MichaelScottsman--LiquidAss-Frame/docs/phase2/wp/C6a-review-R2.md

C6a Steam Settings: review R2 (quick review)

Reviewer: independent agent, quick review (about 20–30 minutes, user-visible blockers and majors only; no native session, no removal sweep, no perf, no full conformance). Package card: PLAN §2.4 "C6a Steam Settings"; builder's log docs/phase2/wp/C6a.md (Status "READY: wp.c6a"). Steam build 11094443. Files under test (md5, unchanged since the builder's 10:53 save): theme/60-settings.css 56b5214c, device/rt/60-settings.js 01963e3f, theme/layers/60-settings.json f82b8617. Main route: /settings/system; main mockup: settings-system.html.

An earlier, deeper R2 pass on the same file hashes (13:58–16:05, cut off) is kept below as an appendix for its evidence. Only the findings in this top section count toward this verdict; the appendix's other items are outside a quick review's scope (polish, other routes, card scope and process) and are left to the builder and V1.

Status

Done (16:25–16:50). Device left: theme on, CSS only; every step used --flags and --mode (popped at the lock exit); no hover, no dialog or menu opened, no setting touched; no hv frame taken. Shots (CDP captures of the main window, no room imagery): shots/p2_c6a_q_sys_laser.png, shots/p2_c6a_q_sys_pad.png.

Verdict

fix. 1 blocker, 1 major. On the main route the package looks and works like the mockup in both input modes, and every gate passes; but destructive confirmations opened from Settings are never tagged, so in native mode they pop forward like ordinary alerts (PLAN §1.7 rule 4), and one of them restarts the device.

What was run

All with --flags wp.c6a,wp.c1a,wp.p3 (wp.p3 is the input-mode stub; wp.c1a gives the window and toolbar the page sits in). Each step's line says native=off.

Command	Result
gates main --route /settings/system --mode laser --shot p2_c6a_q_sys_laser (16:39)	PASS: MOTION, SIZE, TYPE, OUTLINE, AUD all 0 findings (AUD 13 exempt; SIZE skipped 1 obscured and 2 clipped rows that scroll clear)
gates main --route /settings/system --mode pad --shot p2_c6a_q_sys_pad (16:40)	PASS: the same five gates, 0 findings
pad-bfs --route /settings/system --mode pad (16:40, default 150 s budget)	Exit FAIL only because the sweep was TRUNCATED by its budget (46 nodes, 255 moves, 165 s). 0 irreversible moves. Entry focus: the System pill. 16 sidebar pages reached in order (System to In Game) with their content; the 38 unreached focusables are the sidebar's later pages (Compatibility to Developer) and the lower System rows, i.e. budget, not a dead end. Left from a sidebar pill leaves main for the frame menu and comes back (exit:none, back by overlay), as on stock in the earlier pass. B from the entry follows Steam's history
Static read: device/rt/60-settings.js lines 216–224, device/rt/22-menus.js tagDestructive (lines 437–450), theme/layers/22-presentations.json exclude	B1 below

Shots against the mockup

settings-system.html (side by side in the appendix's shots/p2_cmp_c6a_r2_system.png) against the two new shots: the same 400 px sidebar of 72 px pills with colour circles and group gaps; the centred compact hero (60 px grey circle with Steam's gear and a Title 1 "System"); recessed platters with C4a's value capsules and the "Check For Updates" capsule; the "Settings" Large Title with the Back circle; the quiet legend under the glass. Laser: the selected System pill is a lighter fill, no outlines. Pad: the focused and selected System pill is clearly brighter, with A / B glyphs in the legend. Text is legible; nothing overlaps or clips except rows scrolling under the bottom scroll-edge fade. Differences a user would see, none a break: no About / Updates drill-down rows (T3 not built; documented fallback is Steam's single long page), no System badge, the 520 px search capsule straddles the sidebar and the pane (C1a's, REQ C6a->C1a open), and a thin light streak just above the selected pill (polish, appendix m1).

Key functions

Laser: G-AUD PASS (no GONE, HIDDEN, SHRUNK or UNCLICKABLE on the page; the sidebar, the language and update-channel value capsules, Check For Updates, Hostname and the rows stay Steam's own nodes and hit-test as clickable). Pad: pad-bfs reaches every sidebar page it had time for, their content rows and the search field, with no irreversible move. C6a hides and moves nothing in the DOM (read of 60-settings.js: attributes and style properties only).

Findings

Blocker

• B1. Destructive confirmations opened from Settings are not tagged, so they pop +10 mm in native mode (PLAN §1.7 admission rule 4, "never a destructive confirmation"; SET §5 rule 3; reporter.md §1 gives data-lgs-destructive to C1c and C6a). C6a's tag pass scans only the page content: root.querySelector(S.sel.content) then button.DialogButton (60-settings.js lines 217–224), never the open ModalPosition. C1c's tagDestructive (22-menus.js 437–450) tags only Steam's .Destructive buttons and the confirm of a dialog opened from the Power menu. Settings' confirmations carry neither, so C1c's alert rule (22-presentations.json, exclude: ".DialogButton.Destructive, [data-lgs-destructive]") pops them like any alert:

  • Audio "Reset" (tagged on the page) opens a GenericConfirmDialog whose plain "Confirm" (default focus) resets the audio config and restarts the device (SteamClient.System.RestartPC(), webpack module 97849, read in the earlier pass);
  • System "Hostname" opens an alert whose "Change & Restart" is not tagged, although Settings_System_Change_Hostname_Set is in C6a's own label list (line 34): that button exists only inside the dialog;
  • Internet network details "Forget" (inside the dialog).

  The earlier pass measured it in native mode on these same file hashes: sgcheck /settings/system --pre <open Hostname> gave main: 1 pops, depths [0.0, 10.0] mm, modal ['c1c-alert']; sgcheck "passes" only because its destructive check reads the same missing tag. Fix (C6a T2): run the label scan on the top modal card as well, and tag the .Primary confirm of a dialog opened within a few seconds of a click on a tagged page button (as C1c does for Power), so C1c's existing tagDestructive flattens the card (data-lgs-plate="thick"). Recheck: open the Hostname and Audio Reset dialogs with --flags wp.c6a,wp.c1a,wp.p3,wp.c1c and close them by Cancel; in one native-session, sgcheck each with the dialog open: expect no c1c-alert pop.

Major

• M1. In native mode the hero stays popped while it scrolls under the toolbar row. Earlier pass, native probe on these file hashes: at scrollTop 24 and 40 on /settings/system the settings-hero crop stays at +10 mm (slab, hole, contact shadow) at CSS y 92 and 76, inside C6a's own scroll-edge fade and over the bottom of the search capsule; it drops only from 56 on. So after a small scroll a half-faded circle, cut by a strip of the search capsule, floats 10 mm in front of the page. SET §5: the compact hero "slides under the top edge". Fix: drop the pop as soon as the content is scrolled (scrollTop > 0: remove data-lgs-hero, or add a rest attribute the rule excludes), then sgcheck --pre <scrollTop 24>: expect 0 pops.

Not counted (seen, outside a quick review)

• The pad-bfs FAIL is budget truncation only (above); a full-budget run against --stock is V1's.
• Gates fail on other settings routes (type on Bluetooth, Game Recording, Home, Family, Developer, Friends; size on Storage, Family, In Game, Notifications): appendix §1.1, M3–M5. Not the main route.
• Steam's own 86 px page slide between sidebar pages is untouched (appendix M1): stock motion, not a break.
• T3 (drill-down, list page, P-S2), the System badge, the inline title and the M0 re-render are not done (appendix M7, M8): the builder documents fallbacks; scope, not a user-visible break.

Requests

(none: this review edits no other file; B1 and M1 are in C6a's own files.)

---

Appendix: earlier R2 pass (interim, cut off at 16:05; same file hashes)

Reviewer: independent agent (review R2). Package card: PLAN §2.4 "C6a Steam Settings"; builder's log docs/phase2/wp/C6a.md (Status "READY: wp.c6a", 11:40, last touched 12:42). Concept: docs/phase2/concepts/settings.md (SET). Steam build 11094443. Files under test (md5 at 13:58 and again at 15:33, identical on the Frame): theme/60-settings.css 56b5214c, device/rt/60-settings.js 01963e3f, theme/layers/60-settings.json f82b8617. Nothing in the builder's log was taken on trust; every finding below was measured live or read from code, and says which. Raw JSON, probe scripts and outputs: <scratchpad>/c6a_r2/ of this review session. Kept shots: shots/p2_c6a_r2_*.png, shots/p2_cmp_c6a_r2_system.png (CDP captures of the main window; no room imagery is kept; the composites over docs/phase2/mockups/assets/room-*.jpg used for the legibility judgements stay in the scratch).

Status of that pass (superseded by the quick review above)

In progress, session 2 (resumed 15:32 after a usage-limit cut at about 15:14; last update 16:05). Session 1 ran 13:58–15:14. The device is heavily shared: another agent's native session (from about 15:54) contaminated a batch of CSS-tier steps, which are being re-run behind a guard that waits for mode != native. Every CSS-tier verdict below comes from a step whose step: line and result say native=off.

Pending (queued): gates on the remaining route/mode pairs, conformance (laser), perf --ab stock on System and Notifications, the --media contrast shot, one native-session (Audio Reset confirm sgcheck, scrolled-hero hv look and sgcheck, native conformance), and full-budget pad-bfs (System themed vs stock, Audio).

Device state so far: theme on, CSS only after each of my steps; every step used --flags (no flag of mine left); every dialog and menu I opened was closed by its Cancel (value unchanged, modal count back to 0); session 1's two hv frames were viewed and deleted at once on both machines.

Verdict (interim, list not final)

fix. 1 blocker and 9 majors so far. The sidebar, the compact hero, the page keys and colours, the focus and selection looks, removal and fail-closed behaviour are good and pass live. But destructive confirmations opened from settings pages pop +10 mm in native mode against the safety rule (PLAN §1.7 rule 4); several routes fail G-TYPE, G-AUD, G-SIZE or P-23; Steam's 86 px page slide is untouched; the hero stays popped while it slides under the toolbar; and the card's T3 work, M0 and most card tests are absent while the log says READY.

1. What was run

All steps through python glass.py … with --flags wp.c6a,wp.c1a,wp.p3 (plus wp.c1c for dialogs and menus) and --mode pad|laser. "CSS-only" = the step line said native=off.

Id	Command	Result
R-G	gates main --route /settings/<page> --json, both modes, 24 pages	Table in §1.1
R-F	focus main --route /settings/system --pairs <pad pairs> (pad) and <laser pairs> (laser), 14:21 / 14:25	PASS. Pad: P-14 Internet row focus vs rest +91.9 L (row), +98.9 L (right of label); P-15 System focus+selected vs selected +67.3 / +72.8 L. Laser: T-SEL selected vs hovered +42.4 L (right of label) / +37.0 L (row); P-03 hover vs rest +14.5 L (in 10–25). Shots p2_c6a_r2_focpad_{0,1,2}, foclaser_{0,1}
R-ACC	js "$(cat docs/phase2/concepts/settings-accept.js)(<12 routes>)" × 2, laser (T-ACC, T-ROWS)	0 of 24 routes pass (§2 M7)
R-RM	js removal.js with the flags, after the flag step, and with --stock	PASS. With flags: data-lgs-page 25, -hero 1, -title 1, -destructive 1, 1 style property, module installed. Flag popped: all 0, module "off flag wp.c6a is off". --stock: all 0, no runtime
R-IDLE	js idle.js (system, then /library/home)	Install 0.5 ms; 0 DOM mutations in 3 s idle; but the 300 ms tick runs on every route: 0.185 ms per tick on /settings/system, 0.075 ms on /library/home (m3)
R-DES	js destr.js (11 pages): which buttons carry data-lgs-destructive	System Factory Reset, Audio Reset, Developer Format / Clear All / Change User Password, Remote Play Unpair: tagged. Internet "Delete" (Web Browser Data): not (m2)
R-DLG	js dlg_host.js (laser), js dlg_audio.js (pad), --flags …,wp.c1c	Hostname alert: Cancel / Change & Restart, both .Secondary, no tag, no plate. Audio Reset confirm: Confirm (.Primary, default focus) / Cancel, no tag, no plate. Both: card 640 wide, centre (640, 328), buttons 60 tall, scrim black .35 (T-DLG geometry PASS). Closed by Cancel, modals back to 0
R-SRC	js src.js, js src2.js (read-only webpack source text)	Hostname dialog (module 14160): two plain DialogButtons. Audio Reset (module 97849): GenericConfirmDialog whose onOK resets the audio config then calls SteamClient.System.RestartPC(). Network details (module 45935): "Forget" runs accessPoint.Forget() from inside the dialog. Settings page transition: translateY(±12%), 320 ms cubic-bezier(0,0,.1,1) after 80 ms; exit translateY(∓8%) 80 ms (trans.js)
R-MENU	js menu_tz.js (laser), menu_power.js (pad), menu_lang.js (laser), --flags …,wp.c1c	Timezone (64 options): C1c's single scrolling column 560 × 508 at x 264–824, y 108–616, scroll 4918 / 508, focus on the current value, source capsule white .94, Cancel at the end of the scroll. Power (6 + Cancel): compact 64 px rows, slab 400 × 472 at x 556–956, left of the capsule (972–1222). Language (31): column at x 396–956. No list page. Every value unchanged after Cancel
R-LEG	js leg_storage.js, leg_display.js (laser, gamepad focus given with L.gpTake)	Storage: the ornament is a capsule with Uninstall (X), Move Content (Y), Back; each elementFromPoint hits it. Display: More Info (Y), Select, Back, each hit. SET S12 and S13 have a laser path (C1a's ornament)
R-P23	shot … --pre reading the focused rect (pad)	/settings/system entry: System row [16, 108, 368, 72]. /settings/storage after 4 × Down: Proton row [416, 597, 848, 58], bottom 655 (§2 M6)
R-BFS	pad-bfs --route /settings/{system,audio,notifications}, themed (pad, laser) and --stock, default 150 s budget	Truncated on both sides; same exits, same irreversible edge as stock (Back → Down → Output Device on Audio, Back → Down → toast toggle on Notifications), same B effect (Steam's history). Full-budget runs: pending
R-NAT1	native-session 14:38–14:41, steps in native_run.py	§1.2
R-SHOT	shot main p2_c6a_r2_<page> (CSS-only): system (both modes), sys_scroll, storage, audio, notif, sidebar_end, gamerecording, home, family, friends, developer	§1.3

1.1 Gates (CSS-only)

Route	Pad	Laser
system	PASS	pending
internet	PASS	AUD GONE "Other network…" (14:12; re-run pending)
storage	SIZE 10 (sort 156 × 28; art 98 × 45 nested)	same
bluetooth	TYPE 2 ("NOT CONNECTED" 10 px, uppercase, 0.05 em)	same
display	pending	pending
power	PASS	pending
audio	pending	pending
controller	PASS	pending
keyboard	pending	PASS
accessibility	PASS	pending
security	pending	pending
notifications	SIZE 4 (E-SWITCH on %{ParentalWrapper}, hit 62 % own)	same
friends	TYPE 2 (preview friend 15 px / 400, 12 px)	same
downloads	PASS	PASS
cloud	PASS	pending
ingame	SIZE 2 (link "Steam Networking" 24 px tall, P-80 / P-08)	pending
compatibility	pending	pending
family	SIZE 2 (link 19 px) + TYPE 2 (16 px / 400)	pending
remoteplay	PASS	pending
gamerecording	TYPE 8 (15 / 14 px) + AUD SHRUNK %{RecordingModeOption} 952 × 73 → 816 × 70	pending
home	TYPE 2 (12 px; body 20 on 2 lines) + AUD SHRUNK 2 text runs	pending
library	PASS	pending
store	PASS	pending
developer	TYPE 1 (16 px) + AUD SHRUNK (text 952 × 23 → 816 × 19) + OUTLINE (Developer pill top edge ratio 0.559, taken while the list was still scrolling: see m1)	pending

MOTION passes at rest on every route measured (0 running, no lgs-* left).

1.2 Native (P7 complete)

Session 1, one native-session (14:38–14:41), --flags wp.c6a,wp.c1a,wp.p3 (+ wp.c1c for the modal steps):

• sgcheck /settings/system (laser): PASS, main 1 pop, depths {0, 10} mm.
• Hero probe (js_hero.js, __LGS_LAYERS.snapshot() / debug('main') at scrollTop 0, 24, 40, 56, 72): the settings-hero layer is kept at scroll 0, 24 and 40 (texture y 174, 138, 114, i.e. CSS y 116, 92, 76; dz 0.0271, liquid slab, hole, interactive: false) and is dropped only from 56 on ("covered": the search field is under its centre). sgcheck --pre <scrollTop 40>: PASS, 1 pop at 10 mm (M2).
• sgcheck /settings/storage (pad): PASS, 0 pops (no hero on Storage, as designed).
• sgcheck /settings/system --pre <open Hostname> (pad, wp.c1c): main: 1 pops, depths [0.0, 10.0] mm, modal ['c1c-alert'], PASS only because nothing in the alert is tagged (B1).
• sgcheck /settings/power --pre <open the first dropdown>: 1 pop, modal c1c-menu at 10 mm, hero dropped (rule 6). PASS.
• hv c6a_r2_sys --look (laser): glass L 81.1 (55–110), top edge "none" (ratio 0.0), doubling 0.357 (not doubled); --offaxis 30: glass L 71.9, edge none, doubling 0.391. Looked at: the window reads as clear glass over the room, sidebar pills and platters legible, the hero circle in front of the page; verdict withheld by the tool (auto rect). Both frames deleted at once (local look folder and Frame /tmp/lgs/hv-*).

Static check of theme/layers/60-settings.json against contracts/reporter.md §3–§4: version 2, fragment flag wp.c6a, one rule settings-hero on an absolutely placed ::before (P6 crops only those), r 30 on a 60 × 60 circle (rule 5), mm 10 in the allowed set, slab liquid, hole: true, no interactive, wearer.mm 10 (PLAN §1.7 row "Settings hero icon": +10 in both profiles). Decorative, so rule 2 sets no cap. P-48: the CSS circle carries --lgs-shadow-10mm (0 4px 12px, 0.4 px/mm and 1.2 px/mm). Nothing in the fragment keeps the pop off the scroll-edge band under the toolbar row (M2), and C6a sets no data-lgs-destructive inside modals (B1).

1.3 Visual verdict (CSS-only shots, composited over the lounge and studio rooms)

• System, Audio, Notifications, Friends (top), Home (top): this is close to visionOS Settings: a 400 px sidebar of 72 px pills with colour circles and group gaps, a centred compact hero (60 px circle with Steam's own glyph and a Title 1), recessed platters and C4a's controls. Legible over both rooms. p2_cmp_c6a_r2_system.png against settings-system.html: same sidebar, hero and platters; differences are the drill-down rows (not built), the search capsule (C1a's, straddling the sidebar), the 816 px column and the missing System badge.
• Every selected pill: the "specular arc" is a separate light streak 4 px above the pill that starts outside the pill's rounded left corner (pill_zoom.png, every shot): it reads as a detached hairline, not as an edge of the material (m1).
• Game Recording: reads as a skin: the three mode cards keep Steam's 15 / 14 px type, their fills are barely visible, and the unselected radio pips (black .30) almost vanish over the dim room (M3).
• Family, Home's last row, In Game: body text 12–16 px, in-text links 19–24 px tall, the Family primary is a plain left-aligned capsule rather than the page's tinted primary (M3, M5).
• Storage: Steam's layout with new colours: an 8 px usage bar, a 28 px "Size on Disk" sort control, 58 px rows that run to the glass edge and under the ornament when focused (M5, M6, m5).
• Scrolled System (scrollTop 40): the hero circle slides up under the search capsule and is cut by it; nothing takes over the title (no inline title), so the page has no title once scrolled (M2).

2. Findings

Blocker

• B1. Destructive confirmations pop +10 mm in native mode (PLAN §1.7 admission rule 4, SET §5 rule 3 "the safety rule", reporter.md §2.4). C1c's alert rule pops every alert unless it holds .Destructive or [data-lgs-destructive] (22-presentations.json exclude; 22-menus.js tagDestructive flattens the card only when a button is already tagged). For settings, the tag is C6a's job (reporter.md §1: "C1c, C6a"; SET §4.7: "An alert whose buttons include a tagged or .Destructive button is tagged too"). But setTick scans only the page content (60-settings.js lines 217–225: root.querySelector(S.sel.content) → button.DialogButton), so nothing inside a modal is ever tagged, and the confirm that follows a tagged button is not tagged either:
  • Hostname (System): "Change & Restart" (Settings_System_Change_Hostname_Set is even in C6a's token list, but the button exists only inside the dialog). Live: no tag, no plate; native sgcheck: main: 1 pops, depths [0.0, 10.0] mm, modal ['c1c-alert'].
  • Audio Reset: the page's "Reset" is tagged (srcTag: true), but the GenericConfirmDialog it opens has a generic "Confirm" (.Primary, default focus) whose onOK resets the audio config and restarts the device (SteamClient.System.RestartPC(), module 97849). Live: no tag, no plate, so it pops like the Hostname alert.
  • Network details (Internet): "Forget" runs accessPoint.Forget() from inside the dialog (module 45935); never tagged (source read; not opened live, because a wrong row could start a connection). sgcheck passes in all of these because its destructive check reads the same tags, so G-DEPTH cannot see the problem. Fix in C6a's T2: scan the top ModalPosition too (Hostname, Forget, and any other label of SET §4.7 inside a dialog), and tag the confirm of a dialog opened from a tagged source within a few seconds (as C1c does for Power), so C1c's existing tagDestructive flattens the card; then re-run sgcheck with each dialog open.

Major

• M1. Steam's settings page transition is untouched (VP P-54 must, P-58; SET §7 row 1; PLAN S3). Each sidebar step slides the whole content translateY(±12%) (≈ 86 px of a 720 px page) over 320 ms cubic-bezier(0,0,.1,1) after an 80 ms delay, and exits 8 % in 80 ms (Steam's CSS, read live). SET §7 gives C6a the override (≤ 12 px on page, old content out in 150 ms); C1c's 23-transitions.css covers only TopLevelTransitionSwitch and GamepadTabbedPage. Not built, and not listed under "Not built".
• M2. The hero stays popped while it slides under the toolbar row. Native probe: at scrollTop 24 and 40 the settings-hero crop stays at +10 mm with its slab and hole at CSS y 92 and 76 (circle 76–136), inside C6a's own scroll-edge fade (transparent at 84, opaque at 116) and over the bottom of the search capsule (y to ≈ 88); it drops only from 56. So at rest after a small scroll a half-faded circle with a strip of the search capsule floats 10 mm in front over a full slab and contact shadow. SET §5: the hero is a compact header that "slides under the top edge". In CSS the circle is visibly cut by the search capsule and no inline title takes over (p2_c6a_r2_sys_scroll.png). Fix: drop the pop (remove data-lgs-hero from the rule's match, or add a rest attribute) once scrollTop > 0.
• M3. G-TYPE fails on six routes (P-38, P-84 must), both modes where measured: Bluetooth "NOT CONNECTED" 10 px uppercase 0.05 em (SET §4.8: 18 px Semibold title case; C6a only recolours it); Game Recording 8 runs at 14–15 px (C6a restyles %{RecordingModeOption} but leaves Steam's type); Home 12 px description and a 20 px body on two lines; Family 16 px / 400 body and link; Developer 16 px "Steam Play is enabled…"; Friends & Chat preview friend 15 px / 400 and 12 px (SET §4.8: a 48 px avatar and name). T-ACC adds Storage "TOTAL ON-DISK CLIPS / SCREENSHOTS" 12 px uppercase 0.5 px tracking.
• M4. G-AUD SHRUNK (functions-kept gate) on three routes: Game Recording %{RecordingModeOption} (a control) 952 × 73 → 816 × 70; Home two text runs (370 × 20 → 379 × 16, 563 × 17 → 568 × 14); Developer one text run 952 × 23 → 816 × 19. Internet laser had a GONE "Other network…" row in session 1 (re-run pending).
• M5. G-SIZE fails on four routes (P-80, P-08 must): Storage: the sort control "Size on Disk" 156 × 28 (hit 55 %), each row's art %{LibraryItemBox} 98 × 45 as a nested target (SET §4.8 makes the sort a value capsule; that is T1 work, not the T3 row height); Family and In Game: in-text links 19 and 24 px tall; Notifications: E-SWITCH on %{ParentalWrapper} (hit 62 % own over 86 × 80) in both modes. The last one is filed as REQ C6a->C4a (2), which C4a has not answered (no mention in C4a.md); until it is, Notifications cannot reach M2.
• M6. P-23 (must; PLAN §4.4 bounds 124 / 612) fails on C6a's own layout. The sidebar list starts at y 108, so the gamepad entry focus on /settings/system is [16, 108, 368, 72] (conformance --pad: "1 focused rect outside y 124..612"), on every visit. Storage keeps the stock list height and runs to y 720 with no 628 cut: a focused row reaches [416, 597, 848, 58], bottom 655, under the ornament capsule (y 628–712) that shows Uninstall / Move Content exactly then; conformance found 4 such rects. SET §3.5: "A focused row always stays above y 628".
• M7. The card's scope is not delivered, yet the log says READY. PLAN §2.1's definition of done is "every acceptance test in the package card passes". Not built: the T3 P-S1 wrapper (drill-down, VR Settings page; S16, S17, E6), the list page for value menus over 8 options (PLAN §1.12, decision 9: Timezone, Language and Download region open C1c's scrolling column instead, R-MENU), P-S2, the System badge, the inline title, Storage T3. Not run by the builder: T-ACC, T-ROWS, T-DRILL, T-AUD per view, T-CNT, T-HIT, T-MENU, T-SHOT on 24 routes, T-MOT, T-PERF, T-FONT, T-A11Y, T-I18N, T-T3, P-S1/S2/S4. Measured here: T-ACC 0 of 24 routes pass (column 880 > 762 on all, because the 816 px column is drawn as an 880 box with 32 px padding; searchCircle on all (C1a's capsule); rowsAtTop on System, Internet, Storage, Power, Downloads, Family, Game Recording and Developer: 3 rows / 241 px at scroll 0 against ≥ 4 rows or ≥ 260 px; type as M3). T-MENU: the ≤ 8 slab sits left of its capsule, not right-aligned (+8) as PLAN §1.12 and SET §4.5 require (anchoring is C1c's code; C6a never ran T-MENU, so no REQ exists).
• M8. M0 and G-MOCK not done. No C6a-cmp.json, no data-id in any settings-* mockup, settings.md unchanged since the round started, so concept and mockups still show a hidden legend (E-ORN, withdrawn by PLAN §1.10), a .26 selection (S18), a 760 column, the magnifier circle and flat menus (§1.7). The log's "M0 note" table is not M0 (PLAN §2.1: the concept text and mockups updated and re-rendered). glass.py cmp side by side only (§1.3).

Minor

• m1. The selected pill's arc is drawn on the hit extender 4 px above the pill and starts outside its rounded corner, so every shot shows a detached light streak above the selected row. On /settings/developer (pad) G-OUTLINE measured the pill's top edge at ratio 0.559 ("visible") while the list was still scrolling into place (the row at y 616; settled it is at 524–596): the transient is real on every route entry that scrolls the sidebar.
• m2. Internet "Delete" (Web Browser Data, SET §4.7) is not tagged; Developer "Format" is tagged though it only navigates to Storage (onClick: () => Settings("Storage")).
• m3. A 300 ms rt.setInterval runs on every route for as long as the flag is on: 0.075 ms per tick off settings (≈ 0.25 ms/s), 0.185 ms per tick on System (≈ 0.6 ms/s), against RT-7's 1 ms/s for the whole runtime. "Nothing runs at idle" is not met; a route listener plus a MutationObserver on the page content would do.
• m4. Gamepad focus on a sidebar pill: white label on a white .50 composite (≈ 3 : 1), and the grey icon circle disappears into the pill (pill_pad_zoom.png).
• m5. Storage's usage bar is 8 px (SET: a 24 px capsule) and the drive and sort controls keep Steam's proportions.
• m6. P-44 (should): the sidebar's step to the detail pane is < 8 L over the grey and dark rooms (session 1).
• m7. a:is(:hover, .gpfocus, :focus) (in-text links) has no input-mode scope: one of the bundle's P-01 / P-02 failures is C6a's.
• m8. REQ Coordinator->C6a (R2-5, coordinator.md line 240) has no answer; REQ C1a->C6a ×2, C1c->C6a and C4a->C6a ×2 are answered only in C6a's own log.
• m9. The log's "pad-bfs identical to the stock baseline (4 nodes)" is not reproducible: in the same 150 s budget the themed sweep reached 48 nodes and stock 60; both truncated (full-budget runs pending).

Observations for other owners (not counted against C6a): a black "⋯" More circle was captured over the sidebar at (176, 310) on /settings/notifications (14:43, laser; not present in a probe at 14:43 or later): a stale node of C1a's More helper from another route. The 520 px search capsule straddling sidebar and pane is C1a's (REQ C6a->C1a open).

3. Function retention (SET §9)

Everything C6a builds restyles Steam's own nodes; no node is hidden (display/visibility unchanged), moved in the DOM or replaced, and the clip-paths cut only content scrolled under the toolbar row or past y 628 (still scrollable). G-AUD shows no GONE, HIDDEN or UNCLICKABLE on the routes measured except Internet laser (pending re-run).

SET rows	Laser	Gamepad	Evidence
E1–E5, E7, E9	as stock (tab bar, bar menus, Back circle)	as stock	not changed by C6a
E6, E8 (new)	— not built (VR Settings page, SteamVR Back circle is C6b's)	—	nothing lost: SteamVR stays reachable from the tab bar
S1–S4 sidebar, rows	click	Up/Down (selection follows focus), Right, Left, B	pad-bfs, sidebar walk (24 pages in order)
S5, S8–S11 controls	click	A	AUD (no UNCLICKABLE), SIZE except M5
S6 dropdowns	click → C1c's slab or scrolling column, Cancel, outside click	A, Up/Down, A; B	R-MENU (values unchanged)
S7 sliders	drag / click	Left/Right	Audio gates (pending laser)
S12 explainer (Display)	ornament "More Info" (clickable)	Y	R-LEG
S13 Storage Uninstall / Move Content	ornament buttons (clickable)	X / Y	R-LEG
S14 scroll	wheel / drag	focus scrolls (but M6)	P-23
S16, S17 drill-down (new)	— not built	—	M7
S18 Select / Back legends	quiet legend, never hidden (PLAN §1.10)	A / B	R-LEG
§9.3 page functions	Steam's nodes	Steam's nodes	AUD per route (§1.1)
§9.4 dialogs	Cancel / Confirm, outside click	Left/Right, A, B	R-DLG

4. Conformance (P-items for this area)

From conformance --route /settings/system --route /settings/storage --pad (14:52, pad; the System pass overlapped another agent's native session, so its P-39 lines are discarded), G-FOCUS and the probes above. Laser and native conformance: pending.

Item	Result
P-03, P-14, P-15, T-SEL	PASS (R-F)
P-08, P-80	FAIL: Storage, Family, In Game, Notifications (M5)
P-23	FAIL: System entry focus at y 108, Storage rows to 655 (M6)
P-38, P-84	FAIL on six routes (M3)
P-35	alerts centred (640, 328) PASS; menus adjacent (gap 16 px) PASS; anchoring per PLAN §1.12 FAIL (M7, C1c)
P-42, P-43	PASS on 23 routes; m1
P-44 (should)	FAIL over grey and dark rooms (m6)
P-46, P-47, P-48	PASS (1 pop, 60 × 60, shadow 0.4 / 1.2 px per mm)
P-52, P-58	PASS at rest; P-58 FAIL for Steam's page transition (M1)
P-54	FAIL (M1)
P-64	source white .94 while open PASS; rows 72 PASS
P-67	Steam's default focus on "Confirm" in Audio Reset: recorded deviation (PLAN §4.4, S6), not C6a's
P-72	PASS
P-86	scrim .35 PASS
P-89	S12 / S13 legends are visible clickable buttons PASS
PLAN §1.7 rule 4	FAIL (B1)

Requests (that pass)

(none: this review edits no other file; the findings name their owners.)