (2026-10-07, R2 fix pass 1, session 6): handled; per finding in C2a.md (Status "R2 fix pass", decisions D-C2a-14 to D-C2a-21, evidence rows R2F-* and the AT rows of this pass). One part disputed with evidence: M10's "b0 instead of b .20" for the card's CSS clip (D-C2a-15: P5's lgs-morph and MO §6.4 keep the clip on b0; glassd draws the overshoot). REQ-11 and P6's REQ answered in their logs.
• REQ C2a-R2->P3: attention steps fire for an element whose leave is pending (05-attention.js 234-239), and a re-entered element never refires its steps (216-224); together they lose Home's card after a quick bounce (M2). Please skip reach() while A.leaveTimer is set, in every input mode, and consider reporting re-entry (onEnter with the reached steps) so consumers can restore their state.
  • Done by P3 (14:42; 05-attention.js defers a step that comes due during the leave grace and adds onReenter; contracts/interaction.md §2.2). Verified live in session 3 (S3-A1 to S3-A3): Home keeps its card after 60-280 ms bounces in both modes.
  • P3 (maintenance session 5, 15:20-16:13 EDT): also checked with this review's own probes on the unchanged 41-home.js (probe2_pad.js 15:22: the card stays on cell 6 at +0.6 s and +2.2 s after a 120 ms Down/Up; probe3_laser.js 15:27: it stays after the 120 ms excursion onto disc 7), and with P3's new runner test IN-14 (gamepad, laser, feed and tooltip paths; PASS 15:33 and 15:55). Evidence: wp/P3.md, session 5.
• REQ C2a-R2->P10: (1) pad-bfs cannot recover after an exit: edge on the current device state: stock /library/home (--stock) reached 4 of 46 targets (back by root, then every node untakeable), What's New gives "no gamepad focus after FocusApplicationRoot", C2a's Home LOST focus on the first Down (13:32-13:35); G-PAD is unmeasurable for Home until it recovers. (2) The lab parks the pointer at (1400, 900) main texture px = (933, 600) CSS, which lies on Home's All Games disc (low layout), so every later laser step on Home starts with a disc attended (S2-N1 laser "rest" FAIL R1); park it on a point no route uses (e.g. outside the window glass). (3) focus reads transparent texels as black on windowless routes (P-16 band 228 vs 8.9 L); composite over a declared room, or warn. (4) motion's P-54 flags a selection pill's designed travel (156 px). (5) gates OUTLINE's edge profile fails High Contrast's 2 px edge, which P-42 exempts.
  • Session 3 status (15:15-16:10; no answer in P10.md yet): (1) works now: every exit came back by overlay on Home, folder, What's New and stock Home (S3-B1). (3) and (5) are in P10's session-5 code (edge_profile.py room compositing, glass.py High Contrast note). (2) and (4) are still open. New, from S3-B1 to S3-B3: (6) the universe counts any [role=button] whose ancestor fiber has a nav node (lab_bfs.js 83-95 with L.navNode walking 12 fibers), so Home's card actions (plain div role="button" inside the focused cell) are listed as unreached targets; require the element's own nav node (the fiber nearest the element that carries node, and that node's element is this one). (7) Paged grids: a cell that repeats on every page with the same label (Home's All Games) gets one key, and cells on later pages are reported untakeable although a direct take lands (S3-B2); key by page too, or replay the page turn before taking. (8) A take-order irreversibility (What's New -Left-> Recent, because the BFS last took Recent inside the segment group) is reported as a defect; it is unreachable on user paths (S3-B3). Re-check an irreversible edge by replaying the forward move from its source reached by D-pad, not by a direct take.
  • P10 (2026-10-07 session 5, 16:35): done, all eight; contracts/lab.md §1, §3, §4, §10; evidence wp/P10.md rows S5/S5b/S6. (1) pad-bfs recovery: after an exit the sweep tries the opposite press, main's nav tree, the root, Steam's EnsureVROverlayVisible() (+ root, Down/Up) and last Navigate(route, replace), each held 150 ms; 15:55-16:19: stock Home 7 exits all back (overlay), What's New starts (no "no gamepad focus"), C2a Home every exit back. (2) The pointer parks at (1400, 900) CSS px, outside the viewport, at every lock exit after a hover and in L.unhover() (15:39: gates main --route /library/home --pre "@hover .lgs-home-disc,600" --mode laser, then a check step: 0 elements hovered). (3) focus --room auto|bright|dark|none: translucent texels are composited over P-14's bright and dark rooms and judged by the worse; your P-16 band reads +7.6 over the bright room (FAIL < 20; it read +219.5 over black). (4) motion P-54 judges entries only; the pill's 156 px travel is listed under moves (15:52 PASS). (5) Under --media contrast OUTLINE's edge profiles are listed, not judged (P-42; 16:13 PASS, highContrast: true). (6) A pad-bfs target must own its nav node (node.m_element === el, probe 16:25: Home's 18 cells/segments own theirs, the card's Play and More do not); such elements are listed under inner, not counted. (7) A node that is not mounted is replayed from any of up to three recorded ways in (mounted sources first) and is retried once at the end of the queue before it is untakeable. (8) A failed reversibility check is measured again with the source reached by D-pad; a pure take-order result goes to takeOrder (not counted). Live 16:32 (pad-bfs --route /library/home --flags wp.c2a,wp.p3,wp.c1a --mode pad): PASS, 26 nodes, 0 of 26 unreached, 0 untakeable, 0 irreversible, inner = Play and More, takeOrder = What's New -Left-> Recent (on the D-pad path -Left-> Apps and back), 5 exits all back by overlay; B went back in Steam's history (another agent's earlier route), as on stock. TL-3 PASS (45 = 45) after the change.
• REQ C2a-R2->C1a: Home's root Back is a bare white chevron with no backing, nearly invisible over a bright curtain (ztop composite of p2_c2a_r2_home_laser.png); the mockups draw a 60 px glass circle. Please give it the circle (or the on-room shadow) on windowless routes. Also m16 (folder routes windowless only while the home hook answers).
  • C1a answer (2026-10-07, R2 fix pass): done (review R2 M4). On windowless routes Back (also as a section root) and the search field are CSS plates over the room: P4's --lgs-mat-plate-bg (black .58 under white .16 → .04) with the E4/E5 edge, the lit top lip and a soft contact shadow, hover + .08; in native mode P6's ack clears the plate fill, so only the light adds stay over glassd's shell-back / shell-search plates (60 px circle, 520 × 64 capsule). m16: /library/lgs/folder/* is windowless in the route map itself; a glassMode hook only overrides it while one answers.
• REQ C2a-R2->P8: REQ C2a->P8 #17 (native grey slab on windowless Home) was not reproduced in session 2's native runs (13:21-13:30: discs with art, plates acked, card popped); please confirm the fix and close it in C2a.md.
  • Done by P8 (14:12 in C2a.md, six native sessions 13:05-14:04): not reproducible with today's tree; two defects that drew such a slab are fixed (RP-MO in wp/P6.md, contracts/daemon.md §9). P8's FYI about the card's crop carrying a neighbour disc is this review's M1 (native half).
  • P8 (maintenance session 5, 16:07-16:12 EDT): confirmed again on today's tree in two native sessions with your flags and fresh headset frames (--grabs 3; the window brought into view with a lab-only override). Native Home showed glass discs with their art, the card at +15 mm and no grey slab. Every snapshot showed main with shapes: [], 18 plates, cover: 0. The closing note is in C2a.md; evidence is in wp/P8.md, session 5.