MichaelScottsman--LiquidAss-Frame/docs/phase2/wp/C1b.md

C1b Search: evidence log

Package card: docs/phase2/PLAN.md §2.4 "C1b Search". Concept: docs/phase2/concepts/window-nav.md §4 and §9.1 (WN, file owned by C1a; C1b sends its text as requests, below). Inputs: WN §4, §9.1; XC §2; HA §6; PLAN §1.9. Steam build at the time of writing: 11094443 (read from P10's cache /tmp/lgs/steam-build.json on the Frame, 2026-10-07).

C1b owns: theme/21-search.css, device/rt/21-search.js, docs/phase2/mockups/window-nav-search.html, window-nav-results.html, docs/phase2/wp/C1b.md, docs/phase2/wp/C1b-cmp.json.

Status

R2 fix pass done (session 3, 17:45–20:20, 2026-10-07). READY: wp.c1b (with c1bSummary at its default, on). Every major of review R2 is fixed and re-verified live in both input modes and in a native session, except what needs another owner's ruling: G-AUD's by-design relocations (REQ C1b->Coordinator #11) and G-SIZE's sampling of C1a's More circle on its own host (REQ C1b->P10 #18). G-PERF fails on this route for the theme as a whole, also with C1b's runtime off (REQ #19). Per-finding table and the test table: Log, "session 3".

• What the user sees now: activating the field (click, or Up from the top row) presents a thick-glass sheet at R2-6's place over the page you were on, which stays sharp under the .35 scrim while only what lies behind the sheet is frosted. Zero state: Recent Games discs, Recent Searches, See All, the hint, quiet Open / Back. Typing turns it into the All summary: the scope bar (Steam's categories and counts, compact segments), the Top Hit card with Steam's playtime, the next four library matches as posters, the Software cell (HA's programs), "Store (N)" over Steam's own store tiles and See All, more below; X plays (Steam's primary action), ☰ or the More circle opens Steam's own game menu. B anywhere in the sheet dissolves it (content, then glass, by 190 ms) and focus lands on the card you left. In native mode the sheet is one +10 mm modal pop over glassd's thick slab.
• Built files: theme/21-search.css, device/rt/21-search.js, theme/layers/21-search.json, both mockups, docs/phase2/wp/C1b-cmp.json; drafts identical in _wip/; check-theme PASS; hook lists up to date.
• Device left: theme on, CSS only, wp.c1b off, no flag overlay of mine, no search node, no menu, keyboard hidden.

What remains:

Next	Why
Coordinator ruling on G-AUD (REQ #11), P10's More-circle sampling (#18), virtualised rows (#14), exit motion reading (#15), T1 E-SEG selector (#13)	Lab / PLAN rules, not C1b's code
A on the Software cell and on a program Top Hit on the live D-pad path (HA AT-13)	Reachable in pad-bfs; the scripted presses did not land this session
The focused result does not pop on its own in native mode	The reporter never keeps overlapping pops; it rides on the +10 sheet
WN §4 text to the build (REQ C1b->C1a #16)	C1a's file

Interface announced (names are final; code lands at M1 stub, M3 behaviour)

Areas and tests key only on these.

What	Name and values	Notes
Runtime module	search in device/rt/21-search.js, deps ['react'] (P3's rt.sound, C1a's more used when present), flag wp.c1b (built-in default off until V2 accepts); html.lgs-c1b on main while installed (lgs-c1b-presented once the sheet is on screen, lgs-c1b-leaving during the dismiss). Kill switch c1bSummary (unset = on): off gives Steam's own grid on All	Fails closed: with react not ready the route is Steam's, restyled by 21-search.css (T1). Two optional P2 patches (c1b.libresult on Steam's library result memo, c1b.results on Steam's results list memo): without them results keep A and B only, and All is Steam's grid
Provider API (PLAN §1.9)	__LGS_RT.search.addProvider(spec) → {remove()}	Also rt.search.providers() → [{id, slots, calls, ms, failed}] for tests
spec.id	string, unique ('c2a.programs')	A second addProvider with the same id throws
spec.slots	subset of ['tophit', 'software']	Where its candidates may appear. tophit: the Top Hit card. software: the Software cell at the start of the Store row (S-B)
spec.rank(query, ctx)	sync → [{key, name, tier, icon, subtitle, data}], best first	tier: 0 exact title, 1 title prefix, 2 word prefix, 3 substring (case- and accent-insensitive). ctx = {steamBest: {tier, appid} | null, lang}. Must return in ≤ 2 ms; a slower call is skipped for that keystroke and logged
spec.render(candidate, slot)	optional → React element for the cell's content	Our component owns the Focusable, sizes, states, depth tags and legends; without render we draw the standard cell (56 px disc with icon, name, subtitle)
spec.open(candidate, ev)	A or a click	Must go through rt.react.actions (a program: launchNonSteam(cmdline, ev); "and N more": nav.go('/library/home') + Home's Apps section with that program focused, C2a's API)
spec.primary(candidate, ev)	optional, X (S-C)	Through rt.react.actions.primary(appid, ev); shown as Steam's "Play" legend (#GameAction_Play)
Top Hit rule	Among Steam's library matches of the best tier, the most recently played (Steam's last-played time); ties keep Steam's order. A provider candidate becomes the Top Hit only with a strictly better tier than Steam's best	"half" → Half-Life: Alyx (all "Half-Life" titles are tier 1; Alyx was played last). "vlc" → the program VLC media player (no Steam library match)
Liquid Glass	Dropped by the search module from every provider's candidates (HA-10), whatever the provider returns	HA AT-5(c)
Safety	A provider that throws is disabled for the session (logged); providers are removed on lgs off; nothing a provider returns is stored beyond the current query
DOM hooks (live, R2 fix pass)	Zero state: .lgs-search-zero[data-lgs-search="sheet"][data-lgs-search-state="zero"] with [data-lgs-search="recent" | "recent-search" | "seeall"]; results and categories: Steam's %{GamepadSearch} > %{GamepadTabbedPage} tagged data-lgs-search="sheet" + data-lgs-search-state="results" | "category" over our glass .lgs-search-glass[data-lgs-search="glass"] (with the frost .lgs-snap-frost inside it); the snapshot div.lgs-snap (inert) and the scrim .lgs-search-scrim. The All summary: .lgs-all[data-lgs-search="summary"] > .lgs-all-row1[data-lgs-search="library"] (.lgs-all-hit[data-lgs-search="tophit"], .lgs-all-poster[data-lgs-search="poster"]), .lgs-all-soft[data-lgs-search="software"], .lgs-all-storerow[data-lgs-search="store"] (Steam's tiles in .lgs-all-run-store, .lgs-all-seeall), .lgs-all-run-lib, .lgs-all-run-friends. Library cards and Steam's library tiles carry data-lgs-search-app=<appid> (the More circle's hosts). Steam's segments carry data-lgs-exempt="E-SEG", its "No Results Found" data-lgs-q (the echo)	The layer (snapshot, scrim, glass) is portalled into %{BasicHome} under Steam's content (z −1) and keyed, so it is presented once. Depth rules: theme/layers/21-search.json (C1b, R2-16)
Test hooks	__LGS_RT.search.debug() (focus memory, patches, B count), .summary() (where each of Steam's result ids went: tophit, row1, store, rest, friends, storeTab, redirect, software, steamBest), .snapshot(), .recent(), .candidates(q, slot), .setQuery(q)	Read-only except setQuery (Steam's own input handler)
Sounds	rt.sound('sheetIn') / 'sheetOut' on present and dismiss (one each: the Back circle on /search is the sheet's close, so Steam's own HideModal does not play as well); Steam's own tab sound on a scope change; nothing on hover	P3 contracts/interaction.md §5

M0 conformance: every PLAN §1 decision that touches search

PLAN	Decision	Where it landed
§1.1	Search route and sheet: WN → C1b; HA contributes through the provider API	Interface above; REQ C1b->C2a #5
§1.2	/search is a window route: 1280 × 656 glass, 64 px ornament margin	Both mockups: window 656; sheet bottom at y 600, ornament from 628
§1.3	Text buttons 60 capsules, 24 padding, 24 Semibold; icon buttons 60 / 80 hit, ≥ 24 px clear; segmented 64 track, segments 56–60 × ≥ 140 (E-SEG compact 120); mini circle 44 alone (E-MINI); search field 640 × 64 on the search route, no microphone; More circle 60 / 80; Home disc 120	Recent Searches capsules and See All now 60 / 24 Semibold (were 60 / 500 weight); "See All" and "Open" 60 (were 48); the 54 px category capsules (58 px pitch) are gone; scope bar segments ≥ 120 wide in a 64 track; Steam's ‹ › arrows 60 px circles; Recent Games discs 120 (were 112), Home's one-line labels; More circle 60 inside the card
§1.4	Input-mode keying; laser looks on :hover; lift after 80 ms dwell; gamepad focus + .28, spot .16 upper third, arc ×1.5; content cards lift ×1.05, Home discs ×1.10; no row or button scale; glyph badges in gamepad mode only	Recipe block in both mockups (same values as C1c's); zero state in laser mode shows no glyph badges and no focus look in the sheet; results in gamepad mode shows the Top Hit lifted ×1.05 and glyphs in the ornament
§1.5	MO tokens; at rest nothing runs; route overrides within Steam's timeouts (enter ≤ 800 ms, exit ≤ 200 ms)	Proposed WN §4.9: sheet-in 735 (≤ 800); dismiss changed from sheet-out 514 to a 150 ms content-and-glass fade, because leaving is Steam's NavigateBack and React unmounts our route after Steam's ≤ 200 ms exit; glassd dissolves the slab in native mode
§1.6	Search sheet = thick; slab under the pop in native mode; CSS-only blur 30 + tint .30–.40 + edges; no outlines (edge lobe); no glass on glass	Sheet data-mat="thick"; inside the sheet nothing has its own backdrop-filter (P-45): the Open capsule and the More circle are fills, the discs are content with a specular arc and contact shadow instead of Home's glass bezel
§1.7	Search sheet +10 mm non-interactive (wearer +30 → +50); focused content card ≥ 150 px (search result) +15; click-safe on the smallest focusable; ≤ 4 dz; nothing pops while scrolling; only the modal pops while one is open	Mockups annotate sheet +10, Top Hit +15 (its smallest focusable is the More circle's 80 px hit → +15 allowed); proposed text: discs and the Software cell never pop, nothing pops while the sheet scrolls, the snapshot never pops
§1.8	Sheets: Steam's scrim restyled to black .35, never t1	Scrim .35 on the snapshot (lgk-scrim); WN §3.7's "t1 tint .65" row is withdrawn for search (REQ #1)
§1.9	One field, three variants; no microphone; search = WN §4 sheet over a snapshot; HA's split view withdrawn; provider API (S-A, S-B, S-C); Liquid Glass excluded	Field 640 × 64 on the route; Top Hit can be a program; the Software cell beside the Store row (results mockup, real program); X = Play in the ornament; HA's mockup must follow (REQ #6)
§1.10	Legends never hidden; capsule when any action besides A/B; quiet legend when only A/B (in the margin on window routes); A and B in both modes	Zero state: quiet legend "Open / Back" in the margin, labels only (laser); results: capsule "Play · Options · Open · Back" with glyphs (gamepad). The shared builder still drops A/B in laser mode, so the zero state draws the quiet legend inline (REQ #2). Legibility finding: REQ #3
§1.11	More circle: one helper, 60 / 80, inside the host (cards top right, inset 10), dispatches the host's onMenuButton	Results: More circle 60 inside the Top Hit, top right inset 10; the item menu path is in the retention table
§1.12	Sheets ≤ 960 wide, close circle, scrim .35; Steam's order and default focus untouched	Results sheet 1040 → 960 (centred at x 160); the close circle is the window's Back circle above the scrim (no second circle; reasoning in proposed §4.4)
§1.13	Tooltips only on icon-only controls; Steam's sounds; haptics off	The only icon-only controls in the sheet are ‹ ›, the × and the More circle (tooltips from P3); sounds in the interface table
§1.15	Steam's strings by source text; English only for en*	Read-only grep of Steam's English UI strings on the Frame (log below); labels changed: "Recently Played" → Recent Games (#LibraryHome_RecentGames), "Show All Results" → See All (#StoreApp_SeeAll), "In the Store" → Store (158) (#Search_Results_Header_StoreApps_With_Count), category capsules → Steam's section headers with counts, "Close" → Back (#ActionButtonLabelBack, what Steam's legend says on /search today), Top Hit status → Playtime: 6.3 hrs (#BasicGameCarousel_TotalPlayTime_Hours). English-only: "Recent Searches", "Top Hit", the zero-state hint, "and N more"
§1.16	Exemptions	The × in the field is E-MINI; segments E-SEG (compact 120); nothing else in the sheet needs one
§1.17	Safe defaults behind flags	The whole package behind wp.c1b (off). No new sign-off items: the dimmed-page click that closes search is WN §4 behaviour adopted by §1.9 (recorded as a VP P-65 deviation for V2, proposed §4.4)

Proposed WN text (REQ C1b->C1a #1: replace WN §4 with this, and apply the line edits after it)

4. Search: a sheet over the page you were on (built by C1b)

Revision 3 (PLAN §1.9, with §1.2–§1.15). Mockups: window-nav-search.html (zero state, laser mode, keyboard up) and window-nav-results.html (results for "half", gamepad mode, focus on the Top Hit). Both show the device's real library, its real program "Half SBS Toggle", Steam's real counts for "half" and Steam's own strings. Shots: p2_window-nav_search.png, p2_window-nav_results.png.

4.1 Why it changes

Today search is a 1.23° strip whose focused state is a light-grey web form. Entering it replaces the page with Steam's results route; leaving needs Back; an empty query lists the whole library; the keyboard shows no echo 22–28° below the text (shell-nav B.4.1, B.5.2, B.6). visionOS treats search as a system feature: suggestions and recent searches before typing, a scope bar, grouped live results and a no-results view that echoes the query (VP §3.2, P-62).

4.2 The model

Search stays on Steam's own routes (/search, /search/tab/<Category>), so history, B, the keyboard, IME and every result path are Steam's. The route's glass mode is window (PLAN §1.2). It is presented as a sheet over the page you were on:

1. On activation T2 takes a context snapshot: cloneNode(true) of the current page's content element (the child of %{TopLevelTransitionSwitch}) and of the toolbar's Large Title node, every id removed, inert and aria-hidden set, the scrollTop of each scrolled descendant copied, no listeners. It is a decorative picture; Steam's focus never sees it, because focusables register through React context, not through the DOM.
2. The search route renders (T3 override, §4.10): the snapshot under Steam's scrim restyled to black .35 (PLAN §1.8; never the t1 tint), then the sheet.
3. The snapshot is pixel-identical to the page that just left, so the route change is invisible: the page stays while the sheet materializes over it.
4. Leaving (B, the Back circle, a click on the dimmed page, any tab) is Steam's NavigateBack: the real page returns under the same picture, and Steam's focus history puts focus back on the poster you left.

4.3 The trigger

• Activation = the header input receives DOM focus (focusin on %{SearchBox}). A laser click gives it focus; Steam's gamepad focus (D-pad Up from the top row) calls focus() on it.
• On activation T2 takes the snapshot and, if the route is not already /search…, calls Navigate(Routes.Search.Root()). If Steam navigates by itself on focus, nothing is added.
• The keyboard rises as Steam raises it (onKeyboardShow), on activation only, never on arrival (P-63).
• Input mode comes from P3 (rt.input, html.lgs-input-*); Steam's getters are never written.

AT-9a and AT-9b exercise the real trigger with both inputs and record what Steam itself does.

4.4 States

State	Route	Sheet	Tier
Idle	any	— the field: 520 × 64 on section roots, 640 × 64 on nested routes, a 60 px circle on Home, folders, hero routes and the other routes of PLAN §1.9 (C1a)	T1
Activated, empty	/search or /search/tab/All, empty query	Zero state, 880 × 500 at (200, 100)	T2 snapshot + T3
Typing	/search/tab/All + query	Results, 960 × 500 at (160, 100); the width grows 880 → 960 on snappy; the height never changes	T3
Category	/search/tab/{Library,Friends,Store,Tools,Hidden}	Steam's own tabbed page hosted in the sheet, 960 × 500 (AT-9d)	T1 + T3 host
No results	any, with a query	Steam's "No Results Found" (#Search_NoResults) as centred Title 3 and the query echoed under it in Callout, in quotes (P-62)	T1 + T3

• The sheet: thick glass (PLAN §1.6), radius 44, its top 14 px under the field so it reads as growing down out of the capsule, centred on the glass (P-35), ≤ 960 wide (PLAN §1.12); content inset 40 (zero state) or 32 (results).
• The field stays Steam's input in the toolbar row, above the scrim. While it holds DOM focus it carries the only focus ring in the system (3 px soft white .55 + 16 px glow, VP P-17); when focus moves into the sheet the ring goes. No microphone.
• Close control: the window's Back circle at (24, 24), above the scrim; after 0.6 s it grows into a capsule naming the page you return to (§3.2). It is the sheet's close circle in the sense of PLAN §1.12 and VP P-66, so the sheet has no second one.
• A click on the dimmed page closes search. This deviates from VP P-65 ("an outside click does not close sheets"): the search sheet hangs from its field like a popover, closing it loses nothing (the query is kept in Recent Searches), and PLAN §1.9 adopts this section's leaving paths. V2 records it as a known deviation.

4.5 Zero state (880 × 500)

Shot: p2_window-nav_search.png.

• Recent Games (#LibraryHome_RecentGames): five cells of 160 × 180 on a 160 px pitch; each a 120 px game disc (DESIGN2 §9.2: hero, logo at 70 %, specular arc, contact shadow) with a one-line Subheadline label (HA-5). Inside the sheet the disc is content, not glass (no glass on glass, P-45). Data: Steam's recentAppsCollection (rt.react.data.recentGames(5)), shortcuts included as in Steam's own list. A or a click → the game page (rt.react.actions.navigate). Focus: the Home disc lift ×1.10 + shadow; no pop (PLAN §1.7: < 150 px).
• Recent Searches (English only, PLAN §1.15; in other languages the header is omitted and the capsules stay): this session's queries (≤ 8, newest first) as 60 px capsules with a clock glyph, 24 Semibold, 24 px apart. Kept in SharedJSContext memory only, never on disk or in localStorage; lgs off or a Steam restart empties the list and the section hides. A or a click sets the query through Steam's own search store.
• See All (#StoreApp_SeeAll), a 60 px capsule with a grid glyph: Steam's own page for the empty query (today's "list everything") in the sheet.
• Hint (English only): Callout 22 px, white .70: "Type to search your library, friends and the Store."

4.6 Results (960 × 500)

Shot: p2_window-nav_results.png.

• The scope bar at the sheet's top: Steam's categories (#SearchTab_All, _Library, _Friends, _Store, _Tools, _Hidden) as one 64 px segmented control, segments 56 tall and ≥ 120 wide (E-SEG compact), each followed by Steam's count in white .70. Hidden appears only with matches, as Steam does; Friends shows its 0 as Steam does. Steam's own paging arrows ‹ › sit at the sheet's ends as 60 px plain circles (80 px hit). LB / RB step through the categories. The white pill travels on snappy.
• All = the summary, when Steam's own search results are available to T3 (Q6: a finder for Steam's search hook through rt.react.find). Items, order and counts are Steam's; the shell never matches games itself.
  • Row 1: the Top Hit, 372 × 170: hero art, the name in Title 3, Steam's status line (Playtime: %1$s hrs, Not installed, or Steam's display status), an "Open" capsule (60 px, #Generic_Open) that is part of the card's single target, and the More circle inside its top right (PLAN §1.11). Then Steam's next four library matches as 113 × 170 posters in Steam's order. Posters show art only at rest; the focused or hovered poster shows its name and status under it (DESIGN2 §7.8).
  • Top Hit rule: among Steam's library matches of the best match tier (exact title > title prefix > word prefix > substring), the most recently played; ties keep Steam's order. A program (S-A) becomes the Top Hit only with a strictly better tier.
  • Row 2: Steam's own header "Store (%1$s)" (#Search_Results_Header_StoreApps_With_Count) over Steam's first store results as 180 × 84 capsule cards, then a See All capsule that opens Steam's Store tab in the sheet, where "View more in the Store" (#GamepadHome_GoToStore) stays the path to the store search. When programs match (S-B), the Software cell (header #AppType_2; 300 × 84; the best program's icon on a 56 px disc and its name; "and N more" in English only, otherwise "+N") takes the row's start and the store keeps two cards.
  • Below the fold, in Steam's order, one section per other category with matches, under Steam's own headers with counts (Friends (%1$s), Tools (%1$s); Hidden as #SearchTab_Hidden with its count), holding Steam's items. The sheet scrolls under a 26 px fade (P-73), with no scrollbar (P-72); the next header peeks above the fade.
• All without Q6: the sheet hosts Steam's own All grid (every result Steam shows today, with Steam's counts in its tab row). Nothing is hidden or recounted by the shell.
• Categories always host Steam's own tabbed page (steamChildren) in the sheet: Steam's virtualised grid, its tab row restyled as the scope bar, "View more in the Store". Hosting Steam's page in a smaller container is [PLAUSIBLE]: the grid measures its container. Gate AT-9d; fallback: Steam's page full-window, restyled (T1), and the sheet only for the zero state and All.

4.7 Keyboard and echo

The keyboard is SteamVR-placed under the window and built by C4b. Search relies on its echo row (SM SQ4): a display-only mirror of the field on the keyboard's slab, so the typing glance is about 5° instead of 22–28°. If SQ4 fails, the field still updates live in the toolbar row.

4.8 Paths

Step	Laser	Gamepad
Start	Click the capsule (or the 60 px magnifier circle)	D-pad Up from the top row of any page (Main's onMoveUp)
Type	Point at keys	D-pad over keys + A
Leave the keyboard	Done, or click the window	Done / B
Into the results	Point	Down from the field enters the sheet; focus lands on the Top Hit (zero state: the first Recent Games disc), P-22
Switch category	Click a segment or ‹ ›	LB / RB; or Up to the scope bar, Left / Right + A
Open	Click a card, poster, disc, store card or See All	A
Play a library result (S-C)	"Play" in the ornament (acts on the frozen target), or the More circle → Steam's menu → Play	X (Steam's primary action)
Item menu	The More circle on the attended item, or "Options · " in the ornament	☰
Clear	The × mini circle (44 px, E-MINI) in the field	Backspace on the keyboard
Repeat a search	Click a Recent Searches capsule	Focus + A
Leave search	The Back circle, a click on the dimmed page, any tab	B (our page's onCancel → NavigateBack)

4.9 Ornament, depth, motion and sound

• Ornament (§3.4, PLAN §1.10): our Focusables declare Steam's own action descriptions: A "Open" (#Generic_Open), B "Back" (#ActionButtonLabelBack), X "Play" (#GameAction_Play) on library items, ☰ "Options" (#ActionButtonLabelContextMenu). Zero state: A and B only, so the quiet legend in the margin. Results with a library item attended: X and ☰ join, so the 84 px capsule. Glyph badges only in gamepad mode. Steam's legend nodes are never hidden.
• Depth (PLAN §1.7, default profile): the sheet is a +10 mm non-interactive crop over a thick slab, 0 → +10 on depth with sheet-in. A focused or dwelled content card in the sheet whose long side is ≥ 150 px (Top Hit, posters, store cards) rises to +15 mm, capped by the click-safe rule on its smallest focusable (the More circle's 80 px hit allows +15). Recent Games discs, the Software cell, capsules and the scope bar never pop. Nothing pops while the sheet scrolls; the snapshot never pops. Wearer profile (interactivePops): the sheet +30 → +50, interactive.
• Native mode: the snapshot, scrim and sheet content are Steam's DOM, so they reach the headset through the base mosaic and the sheet's crop. CSS-only: in-page blur 30 + tint .40 + the edge lobe + shadow (§8.4); nothing inside the sheet has its own backdrop-filter (P-45): the Open capsule and the More circle are fills there.
• Motion: present on sheet-in 735: glass from the field (scale .97 → 1, origin at the field's centre-bottom), content 25–70 %, scrim 0 → .35 on fade, z 0 → +10 on depth. Zero state → results: width 880 → 960 on snappy, sections cross-fade on fade. Dismiss: leaving is Steam's NavigateBack, and React unmounts our route after Steam's ≤ 200 ms exit (PLAN §1.5), so content and glass fade out in 150 ms linear; glassd dissolves the slab in native mode. Reduce Motion: fades of ≤ 200 ms only. Nothing runs at rest.
• Sound (PLAN §1.13): Steam's ShowModal / HideModal on present and dismiss, ChangeTabs on a scope change on our All page (Steam's own tab row sounds itself); nothing on hover.

4.10 Implementation and fallbacks

• rt.react.routes.override(Routes.Search.Root(), (steam) => jsx(LgsSearch, {steam})) (P2 §4.2) in device/rt/21-search.js, module search, flag wp.c1b.
• LgsSearch reads the query from Steam's search store and re-renders on its change. It renders inside rt.react.ui.Page (onCancel = NavigateBack) with ui.ErrorBoundary whose fallback is steamChildren: the snapshot + scrim (a plain node; a click calls NavigateBack), then the sheet: zero state, the All summary (Q6), or steam for categories, See All and All-without-Q6.
• HA's contributions go in through __LGS_RT.search.addProvider (spec: docs/phase2/wp/C1b.md, "Interface announced"). Every launch and jump goes through rt.react.actions, which logs instead of running in test mode.
• The snapshot lives in a module-level variable and is dropped when LgsSearch unmounts or on lgs off.
• Fallback 1 (no page root found, e.g. a deep link): the sheet sits on plain window glass; the previous page's title shows dimmed in the toolbar row.
• Fallback 2 (T3 override unavailable, or the runtime off): Steam's search page, restyled by T1 (theme/21-search.css): the scope bar, title-case labels, Steam's tiles; row heights unchanged because the grid is virtualised.

Line edits elsewhere in WN (same request):

WN place	Change
Mockup table, rows p2_window-nav_search.png and _results.png	"Search activated: a sheet over the dimmed Library (real art), zero state with Recent Games and Recent Searches, keyboard with echo row; laser mode, quiet legend" / "Results for "half" in the 960 px sheet: the scope bar with Steam's counts, the focused Top Hit (gamepad mode), Steam's next library matches, the Software cell and Steam's store results under Steam's section headers"
p2_window-nav_motion.png row	C1c's storyboard no longer shows the search sheet; remove "search sheet present" from the description (C1b's G-MOTION filmstrips shots/p2_motion_c1b_* cover it at M3)
§1 "The experience", search paragraph	"As you type it becomes results: one scope bar for Steam's categories with Steam's counts, a Top Hit card, your matching games, Steam's first Store results, and a section for every other category with matches." (was "an entry for every other category")
§2 verdict table, Search row	Redesign column: add "scope bar; no-results view that echoes the query"
§3.7 depth table	"Sheets (incl. the search sheet)" row is right; the "Window dim during alerts and sheets: t1 tint .65" row is withdrawn by PLAN §1.8 (scrim .35 only)
§7 motion table	"Search sheet dismiss: sheet-out 514, content first" → "Search sheet dismiss: content and glass fade out in 150 ms linear inside Steam's ≤ 200 ms route exit; glassd dissolves the slab"
§8.1 and §8.4	Add to the search sheet row: "no element inside the sheet has its own backdrop-filter (P-45)"
§9.1	Replace the S rows and H2–H4 with the table under "Function retention" below
§10 tier table	"Search sheet, zero state, Top Results" row → "Search sheet, zero state, All summary, providers"; evidence adds "provider API (C1b), Q6 finder through rt.react.find"
§11 Q6	Default stays "All = Steam's All grid"; owner C1b at M3
§12 deviations	D-17 stays. Add D-18 "DESIGN2 §3.7 sheet width ≤ 75 % of the glass (960): results sheet 960, not WN rev 2's 1040 (PLAN §1.12)"; D-19 "DESIGN2 §7.5 search sheet rows of 64 px: the zero state uses Home's discs and 60 px capsules; results use posters and cards (content)"; D-20 "WN rev 2's column of category capsules (54 px on a 58 px pitch) replaced by the scope bar's counts and Steam's own section headers (PLAN §1.3 sizes)"
§13	Replace AT-9a to AT-9e, AT-10 and AT-16 with the rows under "Acceptance tests" below (ids kept)

Acceptance tests (C1b), updated to PLAN §1

Every live step runs inside P10's locked commands with --flags wp.c1b (and wp.c2a for the provider rows), --mode laser|pad as stated, FocusApplicationRoot() before any D-pad sequence. Launch-type calls are logged by rt.react.actions (test mode), never executed. The keyboard is hidden again with SteamClient.OpenVR.Keyboard.Hide() in the same step. After a synthetic hover the pointer goes to (1400, 900). Typing test queries is allowed; no store purchase flow is ever opened (store cards are never activated with A).

Id	Test	How	Pass
AT-9a	Trigger, laser	--mode laser, route AllGames: L.click('main','%{SearchBox}'), then focus() on it; after 1.2 s read L.route(), .lgs-snap, document.activeElement, keyboard state; shot main p2_c1b_at9a; hide the keyboard; L.back()	Route /search…; the snapshot has 0 [id] and is inert; sheet rect 880 × 500 at (200, 100) ± 2; the field shows its ring; no glyph badge visible; records whether Steam navigated by itself
AT-9b	Trigger, gamepad	--mode pad; focus the first poster; L.pad('up') (A on the field only if it took focus without the sheet; then hide the keyboard)	The zero state appears on the real gamepad path; records Steam's own behaviour (resolves the shell.md / library.md disagreement)
AT-9c	Leaving restores context	After AT-9b: B (DispatchVirtualButtonClick(2)) while .gpfocus is inside the sheet; L.route(), L.focused()	Back on /library/tab/AllGames, focus on the same poster; no .lgs-snap, no sheet node
AT-9d	Steam's category page in the sheet	SEARCH('half','Library'); outline main; L.pad('down', 3); LB, RB	%{TabContentsScroll} inside the sheet (≤ 960 wide); rows render (virtualised); focus moves; LB / RB switch categories. Otherwise the full-window fallback is active and recorded
AT-9e	Parity and no results	For 'half', 'a', 'zzqxjvkw': counts per category in our scope bar against Steam's own tab row (override off, same lock)	Equal for every category; every category with matches has a segment and, on All, a section or row; 'zzqxjvkw' shows "No Results Found" with the query echoed (P-62)
AT-10	Echo	C4b's echo test (SM SQ4) with the sheet open	The echo shows the field's text and placeholder
AT-16	Performance	glass.py perf main --route /library/tab/AllGames; the same with a --pre that opens and closes the sheet 5×	Themed fps within 5 % of stock; no frames > 34 ms; the snapshot costs ≤ 1 long frame per activation
HA AT-13	Providers (with C2a)	AT-9 steps with 'half', 'vlc', 'liquid', --flags wp.c1b,wp.c2a	'half': the Software cell lists "Half SBS Toggle", A on it logs launchNonSteam with Steam's strCmdline; X on the Top Hit logs primary(546560). 'vlc': the Top Hit is the program "VLC media player" and Open logs its launch. 'liquid': no Top Hit or Software cell for Liquid Glass
PLAN-1b-1	Looks like the mockups	glass.py cmp window-nav-search.html LIVE (zero state) and window-nav-results.html LIVE (results 'half', Top Hit focused, --mode pad), with docs/phase2/wp/C1b-cmp.json	Named rects (data-id in the mockups) within ±8 px or the difference explained here; both images viewed and a verdict recorded
G-PAD	Gamepad traversal	glass.py pad-bfs --route /search/tab/All --pre <type 'half'>; and from AllGames: Up to the field, Down into the sheet, B	Every item reached; Down from the field lands on the Top Hit (zero state: the first disc); Down-Up and Right-Left return; B closes search and focus is on the poster you left
C1b-1	Sizes, type, outlines	glass.py gates main --route /search/tab/All --pre <type 'half'> in both modes; and with the zero state	G-SIZE: every target ≥ 60 visible with an 80 px hit (segments E-SEG, the × E-MINI); G-TYPE: no text < 18 px, no weight < 500, no uppercase or tracking in the sheet; G-OUTLINE: no border or outline; edge ratio ≤ 0.35 on the sheet's top edge (WN AT-23)
C1b-2	Ornament	outline main with the zero state, then with results and the Top Hit attended, in both modes	Zero state: #Footer[data-lgs-orn="quiet"] with A and B only; results: the capsule with X "Play" and ☰ "Options"; no legend node with display: none; glyph badges only under data-lgs-vr-mode="gamepad"
C1b-3	Focus and hover	glass.py focus main --pairs on a scope segment (rest, focused, selected), a Recent Searches capsule, See All, the Top Hit; laser hover with --mode laser	G-FOCUS (PLAN §1.4): focus ≥ +40 L over rest, focus ≥ selected + 15 L, glow band ≥ +20 L on the white segment; the Top Hit and posters scale 1.05 on focus and after 80 ms of laser dwell; capsules and segments never scale
C1b-4	Depth (native)	glass.py native-session --step "sgcheck --pre <open sheet, type 'half', focus the Top Hit>"; then the same while the sheet scrolls	The sheet's crop at 10 mm ± 2 (live S × r), interactive: false; the Top Hit at +15 (or capped by the click-safe rule, reported); ≤ 4 distinct dz; no pop while scrolling; no pop for the snapshot, discs or the Software cell; nothing beyond +15 unless interactivePops
C1b-5	Motion	glass.py motion main --pre <activate> --name c1b_sheet_in; --name c1b_results (first keystroke); --name c1b_dismiss (B)	Durations and easings are tokens; present ≤ 800 ms incl. delay; dismiss ≤ 200 ms; glass before content on entry, content before glass on exit; nothing at rest after 1 s; --media reduce: fades ≤ 200 ms only
C1b-6	Recent searches stay in memory	Three searches; then glass.py js reads localStorage / sessionStorage keys on main; frame_ssh grep -r for the queries under /tmp/lgs, /dev/shm/lgs, ~/.local/share/glass-shell; then lgs off, lgs on --css, open search	No query found in storage or on disk; after the restart the section is hidden
C1b-7	Removal	lgs off; navigate to /search/tab/All with a query	Steam's own search page, unpatched; no lgs node, snapshot or provider; rt.react.status().patchedLeft 0
C1b-8	Strings	Static review of 21-search.js: every drawn string goes through rt.react.ui.loc() or the en* gate; DOM: drawn strings on main equal "Strings" below	Equal; no other literal text
C1b-9	Sounds	Intercept PlayAudioURL (IM §9); open and close the sheet; LB / RB on All; hover a card	ShowModal and HideModal once each; ChangeTabs per scope change; nothing on hover

Function retention (search, WN §9.1; complete for audit/shell-nav.md A.1 H2–H4, A.2 S1–S4 and audit/library-apps.md A.7, plus HA §12.6)

#	Function	New place	Laser	Gamepad	Tier	Test
H2	Start a search	The field: 520 / 640 × 64 capsule, or a 60 px magnifier circle on the routes of PLAN §1.9	Click (DOM focus → the sheet)	D-pad Up from a page's top row (same)	T1 (C1a) + T2 + T3	AT-9a, AT-9b
H3	Type / edit the query	Steam's VR keyboard with the echo row (C4b)	Click keys	D-pad + A (same)	C4b	AT-10
H4	Clear the query	The × as a 44 px mini circle inside the field (E-MINI)	Click	Backspace on the keyboard (same)	T1	C1b-1
S1	Switch result category (All, Library, Friends, Store, Tools, Hidden, with counts)	The scope bar in the sheet; Steam's ‹ › as 60 px circles	Click a segment or ‹ ›	LB / RB (same); or Up to the bar, Left / Right + A	T3 (All), T1 (Steam's tab row)	AT-9d, AT-9e
S2	Open a result (game → its page, friend → profile, store item → store page)	All: Top Hit, posters, store cards, friend circles; categories: Steam's grid tiles	Click	D-pad + A	T3, T1	G-PAD, HA AT-13 (store cards are never activated in tests)
S3	"View more in the Store" (store search)	Steam's Store tab in the sheet, its last tile unchanged; reached from See All or the Store segment	Click See All, then the tile	Focus + A	T1 / T3	AT-9d (look only)
S4	Leave search	The Back circle, the dimmed page, any tab	Click	B (same)	T1, T3	AT-9c
LA S4	No results	"No Results Found" + the query echoed	Look	Look	T1 + T3	AT-9e
S-T	See Friends, Tools, Hidden and Store results from All	Scope segments with Steam's counts; a section per category on All; or Steam's All grid without Q6	Click	D-pad + A	T3 / T1	AT-9e
—	Empty query lists everything	Zero state "See All"	Click	Focus + A	T3	G-PAD
NEW-1	Reopen a recent game	Recent Games discs	Click	Focus + A	T3	G-PAD
NEW-2	Repeat a recent search (session memory)	Recent Searches capsules	Click	Focus + A	T3	C1b-6
S-A	Find a program; a program as the Top Hit	Top Hit card (program), Open	Click	A	T3 (C2a provider)	HA AT-13
S-B	Launch a matching program; reach more matches	The Software cell; "and N more" → Home › Apps with it focused	Click	Focus + A	T3 (C2a provider)	HA AT-13
S-C	Play a library result	"Play" in the ornament (frozen target); the More circle → Steam's menu → Play	Click	X (Steam's primary action)	T3	HA AT-13
L8 (search)	A result's item menu (Steam's tile menu)	The More circle inside the attended card; "Options · " in the ornament	Click	☰	T2 (C1a helper)	WN AT-26 on /search
—	Focus returns to the poster you left	Steam's focus history under NavigateBack	—	B	—	AT-9c

H1, H5, H6 and H7 stay C1a's (WN §9.1).

Strings (PLAN §1.15)

Found by source text in Steam's English UI strings on the Frame (~/.local/share/Steam/steamui/localization/*english*, read-only grep, 2026-10-07, build 11094443).

Text	Token	Where
Recent Games	#LibraryHome_RecentGames	Zero state header
Recent Searches	none: English only; omitted elsewhere	Zero state header
See All	#StoreApp_SeeAll	Zero state capsule; Store row
Type to search your library, friends and the Store.	none: English only; omitted elsewhere	Zero state hint
All, Library, Friends, Store, Tools, Hidden	#SearchTab_All, _Library, _Friends, _Store, _Tools, _Hidden	Scope bar
Store (%1$s), Friends (%1$s), Tools (%1$s), Library (%1$s)	#Search_Results_Header_StoreApps_With_Count, _Friends_With_Count, _Tools_With_Count, _OwnApps_With_Count	Section headers
Software	#AppType_2	Software cell header (S-B)
and %1$s more	none: English only; "+%1$s" elsewhere	Software cell
Top Hit	none: English only; no tag elsewhere	Top Hit card
Open	#Generic_Open	Top Hit capsule; A legend
Play	#GameAction_Play	X legend
Options	#ActionButtonLabelContextMenu	☰ legend
Back	#ActionButtonLabelBack	B legend
Playtime: %1$s hrs / Playtime: %1$s min	#BasicGameCarousel_TotalPlayTime_Hours, _Minutes	Top Hit and focused-poster status
Not installed	#BasicGameCarousel_NotInstalled	Status
Steam's display status (Update Required, Downloading, …)	#DisplayStatus_*	Status
No Results Found	#Search_NoResults	No-results view
View more in the Store	#GamepadHome_GoToStore	Steam's Store tab (unchanged)

Requests

• REQ C1b->C1a: #1 merge the search text: replace WN §4 with "Proposed WN text" in this file, apply its "Line edits elsewhere in WN" (mockup table, §1, §2, §3.7, §7, §8.1/§8.4, §9.1, §10, §11 Q6, §12 D-18 to D-20, §13 AT-9a–e / AT-10 / AT-16). Why: C1b builds §4 but does not own the file (your table at the top of WN rev 3 already assigns §4 and §9.1 H2–H4 + S rows to C1b). C1a answer (2026-10-07, session 2): merged. WN §4 is your text; line edits applied (mockup table, §1, §2, §6.2 B label, §7 dismiss and 880 → 960, §8.1/§8.4 P-45 notes, §9.1 with an Owner column — H3's tier set to T1–T3 and AT-3 added to H4 because the field is C1a's —, §10, §11 Q6, §12 D-18 to D-20, §13 AT-9a–e, AT-10, AT-16 plus HA AT-13, PLAN-1b-1, G-PAD and C1b-1 to -9, owner C1b). §3.7's t1 row was already withdrawn in revision 3.

• REQ C1b->C1a: #2 shared builder (window-nav-shared.js / .css) to PLAN §1.10, §1.4, §1.11: (a) buildOrnament keeps A and B in laser mode as quiet trailing members; (b) when the legends hold only A and B it renders the quiet legend (no capsule material, 60 px items, Medium white .70, top: 658px in the margin on window routes) instead of removing the ornament; (c) glyph badges only in gamepad mode (data-mode="laser" → labels only); (d) buildMore 60 visible / 80 hit, black .38 + 10 px blur, and no blur when the host sits inside another glass container such as the search sheet (VP P-45). Why: C1b's zero state draws the quiet legend inline today (.wn-quiet, the same values as C1c's recipe); once the builder does it, C1b switches to data-wn="ornament". C1a answer: done. (a)–(c) were in the builder since session 1; (d) data-inglass on data-wn="more" drops the blur (P-45). data-wn="ornament" with only A and B renders the quiet legend: in the margin on the dim band (see #3), or inside the glass with data-inglass (window-full). The items sit at y 640–700 (the 84 px band from 628, WN §3.4.1), not at a 658 top.

• REQ C1b->C1a: #3 quiet-legend legibility (finding for you to take to the coordinator, since PLAN §1.10 fixes the look). In shots/p2_window-nav_search.png the quiet "Open" sits on the bright room window: background luma p10 181 / p50 209 / p90 230 behind the label; white .70 text there is about 1.1 : 1, far below VP P-39's 4.5 : 1, and the on-room shadow (P-41) does not rescue it (even white 1.0 is about 1.5 : 1). Suggestion: a dark dimming fill under each quiet member (black .35, 60 px capsule, no rim, no blur: a fill, not glass material), or quiet members sitting inside the glass on window routes. Measured with PIL on the render; the same holds on the device wherever the room is bright behind the margin. C1a answer: decided under PLAN §1.17's rule (WN §3.4.1, §11 O3): on window routes the quiet members sit on a dim band (black .62, 60 px, blurred edges, not glass, no slab), labels white .82; flag quietBacking ("window" default, "off" = §1.10's bare look). Measured 5.1–6.7 : 1 over the curtain (L 236) and the sill; bare 1.3–1.5 : 1. Note to the coordinator filed in C1a's log.

• REQ C1b->C1a: #4 route map for /search and /search/tab/*: data-lgs-glass="window", a route key data-lgs-route="search", Back in the nested style, no Large Title in the toolbar row (the snapshot carries a clone of the previous page's title under the scrim), the field in its 640 px variant. Please publish the Large Title node's selector (e.g. .lgs-title) in your interface table so C1b's snapshot can clone it, and confirm the Back circle's 0.6 s titled capsule names the page NavigateBack returns to. C1a answer: done in WN §3.1.1 and §3.2: /search* = window, route key search, Back nested, no Large Title, the 640 field. The Large Title node is #header > .lgs-title (T2, aria-hidden; in C1a's interface table). Confirmed: after 0.6 s the Back circle names the page Steam's NavigateBack returns to (the previous history entry's title from the route map); without a title it does not grow.

• REQ C1b->C2a: #5 conform HA §6 (S-A, S-B, S-C), HA §12.6 and XC §2 to the provider API and geometry in "Interface announced" and the proposed WN §4.6: the Software cell (header #AppType_2, 300 × 84 at the start of the Store row, 56 px disc with the 48 px icon, name 24 Semibold, "and N more" English only) replaces the 704 × 76 bottom row with "Search the Steam Store" (Steam's "View more in the Store" stays on the Store tab); the Top Hit rule (a program wins only with a strictly better tier); your provider registers with __LGS_RT.search.addProvider({id: 'c2a.programs', slots: ['tophit', 'software'], rank, open, primary?}) from device/rt/41-search-apps.js. C2a, 2026-10-07: done: HA §6 (provider c2a.programs, slots tophit/software, tiers 0–3, strictly-better Top Hit, the Software cell, no primary since S-C is your sheet's), §12.6, AT-5(c), AT-13 and XC §2 conformed. Follow-ups for you in REQ C2a->C1b #5 (docs/phase2/wp/C2a.md).

• REQ C1b->C2a: #6 home-apps-search.html (route matrix row /search): it draws the results in the full window, which PLAN §1.9 withdraws (search is a sheet over a snapshot). Please either redraw it as the 960 × 500 sheet over the dimmed page with the Software cell (layout and data as in window-nav-results.html, which now shows "Half SBS Toggle" with its real icon), or retire it and point HA §6 at p2_window-nav_results.png. C2a, 2026-10-07: done: redrawn as your 960 × 500 sheet with your data for "half", gamepad focus on the Software cell, quiet legend (A, B); re-rendered shots/p2_home-apps_search.png.

• REQ C1b->C1c: #8 (withdrawn, see below) native depth for the search sheet (C1b's card is T1–T3 and owns no layer fragment; your theme/layers/22-presentations.json holds sheets at +10). Please add two rules, default profile non-interactive: (a) [data-lgs-search="sheet"] at +10 mm with a thick slab, 0 → +10 on depth with the present, wearer profile +30 → +50; destructive rule n/a; (b) inside it, a content card [data-lgs-search="tophit"], [data-lgs-search="library"] > *, [data-lgs-search="store"] > * at +15 mm on .gpfocus (gamepad) and .lgs-dwell:hover (laser), click-safe capped by P6, never while the sheet scrolls. The snapshot (.lgs-snap), the discs, the Software cell and the scope bar never pop. Why: PLAN §1.7 rows "Sheets, the search sheet" and "search result"; C1b-4 tests it. Not needed before C1b M3.

  • C1c (2026-10-07, session 2): accepted as written, both rules. They land in theme/layers/22-presentations.json with C1c's M4 fragment (next round, after P6's reporter v3); left open until then.
  • C1b, session 2: withdrawn, please do not add search rules. C1b wrote the sheet rule itself in theme/layers/21-search.json (PLAN §2.1 fragment naming, NN = 21; the coordinator's task for this round asks every area for its own fragments). Rule (b) is dropped: the reporter never keeps two overlapping pops (lgs_layers.js: "popped rects never overlap"), so a +15 card inside a +10 sheet cannot exist; the focused result rides on the sheet and keeps its CSS lift ×1.05.

• REQ C1b->P10: #7 glass.py cmp with two mockups per package: C1b's two states need different routes, --pre and mockup origins (zero state: window at (320, 30), the keyboard below it; results: (320, 66)). Please accept docs/phase2/wp/<PKG>-cmp.json as {"mockups": {"window-nav-search": {route, pre, mockOrigin, map}, "window-nav-results": {…}}}, selected by the mockup file name or --name. Both mockups already carry data-id on every named element.

  • P10 (2026-10-07 06:00): done. glass.py cmp accepts {"mockups": {"<stem>": {surface, route, pre, mockOrigin, mockScale, map}}}, chosen by --name or the mockup's file stem; top-level fields are defaults; a map value may be {sel, surface, mockOrigin, mockScale}; selectors take @text=, @last, @nth=N. A repeated data-id keeps its first occurrence. contracts/lab.md §3.

• REQ C1b->C1a: #9 PLAN §6 two-step move, step 2: theme/21-search.css now holds every search-route rule of theme/20-shell.css §6 (live since 2026-10-07 session 2 of C1b). Please delete §6 ("Search route (/search/tab/*): Phase 1 …") and the %{GamepadSearch} entry of §1's "background: none" list (21-search.css paints the T1 scrim on %{GamepadSearch} itself and clears it under html.lgs-c1b).

  • C1a answer (2026-10-07, R2 fix pass): done. theme/20-shell.css §6 (the Phase 1 search route) is deleted and %{GamepadSearch} is out of §1's "background: none" list; §6 is now a one-line pointer to 21-search.css.

• REQ C1b->P10: #10 pad-bfs recovery: on /search/tab/All (any theme state, also with every runtime flag off) the sweep presses Left at the leftmost item, focus leaves the main window (exit:none), recover() runs, and from then on takeDirect fails for every queued node (untakeable), so only 4 of 7 (zero state) or 4 of 34 (results) nodes are visited. A manual D-pad script through the same nodes works. Probably the focus is left in the frame menu after the exit and L.gpTakeEl cannot pull it back; please re-take FocusApplicationRoot() in the main window before the next takeDirect after an exit:* edge. Evidence: C1b log, session 2 (bfs runs with and without wp.c1b).

  • P10 (2026-10-07 session 5): done. Cause (probes 14:00–14:12 on /search/tab/All): Left on the field makes vr-null-tree Steam's active nav tree, the gamepad has left Steam's overlay, and neither FocusApplicationRoot(), D-pad presses, BTakeFocus nor activating main's tree bring it back; the session 4 sweep also counted a moment of focus that Steam undid at once as "back". Now, after an exit: edge, each step is counted only if main still has the gamepad 150 ms later: the opposite press, main's nav tree activated, the root, then Steam's own EnsureVROverlayVisible() (what its Navigate calls) with the root and Down + Up (overlay), and last Navigate(route, replace); a take that does not land activates main's tree and takes once more. Steam's tab-row scroll arrows (%{Arrows}, focusable: false in Steam's own nav node, laser-only) are listed as notFocusable, not unreached. Live with --flags wp.c1b,wp.p3,wp.c1a --mode pad: zero state PASS 7 of 7 (3 exits back by overlay, 14:16); results ("half") PASS 32 of 32 targets, 9 exits all back, 0 irreversible, the 2 arrows listed apart (16:02, <P10 scratch>/s5/bfs_search_half2.json). Before: 4 of 7 and 4 of 34. contracts/lab.md §4 pad-bfs.

• REQ C1b->Coordinator: #11 G-AUD on /search (review R2 M10): two relocations by design need a §1.16 line, or a ruling that they are not GONE. Neither loses a function; both are checkable by an agent.

  • (a) The zero state replaces Steam's empty-query "list everything" page (WN §4.5, PLAN §1.9): gates main --route /library/tab/AllGames --pre <activate> gives GONE 136 in both modes (C1b log, R2 fix pass: g_s_zero_*.json), every one a tile or tab of Steam's page for the empty query. The zero state's See All shows exactly that page, Steam's own, one press away.
  • (b) The All summary lays Steam's All results out as the Top Hit, posters, the Store row and sections (WN §4.6, Q6): GONE 50–51 on /search/tab/All for "half" (g_s_res_*.json), Steam's result tiles at their stock DOM paths. Each result is in the summary (library matches as C1b's cards with Steam's actions; store, friend and further library results as Steam's own tiles through Steam's own grid component) or one press away (See All opens Steam's Store tab, where "View more in the Store" stays). __LGS_RT.search.summary() lists where every one of Steam's result ids went (row1, store, rest, friends, storeTab, redirect), so a probe can check it against Steam's results; the counts in the scope bar stay Steam's.
  • Suggested line, "E-RELOC (search)": AUD's GONE only (HIDDEN, SHRUNK, UNCLICKABLE, CONTRAST never waived), on /search and /search/tab/All, while (a) See All opens Steam's own empty-query page and (b) every stock result id is listed by search.summary() and the ids under storeTab / redirect are on Steam's Store tab. The kill switch c1bSummary (below) brings back Steam's grid on All if you rule (b) out.

• REQ C1b->Coordinator: #12 FYI, PLAN §1.17: a new kill switch, c1bSummary (default on, read as on when unset): off gives Steam's own grid on All in the sheet (WN §4.6 "All without Q6"). Please list it in the register if you keep it.

• REQ C1b->P10: #13 lab/exemptions.json E-SEG: please add %{GamepadSearch} %{GamepadTabbedPage>Tab}. With the runtime (T2) C1b tags these segments data-lgs-exempt="E-SEG" (criterion PASS live: 120–123 × 60, 20 px labels, gap 0); in T1 (runtime off, the fallback of PLAN §5.2 rung 3) nothing tags them, and the same compact segments are then judged by P-08 alone.

• REQ C1b->P10: #14 AUD on Steam's virtualised result grid: on a category tab (and on All with c1bSummary off) one or two tiles are GONE only because Steam's virtualiser did not mount that row in the 392 px sheet (GONE 1 on /search/tab/Library, g_cat_pad.json; T1 shows the same, g_t1_pad.json); the D-pad reaches them (Steam's grid scrolls). Suggest: a stock record inside a grid with aria-rowcount whose aria-rowindex row the themed grid did not mount is listed as virtualised, not GONE, when the themed scroller can scroll that far (scrollHeight − clientHeight > 0).

• REQ C1b->P10: #15 motion, P-54 on exits that unmount: in p2_motion_c1b_r2f_dismiss* the dismiss fades (lgs-fade-out 150 ms, opacity only) read "starts 640 px from rest", because the f = 1 frame comes after the route has gone and the target's rect is 0 × 0. A target with no box in a frame is not a move; please skip it (or list it as gone).

• REQ C1b->C1a: #16 WN §4 to the R2 fix pass (C1b log, "R2 fix pass"): (1) §4.4: sheets at R2-6's place, the zero state 880 × 488 at (200, 108), results 960 × 488 at (160, 108); the page you left stays sharp under the .35 scrim and only what lies behind the sheet is frosted; (2) §4.6: the All summary as built (Top Hit 372 × 170 and four posters in row 1; the Software cell, "Store (N)" over Steam's own first store tiles and See All in row 2; the other library matches and friends below in Steam's own grid; "View more in the Store" on the Store tab); the scope bar's compact segments (20 px, E-SEG compact); (3) §4.8: B anywhere in the sheet leaves search at once; Down from the field reaches the scope bar first, then the Top Hit (P-22: the scope bar is the sheet's top row); the item menu from ☰ or the More circle returns focus to its tile; (4) §4.9: the dismiss plays first (content 150 ms, then glass and scrim, done by 190 ms) and Steam navigates back when it has played; the zero state cross-fades into the results (no second present); the present's swell starts at .985 (VP P-53); (5) §9.1: S-A, S-B, S-C and L8 (search) now have both paths (rows below in this log), and the H-rows stay; (6) §13: C1b-5 reads "dismiss plays before the route changes; ≤ 200 ms". Please also add D-21 "VP P-22: on /search the field's Down lands on the scope bar, the sheet's first row".

• REQ C1b->P10: #18 G-SIZE and C1a's More circle: on a focused card that hosts the More circle (PLAN §1.11: inside the host, top right, inset 10), the 80 px sampling box hits the circle: "lgs-all-hit … hit 91 % own, 9 % other (lgs-more lgs-more-show "Options")" (g_g_res_pad.json, the Top Hit focused); C2c's review R2 M10 found the same on square posters. The circle is the host's own control (it calls the host's onMenuButton); please count .lgs-more placed on a host as part of that host, as "part of its host" does for hosted glyphs.

• REQ C1b->Coordinator: #19 G-PERF on /library/tab/AllGames → search ("half", perf --ab stock, 20:05–20:15): ratio 0.871–0.921 with wp.c1b, 0.866 with the summary off, and 0.86 with C1b's runtime off (the theme alone, T1); review R2 measured 1.002 at 13:16. The drop is theme-wide on this route since then, not C1b's runtime (C1b log, test table). FYI for whoever owns the theme-wide perf (V2).

• REQ C1b->P2: #17 FYI (no change needed now): C1b and C2a both carry the same finder for Steam's library app menu (41-home.js homeFindAppMenu, 21-search.js searchFindAppMenu: #GameAction_DismissPlayNext + bFitToWindow). If you add it to P2's finders (M.AppMenu, or actions.appMenu(overview, anchor)), both can drop their copies.

Requests to C1b, handled

Session 1: none. Session 2 (grep -n "REQ [A-Za-z0-9]*->C1b:" docs/phase2/wp/*.md → three):

• REQ C1a->C1b (C1a.md: quiet legend in window-nav-search.html): deferred. This round runs in expedited mode (no mockup re-renders). Nothing on the device depends on it: the live zero state declares only A and B, so C1a's own ornament draws the quiet legend; the mockup switch to data-wn="ornament" is a one-line change for the next mockup pass.
• REQ C1a->C1b (C1a.md: PLAN §6 two-step move of 20-shell.css §6): done. theme/21-search.css §2–§3 restate every §6 rule with the Phase 2 look (tab-row capsule → the 64 px recessed track without blur, tabs → 56 px segments, arrows → 60 px circles, art radius, the focused-tile ring → the ×1.05 lift and contact shadow, labels, "View more in the Store", no results). While §6 is still in 20-shell.css my rules win (same selectors, later file, !important where §6 has it). C1a may delete §6 now.
• REQ C2a->C1b #5 (C2a.md: Software cell details (a)–(d)): confirmed as written: (a) the cell shows the best program that is not already the Top Hit, absent when none is left; (b) one target (open launches with one match, "and N more" goes to Home › Apps through rt.home.reveal); (c) a program Top Hit has no More circle and subtitle = #AppType_2; (d) with a program attended only A and B are declared. Status: the provider API is live in device/rt/21-search.js (__LGS_RT.search.addProvider / providers / candidates, Liquid Glass dropped, a throwing provider disabled, a slow one skipped); the cell itself renders with the All summary (Q6), which is not built this round: results are Steam's own grid in the sheet (WN §4.6 "All without Q6").

Log

2026-10-07, session 1: M0 conformance

• Start-of-session request check: grep -n "REQ [A-Za-z0-9]*->C1b:" docs/phase2/wp/*.md → no docs/phase2/wp/ directory at the start; re-run after it appeared: no matches.
• Read: PLAN §0–§2 (all), §3, §4.1–§4.6, §5–§7; WN (all, and C1a's revision 3 changes in the working tree); XC; HA §6, §12.6, §14; VP §3.2, §5, §6 (P-01 to P-89); D2 §3.5, §3.7, §5.3, §7.5, §7.8–§7.10, §9.2–§9.4, §10, §11.2–§11.7; contracts react.md (§0–§8), interaction.md (§1, §2, §5), lab.md (§1, cmp); C1c's mockup recipe (window-nav-menu.html working tree) for one illumination model across mockups.
• Strings on the Frame (read-only): frame_ssh.py put of a read-only lookup script to /tmp/lgs/, frame_ssh.py run "python3 /tmp/lgs/c1b_locgrep.py; rm -f /tmp/lgs/c1b_locgrep.py", the same for c1b_locgrep2.py (token prefixes SearchTab_, Search_Results_Header_, DisplayStatus_, Playtime_, AppType_, LibraryTab_). Result: the "Strings" table. Both scripts removed from the Frame in the same command; nothing else touched. Steam build: cat /tmp/lgs/steam-build.json → 11094443.
• Real data used by the mockups (no device access needed; docs/refs/library/* from C2c's fetch-library-refs.py, git-ignored, with title-card fallbacks): Recent Games = the first five of lists.json recent (DolphinXR, Halo: Combat Evolved VR (Native), Rise of the Tomb Raider, Balatro, Stardew Valley); library matches for "half" = app types 1 (9 games; Alyx and the HL2 VR mod installed), tools 90, 254430, 255470 (Steam's live count: Tools 3); programs: "Half SBS Toggle" (icons/half-sbs-toggle.png); counts All 162 / Library 8 / Friends 0 / Store 158 / Tools 3 / Hidden 4 and the first store capsules from Steam's own /search/tab/All page (shots/p2_audit_search_on.png, row 2, columns 3–4, cropped by CSS; no file copied).
• Mockups edited: docs/phase2/mockups/window-nav-search.html, docs/phase2/mockups/window-nav-results.html (changes listed in the conformance table).
• Renders:
  • python tools/mockshot.py docs/phase2/mockups/window-nav-search.html shots/p2_window-nav_search.png → 1 984 359 bytes. Viewed. Fixed after the first render: Steam's strings (Recent Games, See All, Back). Verdict: the page you were on reads through the .35 scrim; the field's ring is the only ring; discs match Home's vocabulary; every control 60 px; no glyph badges in laser mode. Open issue: the quiet legend's contrast over the bright room (REQ #3).
  • python tools/mockshot.py docs/phase2/mockups/window-nav-results.html shots/p2_window-nav_results.png → 2 120 667 bytes. Viewed. Fixed after the first render: store cards needed position: absolute; the "Tools (3)" peek moved under the 26 px fade. Verdict: the scope bar (Steam's counts) fits 960 with Steam's arrows; the focused Top Hit lifts ×1.05 with the More circle inside; Steam's section headers replace the category capsule column; the Software cell sits beside the Store row; the ornament is a capsule with glyphs in gamepad mode.
  • Contrast measurement for REQ #3: PIL on shots/p2_window-nav_search.png, luma (0.299 R + 0.587 G + 0.114 B) over x 880–945, y 705–735 (behind "Open") → p10 181, p50 209, p90 230; over x 985–1040 (behind "Back") → p50 81, p90 191.
• No theme or runtime code written (M0). No device state changed: no Steam, SteamVR, theme or native-mode action; the two scripts in /tmp/lgs/ were deleted in the same command that ran them.
• End-of-session request check: see the last line of this log.

2026-10-07, session 2: M2, M3, native fragment (expedited round)

• Start-of-session request check: three REQs to C1b (C1a ×2, C2a #5); answers in "Requests to C1b, handled". P7 reports M1 (not M3).

• M0 note (PLAN §1 conflicts found while building): (1) PLAN §1.7 lists both "the search sheet +10" and "search result +15", but the reporter never keeps two overlapping pops (device/lgs_layers.js "popped rects never overlap"); the sheet wins (it is the presentation) and the focused result rides on it with its CSS ×1.05 lift. (2) WN §4.3 navigates to Routes.Search.Root() on activation; Steam itself only navigates on the first keystroke and then replaces the entry when already on /search/tab/All (probe: AllGames → go(/search/tab/All) → two keystrokes → one back = AllGames; typing from /search itself was not probed), so activation goes to /search/tab/All and one B leaves. (3) Steam's search store keeps the last query and shows it again on /search; a fresh activation therefore clears a stale query within 1 s and before any keystroke (to Recent Searches, then the field emptied through Steam's own handler, what its × does), so "Activated, empty → zero state" (WN §4.4) holds.

• Live probes (locked glass.py js steps): focusing the field does not navigate (Steam); setting the field through the native value setter + an input event runs Steam's own onChange (route → /search/tab/All, results for "half": All 162 / Library 8 / Friends 0 / Store 158 / Tools 3 / Hidden 4); clearing it on /search keeps the route and empties Steam's store. %{GamepadTabbedPage} is position: relative 680 tall, %{TabContentsScroll} padding-top 58, tabs max-height: 22px (Steam's .Tab.HasAddon), header z 6000, #Footer z 7000.

• Built: theme/21-search.css (T1 + T3 hooks; draft identical in theme/_wip/), device/rt/21-search.js (module search, flag wp.c1b; draft identical in device/rt/_wip/), theme/layers/21-search.json (sheet +10 mm, thick slab, behind wp.c1b). Workflow each time: brace_error, copy, python glass.py check-theme (clean), python glass.py sync.

• Fixes after viewing shots: sheet height 720 → !important geometry (Steam's own height rule); segments 22 → 56 tall (max-height: none); the snapshot's z-indexed rows (library tab row, Sort/Filter) painted above the scrim and through the sheet → .lgs-snap is its own stacking context (z 0, isolate), scrim z 1, sheet z 2; DolphinXR's disc empty → art chain hero → portraits → header → monogram, each failing URL steps on; the zero state showed a stale "half" → the stale-query clear above; a field that kept DOM focus gave no focusin → a pointerdown on the field activates too.

• Shots viewed: shots/c1b_now_all.png (before: the tab row under the field, uppercase 12 px labels), c1b_t1_half2.png (T1 sheet), c1b_t3_half.png (T3 results over the snapshot), c1b_t3_zero3.png (T3 zero state). The soft grey capsule behind the second disc in the zero state is the library's white "All" segment seen through the 30 px blur (material behaviour, not a node of ours). c1b_lib_ref.png was taken while another agent's native session was on (CSS glass dropped, black in CDP shots); not ours.

• Backdrop finding (session 2, 09:30–09:45): the sheet's in-page backdrop-filter did not frost the snapshot (the page's text read sharp through the sheet). Probes: brightness(.15) added to the sheet's backdrop-filter left the page behind untouched, an opaque red glass hid it (so the stacking was right); the same on Steam's tabbed page, on our zero sheet and on a glass slab of ours portalled into %{BasicHome} outside Steam's transformed content. Same as C1c's open M2 finding over the zoo page. Fix: the snapshot itself is blurred (filter: blur(14px) saturate(1.15), static) and the sheet gets a denser tint under text (+ rgb(26 28 36 / .34) over the thick preset; both dropped under [data-lgs-pop] so glassd's slab is the glass in native mode).

• Structure after the fixes: no wrapper around Steam's page; the snapshot, scrim and the sheet's glass are portalled into %{BasicHome} > .lgs-search-layer (last child, z-index: -1, inert), so Steam's DOM paths are unchanged (AUD went from 108 GONE to the reflow-only GONE) and Steam's content paints above the glass. A click on the dimmed page is a document click listener (inside %{PopupBody>Content}, outside the sheet → NavigateBack). Arrows 80 px boxes with the light inset 10; segments 60 tall in the 64 track; results inset 14 px so Steam's 5 columns stay ≥ 85 % of the 6-column tile (177 × 82).

• Verification runs (all --flags wp.c1b unless noted; JSON in C1b's scratch): gates results pad / laser, zero laser / pad, T1 laser (no flags): SIZE, TYPE, OUTLINE, MOTION pass; AUD as in Status. Two earlier runs at 09:31 were contaminated by another agent's open context menu ("3dSen") and native session (SIZE checked 0, chrome CONTRAST 1.5) and are discarded. Scroll test: 25 tiles + "View more in the Store" reachable in the sheet. Path tests (pad and laser): see Status. Error hooks: none.

• Native: python glass.py native-session --flags wp.c1b --mode pad --pre <search "half"> --step "sgcheck --route /search/tab/All …" --step "hv c1b_sheet_native --look" (exit 1: C1a's legacy R2 pops; the hv frame was taken while the route re-entered and shows an empty slab, so a second session ran with only hv and a 45 s flag overlay pushed in its pre: c1b_sheet_native2 shows the sheet with content at +10 mm over the blurred page). Both frames viewed and deleted the same minute; Frame /tmp/lgs/hv-*.png: 0.

• End-of-session request check: grep -n "REQ [A-Za-z0-9]*->C1b:" docs/phase2/wp/*.md → the three handled above; new: REQ C1b->C1a #9, REQ C1b->P10 #10; REQ C1b->C1c #8 withdrawn.

2026-10-07, session 3: R2 fix pass (review C1b-review-R2.md)

• Start-of-session request check: grep -n "REQ [A-Za-z0-9-]*->C1b" docs/phase2/wp/*.md → open: REQ C1a->C1b (quiet legend in window-nav-search.html), REQ C2a->C1b #5 (Software cell), REQ Coordinator->C1b (R2-16, FYI). All three handled and marked [x] where they were filed (answers there); REQ C1b->C1c #8 marked [x] (withdrawn in session 2).
• State found: an earlier attempt of this pass had drafted fixes at 17:25–17:31 (E-SEG tags, the query echo, focus memory by position, the dismiss class, the Title clone's type, A/B-only discs, the fragment's hitTest: false / modal / scrim hole) and synced them, but logged nothing and verified nothing (its last device call timed out at 17:23). This session verified each, kept what held and replaced what did not (focus memory, dismiss timing, native look).
• Files changed (owned): theme/21-search.css, device/rt/21-search.js (drafts identical in theme/_wip/, device/rt/_wip/), theme/layers/21-search.json, docs/phase2/mockups/window-nav-search.html, window-nav-results.html, docs/phase2/wp/C1b-cmp.json (new), this log. python docs/phase2/fontkit.py --hooks (P4's generator, any package may run it, tokens.md §2) rewrote the generated host lists in theme/04-states.css for the summary's --lgs-ill hosts. Workflow each time: node --check, python glass.py check-theme (PASS), --hooks-check (up to date), python glass.py sync, python glass.py reload.
• Every live step: python glass.py … --flags wp.c1b,wp.p3,wp.c1a (+ wp.c2a where the Software cell is tested), --mode pad or --mode laser, native off unless stated; Steam build 11094443. Probe scripts and raw outputs: C1b's scratchpad c1b_f2/.

Findings, what changed, and the live evidence (Frame times, 2026-10-07; JSON and probe outputs in c1b_f2/):

Finding	Fix	Evidence	Verdict
M1 scope bar	Steam's 36 px side padding on the tab header row removed (the arrows sit at the sheet's ends, the track gets 800 px); compact segments (20 px Semibold labels, 16 px padding: PLAN §1.10's compact member, E-SEG compact), sized to their content and never shrunk; the count 8 px after its label (Steam's 8 px title margin moved into the segment's gap), 32 px before the next label; T2 tags each segment data-lgs-exempt="E-SEG"	a5.out (laser, "half"): six segments 120–123 × 60 from x 277 to 1004 in the 800 px row (240–1040), the arrows' 80 px boxes at 160 and 1040, the track's ::after insets 114.5 / 114.5 (P-36: equal); elementFromPoint at "Hidden 4"'s right end hits the segment. No results: five segments centred, 338–942. SIZE PASS with every E-SEG criterion met ("120 x 60, label 20 px (compact), gap 0") in g_none_laser.json, g_res_pad.json and the later runs	Fixed
M2 ring and light	The ring is gone with C1a's §6 (REQ #9, done) and a doubled-class box-shadow: none !important; each segment is the containing block of P4's ::after; the focused (= selected) segment adds --lgs-white-glow on itself as well (blurred, never a ring)	b7.out: focused "All 162" box-shadow rgba(255,255,255,.55) 0 0 22px 8px, ::after 120 × 60. OUTLINE PASS in every gate run, also with --flags wp.c1b alone (g_s_c1bonly_laser.json). G-FOCUS in the test table	Fixed
M3 Large Title clone	The clone takes the live title's computed font, size, weight, colour and box, inline	a5.out: the clone 46 px / 700 at (100, 14, 145 × 80), white .96. TYPE PASS in the zero state in both modes (g_f_zero_*.json)	Fixed
M4 focus after B	The card that Steam's gamepad focus last held on the page (its vgp_onfocus event; DOM focus does not follow every D-pad step) is remembered by identity (first class and the art's URL, else its text) and takes focus again through its own nav node after the leave, never by position (each take scrolls the grid); no such card: FocusApplicationRoot() (P-13). B anywhere in the sheet leaves at once (m11)	b7.out (pad): zero state, B → /library/tab/AllGames, focus on the card the Up path left from (1147940); results: Right to the next top-row poster (954280), Up to search, "half", Down to the scope bar and the Top Hit, X, ☰, B (the menu closes, focus back on the Top Hit), B → AllGames with focus on 954280 (g_same: true), no layer left; debug().cancels 2	Fixed
M5 discs declared X	Discs declare A and B only	b7.out: a disc gives #Footer "quiet Open Back"	Fixed
M6 native	search-sheet pops C1b's glass with hitTest: false (the glass is pointer-events: none in an inert layer), modal: true, hole.fill: "scrim"; the T1 rule modal and the scrim hole too; once the pop is acknowledged, the CSS tint, frost and shadow go and the page picture and scrim are cut out under the sheet's rounded rect, so the crop carries only the sheet's content over glassd's slab	Native session (below)	see below
M7 motion	(1) dismiss: B in the sheet, the Back circle and the dimmed page start it (content out on page-out 150 ms; glass and scrim on page-out after the 40 ms page delay) and Steam navigates back once the glass's fade has ended (animationend, 600 ms fallback); Steam's render of the returning page, which holds the main thread for about 0.5 s, then happens under an invisible sheet, over the picture of that same page, which stays until the route unmounts; (2) first keystroke: the picture layer is keyed and never remounts (no re-present); Steam's page fades in on fade, the zero state's content fades out over it; (3) the 880 → 960 morph is a scale transition of the 960 glass (never left / width), a cut under Reduce Motion; (4) the swell starts at .985 (P-53)	tl5.out (pad): close from the zero state: content 1 → 0 over 112–270 ms, scrim and glass 1 → 0 over 158–315 ms, then the route changes; from results the same (117–335 ms). pt4.out: scrim and glass stay at 1 through the first keystroke. motion (laser): c1b_r2f_results PASS; c1b_r2f_results_reduce (--media reduce) PASS, fades of 180 ms only; c1b_r2f_dismiss: tokens only (150 ms linear, delay 40), content before glass; its P-54 lines read targets that are gone in the last frame as moves (REQ C1b->P10 #15)	Fixed
M8 P-62, P-01 / P-02, P-06	The query is echoed under "No Results Found" (generated content from data-lgs-q; Steam's text node untouched); every .gpfocus rule in 21-search.css is scoped :not(.lgs-input-laser); the disc lift keys on .lgs-dwell:hover / .lgs-attend-80 under the laser	a5.out: ::after content “zzqxjvkw”, 22 px. grep -n gpfocus theme/21-search.css: every selector mode-scoped. Conformance in the test table	Fixed
M9 retention rows	S-C: Steam's primary action as X on every library result, labelled as Steam's tile menu labels it (Play, Install, …, actions.primaryInfo); L8: Steam's own library app menu on ☰ and on C1a's More circle (registered on [data-lgs-search-app]), focus back on the tile when it closes; S-A, S-B and the results capsule: the All summary	b7.out (pad): Top Hit ornament "capsule Play Options Open Back"; X logs primary(546560); ☰ opens Steam's menu "Half-Life: Alyx · Play · Add to Favorites · … · Properties… · Cancel", B closes it, focus back on the Top Hit. a5.out (laser): the More circle (80 px box) on a library result, labelled Options; a click opens Steam's menu for that game; B closes it. HA AT-13 in the test table	Fixed
M10 AUD	Not a code defect: the zero state (GONE 136) and the All summary (GONE 50–51) relocate by design; a category tab keeps GONE 1 from Steam's virtualiser (T1 the same)	REQ C1b->Coordinator #11 (an "E-RELOC (search)" line with a checkable criterion and search.summary()); REQ C1b->P10 #14 (virtualised rows)	Open with the coordinator
M11 status	Every card test re-run (test table); PLAN-1b-1 with the new C1b-cmp.json	Test table	—
m1 idle work	No timer runs while the sheet is open (Steam's × is read again after a click in the field's container)	rm1 idle counters	Fixed
m2 blurred page	The page stays sharp under the scrim; only what lies behind the sheet is frosted: the picture is blurred and darkened as one static layer (blur(14px) saturate(1.15) brightness(.8)) and a sharp copy (.lgs-snap-sharp) lies over it everywhere except under the sheet's rounded rect (the mask that native mode uses for the cut-out). A first version with the frost inside the glass bled the sharp tab row through the sheet and was replaced	p2_c1b_r2f_zero.png (first version), p2_c1b_r2f_focus_zero_0.png, p2_c1b_r2f_results_pad.png	Fixed
m3 entry focus	Down from the scope bar lands on the Top Hit (the summary's first row); from the field, Down reaches the scope bar first (the sheet's top row)	b7.out f_tile "lgs-all-hit"; REQ C1b->C1a #16 (D-21)	Fixed (deviation recorded)
m4 provider timing	Skip at SEARCH_RANK_MS (2 ms); providers get ctx.steamBest = {tier, appid}	at13b.out: steamBest {tier 1, appid 546560} for "half"	Fixed
m5 DolphinXR art	The disc steps through custom art first	a5.out: every disc has an image	Fixed
m6 mockups	window-nav-search.html draws the builder's quiet legend (REQ C1a->C1b); both mockups at R2-6's geometry with the compact scope bar and the Software cell as a capsule; re-rendered shots/p2_c1b_mock_search_r2.png, p2_c1b_mock_results_r2.png (viewed)	PLAN-1b-1 in the test table	Fixed
m7 keyboard after the dimmed page	That path hides Steam's VR keyboard and the field lets go of DOM focus	a5.out e_route AllGames, e_kbActive BODY	Fixed
m8 R2-6	Sheets from y 108, 488 tall: zero 880 × 488 at x 200, results 960 × 488 at x 160	a5.out	Fixed
m9 disc halo	Kept: Home's disc recipe exactly (41-home.css C3: shadow .40 and 0 0 26px 2px white .22); the lift reads --lgs-disc-lift	—	Disputed (Home parity, PLAN §1.4 "Home discs ×1.10")
m10 disc labels	Cells 160 × 180, the whole cell the target	cmp recent-0…4 PASS (dy −0.4)	Fixed
m11 two B	B anywhere in the sheet leaves (vgp_oncancel caught in the capture phase, inside the sheet only; menus and modals over it keep theirs)	b7.out	Fixed
m12 bookkeeping	REQs marked [x] where they were filed; REQ #8 marked withdrawn	this log	Fixed
m13 stale query in Recent Searches	Steam's stored query is cleared, never remembered	searchClearStale	Fixed
m14 All = Steam's grid	The All summary	below	Fixed

The All summary (Q6), built this pass. device/rt/21-search.js patches Steam's results list (a memo whose only prop is results, inside %{GamepadSearch}; optional, c1b.results) and lays All out as WN §4.6 asks, in Steam's order and with Steam's counts. Row 1: the Top Hit (372 × 170: hero art, "Top Hit" in English only, the name in Title 3, Steam's status line, the Open fill; the best library tier, the most recently played; a program only with a strictly better tier) and Steam's next four library matches as 113 × 170 posters (name and status under the attended poster, keyed on the input mode). Row 2: the Software cell (S-B, a 300 × 84 capsule; "and N more" in English, "+N" elsewhere), "Store (N)" over Steam's own first store tiles (rendered through Steam's own result grid component with Steam's scroll element, so each tile and handler is Steam's) and See All (clicks Steam's Store tab). Below: the other library matches, then friends, in Steam's own grid. Library cards carry X = primary, ☰ = Steam's menu and the More circle; programs only A and B. The first time the patches attach in an install, Steam's page is remounted once (memo props keep their identity). Kill switch c1bSummary.

Card tests after the pass (--flags wp.c1b,wp.p3,wp.c1a, + wp.c2a where noted; native off unless stated; build 11094443):

Test	Command / probe	Result
AT-9a/b trigger	js a5_laser.js --mode laser, js b6_pad.js --mode pad	PASS: a field click and Up from the top row open the zero state (880 × 488 at 200, 108), snapshot inert, 0 ids, 0 focusables, keyboard on activation only
AT-9c B restores	b7.out	PASS: one B from the zero state and from a results card; focus back on the card left (by identity), no layer left
AT-9d category in the sheet	gates … pre_cat.js --mode pad (g_s_cat_pad.json), pad-bfs results	PASS: Steam's tabbed page in the 960 × 488 sheet; LB / RB and segments switch; rows render
AT-9e parity, no results	a5.out, at13b.out	PASS: the scope bar is Steam's tab row (counts are Steam's: All 162 / Library 8 / Friends 0 / Store 158 / Tools 3 / Hidden 4); "zzqxjvkw": "No Results Found" + “zzqxjvkw” (P-62)
HA AT-13 providers	at13b.out, at13c.out (+wp.c2a)	PASS for the summary and S-C: "half" Top Hit Half-Life: Alyx (tier 1, last played), Software cell "Half SBS Toggle", X logs primary(546560), A logs navigate(/library/app/546560); "vlc": the program "VLC media player" is the Top Hit (no Steam library match); "liquid": no Liquid Glass anywhere. Not shown on the live D-pad path: A on the Software cell and on the program Top Hit (the probes' D-pad did not land on them; the cells are reachable in pad-bfs, su1.out). See the disclosure below
PLAN-1b-1 cmp	cmp window-nav-search.html --name window-nav-search (laser) and window-nav-results.html (pad, +wp.c2a) with C1b-cmp.json	Every content rect within ±8 (sheet 0; discs dy −0.4; headers, capsules, hint, Top Hit, Open, posters, Software cell, Store header and tiles dy +7, See All). The FAILs are by construction and listed in C1b-cmp.json _doc: hit boxes vs visible (Back, field, arrows, More circle), C1a's 84 px ornament rows, the picture layer covering 720 vs the mockup's 656, the recent-search row +8.6 (each capsule +2.5–3.3). Viewed shots/p2_cmp_c1b_window-nav-search.png and p2_cmp_c1b_window-nav-results.png: the live sheet reads as the mockup (the zero state's laser legend says Steam's "Select" where the mockup says "Open")
G-PAD	pad-bfs --route /search/tab/All --pre pre_zero.js / pre_half.js (--mode pad)	PASS zero state 7 of 7, results 22 of 22 (Top Hit, posters, Store tiles, See All, the rest, segments), 0 irreversible, Steam's ‹ › listed apart; B from the zero state leaves to AllGames
C1b-1 sizes, type, outlines	gates (13 runs)	SIZE, TYPE, OUTLINE, MOTION PASS in both modes on zero, results, no results, a category, T1, wp.c1b alone. One exception: SIZE P-08 on the Top Hit when C1a's More circle is shown on it ("91 % own, 9 % other (lgs-more)"): the circle is the host's own, inside it by PLAN §1.11 (REQ C1b->P10 #18; C2c's review R2 M10 saw the same). AUD: REQ #11, #14
C1b-2 ornament	b7.out	PASS: zero state quiet "Open Back"; Top Hit capsule "Play Options Open Back"; no legend display: none
C1b-3 focus	focus … pairs_res.json (f_res2.json), pairs_zero.json (f_zero2.json), pad	PASS: P-16 glow band +38.3 L on the white segment; P-14 Software cell +62.1, See All +101.4, Recent Searches capsule +99.7, zero See All +107.7. P-15 inside the white pill 0.0 (white .94 is the ceiling; focus selects in Steam's tabbed page, so the focused segment is always the selected one and the glow band carries the difference, as in C4a's segments)
C1b-4 depth	native session (below)	PASS
C1b-5 motion	motion ×6	PASS (see M7); the dismiss strips' P-54 lines are REQ C1b->P10 #15
C1b-6 memory	code unchanged since review R2's PASS (no storage access added)	PASS (review R2 m1.out)
C1b-7 removal	js r1_flagoff.js (rm1.out), js r2_stock.js --stock (rm2.out)	PASS: wp.c1b off with results open: no layer, snapshot, sharp copy, summary, data-lgs-search-app, override or c1b.* patch; patchedLeft 0; back on reinstalls. Stock: no lgs mark, Steam's page
C1b-8 strings	static review	PASS: new drawn strings: Steam's #SearchTab_Library, _Store, _Friends, #Search_Results_Header_StoreApps_With_Count, _Friends_With_Count, #BasicGameCarousel_TotalPlayTime_Hours/_Minutes, #BasicGameCarousel_NotInstalled, #ActionButtonLabelContextMenu, #AppType_2 (from the provider); English only: "Top Hit", "and N more" ("+N" elsewhere)
C1b-9 sounds	b7.out, a5.out	PASS: one show_modal per open, one hide_modal per leave by B, by the Back circle (it no longer plays twice) and by the dimmed page; Steam's tab sound on LB / RB; nothing on hover
AT-16 / G-PERF	perf main --route /library/tab/AllGames --pre pre_half.js --ab stock	FAIL, not C1b's runtime: ratio 0.871–0.921 with wp.c1b; 0.86 with C1b's runtime off (T1, the theme alone, same step, 20:15) and 0.866 with the summary off; review R2 measured 1.002 at 13:16. The drop is theme-wide on this route since then; C1b's frost layers were restructured to one filtered layer (no gain). Reported to the coordinator (REQ #19)
AT-10 echo	C4b's	not run (C4b)

Native session (python glass.py native-session --step … ×7, 19:32–19:41, ready in 2.6 s; back to CSS only: yes): n_probe2.js: zero state and results each one pop on main, search-sheet on the glass, 1320 × 732 / 1440 × 732 tex (880 / 960 × 488 CSS) at dz 0.0271 (10.0 mm), thick, modal, hole fill rgba(0, 0, 0, 0.35); the glass acknowledged (data-lgs-pop="self"): shadow none, tint ::after none; the picture and the scrim masked with the hole at the glass's rect; after B no pop left. sgcheck zero, results (Top Hit focused) and a category: "main: 1 pops, depths [0.0, 10.0] mm, modal ['search-sheet']", PASS ×3. hv --look zero, results and results at 30° off axis: the sheet stands in front of the window as glassd's thick slab with the content on it, no doubled element; off axis it sits visibly ahead of the window. G-HV verdict withheld (auto rect, exit 3, as for every hv without a wearer's rect). All three frames viewed and deleted the same minute; 0 left in %TEMP% and in /tmp/lgs on the Frame.

Disclosures.

• 19:4x, js p_at13c.js ("vlc", pad): two D-pad Downs from the field went to a Store tile, not the program Top Hit, and the probe's A then opened that store item's page (/steamweb), a page view only: no purchase flow, nothing changed; the probe's finally went back (route /library/tab/AllGames checked). Store tiles must never be activated in tests; the probe now presses A only after checking the focused element's data-lgs-search.
• perf --ab re-applies the step's flags in each subject run, and those overlays were not popped (the lab bug review R2 n1 describes): I popped mine (32 overlays with wp.c1b, or wp.p3+wp.c1a only, from my steps) at 19:5x and 20:16; wp.c1b is off and search not installed.

Device left (20:17): theme on, CSS only, no flag overlay of mine, wp.c1b off, __LGS_RT.search undefined, no .lgs-search-layer, no menu open, route /library/home (another agent's step), pointer parked by the lab.

• End-of-session request check: grep -n "REQ [A-Za-z0-9-]*->C1b" docs/phase2/wp/*.md → none open. New: REQ C1b->Coordinator #11, #12, #19; C1b->P10 #13, #14, #15, #18; C1b->C1a #16; C1b->P2 #17.