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

Coordinator: decision log and requests

Owner: the coordinator (PLAN §2.6). This file records the coordinator's decisions on the requests packages address to it (REQ <ID>->Coordinator), the evidence read for each, and the coordinator's own requests to the packages that must change a file. The decisions themselves are PLAN amendments: docs/phase2/PLAN.md §1.19 (round R2), applied in place and marked [R2-n].

Status

• Round R2, second pass done (2026-10-07 13:00–13:30, Steam build 11094443).
  • grep -n "REQ [A-Za-z0-9-]*->[Cc]oordinator" docs/phase2/wp/*.md found one open request: P10->Coordinator (E-GRID). It is decided (R2-15) and marked [x] in wp/P10.md. The nine requests of the first pass are still [x], and no requester added an objection under its answer.
  • Also decided without a REQ line of their own: the label runs that E-GRID was waiving (C2b review R2, M1 and M4; folded into R2-15), and the owner of theme/layers/21-search.json (C1b review R2, m3; R2-16).
  • C6b's "Observation for the coordinator" (SteamVR Display Brightness read 40 %, then 120 %) is checked and closed: the stored value is unchanged from before Glass Shell (below, "Other items seen").
  • New requests from the coordinator: P10, C2b, C2a, V2, C1b (FYI) and C6b (FYI); the R2-15 / R2-16 lines under "Requests".
  • Device: read-only. One SSH read of SteamVR's settings file (ls, grep, diff against the pre-Glass-Shell copy); nothing changed, no lab lock taken, no native session, no imagery.
  • The first-pass request to P10 is being closed by P10's session 5 (its Status, 12:59): code and contract are in (lab_gates.js E-MENU rows and Cancel, tools/p2/conformance.py P-23 at 612, perf --ab); P10 answers under its line here.
• Round R2 done (2026-10-07, Steam build 11094443). All nine open requests to the coordinator are decided and marked [x] in their logs, each with a one-line answer that points to its amendment:
  • C1a ×1 (it covers WN §11 O1 to O4);
  • C1c ×2;
  • C2c REQ-7 and REQ-8;
  • C4a ×1;
  • P2 ×1;
  • P3 ×1;
  • P7 ×1.
• Also decided, without a REQ line of their own: P9's morph sign-off (R2-12), P1's RT-7 deviation (R2-13) and C7's /chat circle (R2-14). WN's open items O1 to O4 came in C1a's request and are covered by R2-10, R2-11, R2-1 and R2-3.
• Amendments: PLAN §1.19 R2-1 to R2-16 (R2-15 and R2-16 from the second pass). DESIGN2 follow-ups are listed as PLAN §1.18 A14 (P4, done 11:40); R2-15 and R2-16 need no DESIGN2 change (DESIGN2 has no launcher-cell or ownership text that disagrees).
• Open (13:30): first pass: C1c, P10 (being closed by P10's session 5), C2c, C1a, C2a, C4a, P3, V1, V2, C5a, C6a, C7 and C3b (P4's is done). Second pass: P10, C2b, C2a, V2, C1b (FYI) and C6b (FYI). PLAN-2b-1 stays FAIL until C2b's labels meet E-GRID (labels), and that blocks only C2b's own M3 claim.
• Device: none. These are document decisions, so no device step was run and the device was not touched. No native session was run, and no room imagery was taken.
• How to re-check.
  • grep -n "REQ [A-Za-z0-9-]*->[Cc]oordinator" docs/phase2/wp/*.md: every hit is [x].
  • grep -n "REQ Coordinator->" docs/phase2/wp/coordinator.md lists the follow-ups. Each owner answers under its line here, as the cross-file request rule says (PLAN §2.6).

Decisions

Rules applied (PLAN §1.19):

• §1 decides conflicts.
• A Steam function, with both its paths, is never traded for a look.
• Between two looks, the one closer to visionOS wins when it can be measured.
• A VP "must" passes or is a recorded deviation.
• Every rule has an agent-runnable check.

Request	Decision	Amendment
C1a->Coordinator (quiet legend, slot arithmetic, WN O1, O2, O4)	The dim band adopted (S27 quietBacking); the slots sum corrected (840 + gaps + padding = 880); achievements stay window; P-23 scored at 612; source card: rule 6	R2-1, R2-2, R2-10, R2-11, R2-3
C1c->Coordinator #1 (REQ 6a–c)	Cancel raised to 60 with its own E-MENU line; P-65 and P-67 recorded deviations; placement on the glass confirmed	R2-5, R2-6
C1c->Coordinator #2 (E-MENU width, D28, Cancel line)	E-MENU ≥ 280 in the grid; 7–10 take the grid, 11–14 scroll inside the slab; Cancel's own line	R2-5
C2c REQ-7 (source card)	The source card drops to 0 with a glow in both modes and profiles	R2-3
C2c REQ-8 (library ornament, PLAN-2c-1)	D-C2c-12 adopted: the ornament hugs its members; the laser pill is the toolbar's trailing group, ≥ 20 px clear of the search field; fixed slots behind libFixedSlots (S28); PLAN-2c-1 replaced	R2-2
C4a->Coordinator (white-fill glow)	§1.4 names --lgs-white-glow and --lgs-focus-add (.32)	R2-7
P2->Coordinator (actionsLive)	S26 added; §4.5 item 4; §4.6 item 9	R2-9
P3->Coordinator (IN-7, P3-D4)	§1.13's delays stand; IN-7's criterion amended with tolerances; tipQuick off (S29)	R2-8
P7->Coordinator (SG-5)	§1.8 states the measurement; CC-A window.dim 0.6 alone; surface dim for a dim behind a pop	R2-4
P10->Coordinator (E-GRID), second pass	E-GRID added as proposed (cells: AUD SHRUNK only, live criterion); a separate E-GRID (labels) line for the text runs; C2b's unbuilt "…" not accepted as a deviation; PLAN-2b-4 (the full name on attention, T3 and the T1 fallback)	R2-15
C1b review R2 m3 (no REQ line), second pass	theme/layers/21-search.json is C1b's; C1b's tiers T1–T4	R2-16

Why, request by request

C1a: the quiet legend (R2-1).

• Evidence read:
  • WN §3.4.1 and §13.1: bare labels at 1.3–1.5 : 1 over the curtain; the band at 4.9–6.7 : 1 (sill) and 5.1–6.0 : 1 (curtain).
  • C1b #3: 1.1 : 1 over L 209.
  • C1c REQ 4: "Select" invisible over L 236.
• Alternatives considered:
  • (a) Keep §1.10's bare look. It fails P-39, a must, so a legend a person cannot read is a lost path.
  • (b) Make every A/B-only legend a glass capsule. That brings back the chrome §1.10 removed on purpose ("chrome only where it carries something"), and it adds a slab in native mode.
  • (c) The band. It is a shadow with no edge and no frost: it passes P-39 and keeps G-OUTLINE and "no capsule material".
• Decision: (c).

C1a: slot arithmetic, O1, O2 (R2-2, R2-10, R2-11).

• 300 + 130 + 156 + 134 + 120 = 840. With four 4 px gaps and 12 + 12 px padding the capsule is 880, which is what WN §3.4.1 says. Corrected in §1.10, now inside R2-2's flag variant.
• O1: §1.2's table and its note disagreed. SM revision 3, WN §3.1.1 and C5a all build achievements as window, so the table holds and the note is fixed.
• O2: A1 moved the ornament from 636 to 628, so P-23's 620 becomes 612 (C1a's --lgs-guard-bottom). VP stays frozen; the bound is in PLAN §4.4. P10's tools/p2/conformance.py:118 still uses 620 (request below).

C2c REQ-7, C1a O4: the source card (R2-3).

• §1.7's row contradicted admission rule 6.
• P6's reporter applies rule 6 in both profiles (contracts/reporter.md step 6: "While the surface's modal selector (or a modal: true rule) matches a visible element, only modal: true rules pop"; the wearer profile skips only rule 2).
• C2c, C1c (D6b) and C2a (D-C2a-3) already build it, and C2c's S2-N2 shows it live.
• A +15 source beside a +10 menu would also invert the visual order.
• Rule 6 wins; the row is amended.

C2c REQ-8: the library ornament (R2-2).

• Evidence read:
  • C2c's log: D-C2c-12; S2-G1, S2-G2 and S2-G6 PASS in both modes; S2-N1 (native look with the toolbar pill).
  • theme/40-library.css §7: the pill is position: fixed, right: 24px, 80 px tall, max-width 420, or 288 on nested routes.
• Accepted:
  • The pill cannot live in the ornament: the window clip is C1a's and must stay, so that nothing floats in the margin.
  • A fixed 880 capsule with empty slots is not visionOS; a toolbar keeps its items, it does not show holes.
  • visionOS puts a window's sort and filter in the top bar's trailing group (DESIGN2 §3.2 "Trailing actions"), so the move is a conformance gain.
• Added: the trailing group must stay ≥ 20 px clear of the search capsule (DESIGN2 §3.2's 20 px between trailing actions).
  • Measured from C2c's own numbers it is 6–8 px today: section roots x ≈ 906 against the capsule's 900; collection pages 968 against 960.
  • Two 60 px capsules 6 px apart read as one control, not as a search field and a separate group.
  • The rule is measurable and keeps every function: Steam's sort name ellipsizes inside the group, while the gap and the inset hold.
• Also added: the account-alert circle (#header_profile, WN §3.2) stays outermost when it shows, so the two never overlap.

C1c: Cancel, E-MENU, D28 (R2-5).

• Evidence read:
  • C1c's gates run 3 and run 4: the only G-SIZE red lines left were the E-MENU width clause and Cancel.
  • D23 and D28 (the 508 px box: 5 rows → 508, 6 compact → 502, 7 compact → 564).
  • D5 and HA-11 (56).
  • P10's E-MENU check (lab/exemptions.json).
• Cancel: 60, not an exemption at 56.
  • P-80 is a must ("no control smaller than 60 px visible on its short side").
  • C1c's Cancel element is already 60 tall (56 visible inside a 2 px clear border), so 60 visible costs no height: the 508 px budget is unchanged.
  • The quiet look comes from the fill (.08) and the label (.70), not from 4 px of size.
  • The ≥ 4 px clear rule keeps it apart from the last row. Compact layouts may need a 2 px margin: 502 → 504, still ≤ 508.
  • GP §3.4 and SET already drew Cancel at 60. C1c's requests asking C5a, C6a, C7, C2c and C3b for 56 are superseded.
• E-MENU width.
  • 2 × 320 + 8 + 16 = 664 > 592, and 592 is also the clip-path morph limit (≤ 600).
  • Steam's own row min-width is 280, and 284 × 72 rows still meet P-08 by hit.
• 7 → grid.
  • It turns 7 compact rows (60 on 64) into 72 px rows, which is better on P-64 and P-08.
  • Power (7) was a grid already (S24).
• 11–14 scroll.
  • Seven 72 px rows per column cannot fit 508 px with a title and Cancel. Three columns would break the 592 / 600 limit and the 280 width.
  • So the row area scrolls inside the slab, with the scroll edge and no scrollbar at rest. These menus are rare on this device: Sort has 10 items, Downloads Options 8.
  • AT-11b's no-scroll check now runs up to 10.

C1c: P-65, P-67, P-35 (R2-6).

• P-35: C1c's D7 already meets it (live: 958 × 487 at (161, 109); alert centre 328). Confirmed and written into §1.12 with numbers.
• P-65: the outside click is Steam's ModalClickToDismiss.
  • It only ever cancels; an outside click never confirms. So the loss P-65 guards against cannot come from it.
  • With the runtime off, C1c's sheet close circle is absent. For a sheet without a button of its own (ZOO's Scroll Panel), the outside click is then the only laser dismiss.
  • A guard would also be new T2 code intercepting Steam's dismissal, which no wearer can validate.
  • Recorded as a deviation; modalOutsideGuard is reserved (S30).
• P-67: decided by S6 at plan time (Steam's default focus untouched). C1c asked only for it to be listed; it is now in §4.4.

C4a: the glow token (R2-7).

• P4's contracts/tokens.md §1.6: --lgs-white-glow 22/8/.55, at +23 L bright and +27 L dim.
• §1.4's 18/2/.30 measured +4 to +12 L, so PLAN contradicted its own gate.
• PLAN now names the token rather than a value, so P4 can tune without reopening PLAN.
• DESIGN2 line 685 already names it (A5).
• The same edit records P4-D8's final focus add (.32), which C1a's REQ to P4 set from G-FOCUS.
• Seen, no change needed: theme/50-appdetails.css:268 uses 0 0 18px 2px white .22 for the laser hover on Play. Hover is allowed to be weaker than focus, and focus there uses the token.

P2: actionsLive (R2-9).

• Accepted as proposed. A dead Play button that every gate passes is the worst kind of lost function.
• Added a §4.5 rule as well: a T3 action's ledger row is partial until the shipped-state check reads live, so the ledger cannot report the launch paths as kept before then.

P3: IN-7 (R2-8).

• §1 decides, and it agrees with D2 §11, MO §4.17 and Apple's WWDC24 sample (delays).
• The card's literal numbers cost visible flicker under laser jitter (tipQuick: a dip to about 0.6 on a return within 180 ms).
• P3's proposed criterion is adopted, with the tolerances set from P3's own measurements:
  • leave start 210–230 ms, inside 200 ± 40;
  • label 0.9 at about 1050 ms, at or before 1100;
  • label gone at 400–420 ms, at or before 460;
  • node hidden at 560–610 ms, at or before 640.
• P3's recorded runs already pass it.

P7: SG-5 (R2-4).

• Read wp/P7.md's SG-5 row and contracts/sg.md §5.
• The inference in §1.8 was wrong. The rule it supported ("no t1 for in-window modals") is still right, for the opposite reason: t1 dims the modal's own crop.
• CC-A's dim must be window.dim alone; P7 measured that the two compound.

Items without a REQ line (R2-12, R2-13, R2-14).

• P9's review R1 m2: glassd has no morph. The CSS clip-path is the visible morph in both tiers, so this is signed off.
• P1's RT-7 deviation (runtime.md §9) is accepted, and the same statistic is applied to G-PERF, because every owner faces the same shared-device noise.
• C7's SM-D2 revision 3 moved the /chat circle with the 512 px sidebar. It is written as a rule, not a coordinate.

P10: E-GRID, and the label runs it waived (R2-15, second pass).

• Evidence read:
  • wp/P10.md REQ and session 4 (live gates barpopup, 11:15: 35 E-GRID waivers, each cell 72 × 96, gaps 0 / 0, hit 100 % own; SIZE PASS on 24 cells); lab/exemptions.json (_scope E-GRID: aud / SHRUNK); lab/lab_gates.js exemptCheck case E-GRID; lab/lab_helpers.js snap (a text record inside a cell takes the cell's match through exemptMatch(..., el, 'aud'), so the cell's result waives its label too) and diff (SHRUNK = themed area under 85 % of stock).
  • C2b's log (Status: 24 cells and 10 label runs SHRUNK; "not built: the word-aware … inside a single long word, T1 and T3 clip it at 70 px").
  • C2b review R2 (in progress): M1 (11 of the 35 waivers are label runs, not cells) and M4 (9 of 23 labels cut mid-glyph, scrollWidth 73–78 against 70).
  • HA §4 (cell, hit region, Labels row, the plate), §0.3 (11 of 23 names cannot fit two lines), §1.3 (stock hides 40 %), §11 (plate timing), AT-12; VP P-08 (its check is "AUD SHRUNK = 0").
• Decision on the cells: as P10 proposed. SHRUNK is an area proxy for P-08, and the cells beat P-08 on both sides (87 × 116 main px, whole-cell hit) where Steam's rows were 48 main px tall. The criterion is live, so a cell that shrinks, opens a gap or loses hit area fails again.
• Decision on the label runs: they are not cells, and the cell criterion says nothing about them. Three options:
  • (a) Keep waiving them with the cell. That hides M4: a cut glyph would pass every gate.
  • (b) Never waive them. Then PLAN-2b-1 fails on every label that HA §4 shortens by design (11 names cannot fit two lines of 70 px) for as long as the grid stands. The alternative grid (3 columns of 92 × 100, 950 px, 35°) was rejected in HA §4 and §17 row 9.
  • (c) Waive them under their own line, which states what visionOS does with a long name: an ellipsis, never a cut glyph, and the whole name on attention.
  • Decided: (c), with four measurable clauses (inside the cell, ≤ 2 lines, no glyph cut with breaks only between words or after "/", the full name kept in the DOM).
• C2b's unbuilt "…" is therefore not a deviation: it fails clause (3). The fix is CSS (an ellipsis on the label box) or T2 (C2b review R2's data-lgs-fit suggestion).
• The full name on attention is the half of the trade that keeps the function, so it becomes a card test (PLAN-2b-4). HA's plate is T3 only in the card. R23's fallback (the "+" patch fails) would then lose 9–11 full names, so the T1 floor shows the attended label whole, in CSS. The test runs in both tiers and both input modes.

C1b review R2 m3: theme/layers/21-search.json (R2-16, second pass).

• The file exists (2.1 KB, two rules behind wp.c1b), it follows §2.1's naming with C1b's NN, and C1b's log explains it (REQ C1b->C1c #8 withdrawn, the round's task asked areas for their own fragments). Nobody else claims it.
• Confirmed for C1b, with a T4 line in its card that restates §1.7 rule 6 and §1.8. The fragment today has "fill": "auto" and no "modal": true (review m1, m2): those are C1b's fixes under rules PLAN already had, not new ones.
• Ownership sweep (a script over §2.6's rows, run in the coordinator's scratchpad): nothing else under theme/, device/, lab/, tools/, native/ or docs/phase2/ is unlisted, apart from the frozen inputs (audit/**, research/**, capabilities/**) and _wip folders (theme/layers/_wip/22-presentations.json is C1c's draft; lgs.py skips _wip).

Requests

• REQ Coordinator->C1c: PLAN R2-5, R2-6. (1) Cancel at 60. Draw Steam's appended Cancel 60 visible on its 60 px element (drop the 2 px clear border), keep min-width 192, centred under the rows (across both columns in the grid), and keep ≥ 4 px clear of the last row's visible fill in every layout (compact may need a 2 px margin: 502 → 504 ≤ 508). It keeps data-lgs-exempt="E-MENU" and is judged by E-MENU's new Cancel line. (2) AT-11b and C1c's tests: no scroll up to 10 items. For 11–14 the grid's row area scrolls inside the slab (scroll edge, no scrollbar at rest, P-72, focused row fully inside the slab); record it. (3) Mark your REQs to C2c, C3b, C5a, C6a and C7 that say "Cancel a 56 px quiet capsule" as superseded: it is 60 (GP and SET already say 60). Re-render your own mockups when you next touch them (the 4 px is inside G-MOCK's ± 8). Placement (D7), D9 and D10 need no change: they are now PLAN §1.12 and §4.4.
  • C1c (2026-10-07 20:25, R2 fix 1): (1) done: Cancel is 60 visible on its 60 px element (border 0, background to the border box), 192 wide, centred, 14 px clear of the last row; gates tile pad/laser and Power pad/laser: E-MENU Cancel line PASS ("fill 192 x 60 on a 192 x 60 element"). (3) done: C1c's REQs that said 56 are marked superseded in its log. (2) the 11-14 grid scroll is not built yet; it is listed as open in C1c's Status.
• REQ Coordinator->P10: PLAN R2-5, R2-11, R2-13. (1) E-MENU in lab/exemptions.json / gates: row width ≥ 280 when the row sits in a two-column grid (its slab has rows at two distinct x positions, or C1c's lgs-menu-grid), ≥ 320 otherwise. New line for Steam's appended Cancel (the last item of %{*BasicContextMenuModal>contextMenuItem} after the actionable rows): visible height ≥ 60, width ≥ 192, ≥ 4 px clear of the previous row's visible fill, inside the slab, elementFromPoint at its centre hits it; listed under SIZE exempt with pass/why. (2) tools/p2/conformance.py P-23 (line 118: > 620): bottom bound 612 when the route has a bottom ornament (any #Footer legend rendered), else the glass bottom − 16 (704 for window-full and windowless); top 124 unchanged; update contracts/lab.md (the P-20/22/23 row) and the stock baseline note. (3) perf ABBA mode for R2-13: for example perf SURF --route R --ab stock|theme --rounds 2. It alternates the reference (--stock, or theme-only with the runtime off) and the subject, and prints per run fps, long frames (> 34 ms) and native, then the median fps ratio, the median extra long frames, the reference's A/A spread and PASS/FAIL by R2-13. It pools a second round when the first fails. Today RT-7 computes this inside P1's selftest and G-PERF has no tool for it.
  • P10 (2026-10-07 sessions 4–5): done, all three; contracts/lab.md §4 (conformance, perf), §6 (E-MENU), §10. (1) E-MENU: rows: visible fill ≥ 60, ≥ 280 wide in the two-column grid (rows at two distinct x in the slab, or .lgs-menu-grid), ≥ 320 otherwise, pitch 64 compact / 78 regular. Steam's appended Cancel (the last item of a top-level menu's first slab) has its own line instead of P-08 and the row clauses: visible fill 60 tall and equal to its element, ≥ 192 wide, ≥ 4 px clear of the previous row's visible fill, inside the slab, elementFromPoint at its centre hits it; listed under SIZE exempt with pass/why. Live (gates main --pre <a lab context menu of N inert items through Steam's own menu component> --only size --flags wp.c1c, 13:37–13:42): rows pass at 5 (column 384 × 72, pitch 78), 6 (compact column, pitch 64), 7 and 10 (grid 284 × 72), in pad and laser; Cancel fails its own line every time ("fill 188 x 56 on a 192 x 60 element"), which is C1c's open REQ (1). Fixture cases 9a–9d. (2) P-23: bottom bound 612 when the route renders a #Footer legend, else the glass bottom − 16 (704 on window-full and windowless, 640 on window), top 124; the step records the route's layout; the lab.md P-20/22/23 row and the stock-baseline note are updated (the baseline ran without --pad, so nothing in it changes). Live (13:44): 612 on /library/tab/AllGames, 704 on What's New. (3) perf SURF --route R --ab stock|theme [--rounds N] [--seconds S]: ABBA × 2 per round, per-run fps, long frames (> 34 ms), p95 and native; the verdict pools native=off runs only and is R2-13's (median fps ratio ≥ 0.95, median extra long frames ≤ the reference's A/A spread, min 1), a second round pooled while it fails; it is BLOCKED (exit 3) with fewer than 2 pooled runs a side (12:11: all 16 runs fell inside another agent's native session, so BLOCKED, as designed). Since session 5 the subject runs get the step's flag overlay, mode stub and action logger again after each lgs on (reapplied). Live (perf main --route /library/home --ab theme --rounds 1 --seconds 2 --flags wp.c2a --mode pad, 16:19, 8 runs native off): fps median 87.75 (theme only) vs 89.0 (with the runtime), ratio 1.014, extra long frames −0.5 within the A/A spread 5 → PASS; every subject run says reapplied.
• REQ Coordinator->C2c: PLAN R2-2, R2-13. (1) Trailing group spacing. Its visible capsule must stay ≥ 20 px clear of the search field's visible capsule: today it is 6 px on section roots (x ≈ 906 against 900) and 8 px on collection pages (968 against 960). That means max-width about 336 on section roots and 276 on nested routes, with the sort name ellipsizing first and the 24 px inset kept. When Steam's #header_profile circle shows, the group sits 20 px to its left. (2) PLAN-2c-1 is now R2-2's three checks (hug, the toolbar pill's rect and hit, nothing hidden), in both modes with focus in the grid and on the tab row; record them. The 880 check runs only with libFixedSlots. (3) G-PERF on /library/tab/AllGames. P4 measured themed 61.1 fps against stock 83.4 (4 s, during another agent's native session, wp/P4.md), and P1 measured theme-only at 64–68 fps. Please run G-PERF with R2-13's statistic in CSS-only (native=off runs only). If it fails, measure the library CSS's share (A/B with a --pre that disables groups of 40-library.css rules, as P4 did with fontkit.py --perf-js), fix what is yours and report the rest here. You own the route (§4.2). (4) Your mockups' tile menu draws Cancel at 56 (M0-G1). It is 60 now (R2-5; C1c's REQ to you is superseded on that point); re-render when you next touch them.
  • C2c answer (2026-10-07, R2 fix pass 1; evidence in wp/C2c.md, R2F rows): (1) done: the trailing group is capped at 336 px on section roots and 276 on collection pages (padding 18, gap 8, so "Alphabetical" fits whole: x 926-1256, 26 px clear of the search capsule's 900); the sort name ellipsizes first; when Steam's #header_profile circle shows the group moves 20 px left of it (right 104, caps 256 / 196). (2) R2-2's three checks are recorded as PLAN-2c-1 (R2F-R22), both modes, focus in the grid and on the tab row, AllGames and the VR collection; the 880 check runs only with libFixedSlots. (3) G-PERF: perf --ab stock FAILs at 0.707 (85.3 → 60.3 fps, R2F-P6). C2c's share, measured in-step (R2F-P1..P5): 40-library.css costs nothing measurable; the library module without the scrubber costs nothing; the letter scrubber costs about 5 % after two fixes (a per-frame layout read, 13 %; a fixed node over the scroller). The rest of the route's gap (about 85 → 66 fps with no C2c code running) is outside C2c's files; reported in REQ-13 (3) of wp/C2c.md. (4) done: the three mockups are re-rendered with Cancel at 60, the hugging ornament and the toolbar group (shots/p2_home-apps_library*.png, _filter.png).
• REQ Coordinator->C1a: PLAN R2-1, R2-2, R2-3, R2-5, R2-10, R2-11. window-nav.md updates:
  • §3.4.1 "Note for the coordinator" → adopted (PLAN §1.10, S27).
  • §3.4.1 "Library routes" row and §3.4.2 slot 1: the fixed slots run only behind C2c's libFixedSlots (off). In laser mode the library's Sort & Filter pill is the toolbar row's trailing group, not slot 1. AT-6's fixed-slot part and AT-24 (b) run only with libFixedSlots.
  • §3.2: on library routes the toolbar's trailing area holds C2c's group (with Steam's account-alert circle outermost). Keep it free of shell items; this answers C2c's REQ-9 in substance, so please answer it there.
  • §3.7 and §11: the source-card row is rule 6 (R2-3); close O1 (R2-10), O2 (R2-11, PLAN §4.4), O3 (R2-1) and O4 (R2-3).
  • §0's §1.10 row ("Sort/Filter in slot 1 in laser mode; library five slots (880)") and §3.2's "V2 is asked to score P-23 at 612" → PLAN §4.4.
  • §5.1.1, §9.5 C4 and §13 AT-11b: the count table (6 compact, 7–10 grid, 11–14 grid scrolling inside the slab), Cancel 60, and AT-11b's no-scroll check up to 10 (R2-5).
  • Mockups that draw the library's fixed slots may keep them as the flag variant, but the default strip should show the hugging capsule and the toolbar pill (window-nav-ornament.html, -anatomy*.html) when you next re-render.
  • C1a answer (2026-10-07, R2 fix pass): done. window-nav.md: §0's §1.10 row, §3.2 (the trailing group row: kept free of shell items, ≥ 20 px clear of x 900), §3.4.1 (library routes hug; the note → adopted), §3.4.2 slot 1, §3.7 (R2-3), §5.1.1 count table (6 compact, 7–10 grid, 11–14 grid scrolling inside the slab; Cancel 60; AT-11b up to 10 no scroll), §9.3 L6/L7, §9.5 C4, §11 O1–O4 closed. Mockups window-nav-anatomy*.html, -chrome.html show the hugging capsule and the toolbar's Sort & Filter group; -ornament.html strips 1–2 are labelled the libFixedSlots variant; re-rendered (shots/p2_window-nav_*.png).
• REQ Coordinator->C2a: PLAN R2-2, R2-3, R2-5. home-apps.md:
  • §7.1: the five fixed slots, and the laser pill in slot 1, become the libFixedSlots variant. The default is the hugging ornament plus the toolbar's trailing Sort & Filter group (D-C2c-12). C2c's text for its sections can be requested from C2c.
  • HA-11: Cancel 60 (P-80), not 56.
  • Line 130 (§1.7 conformance row) and §16 #18: resolved by R2-3 (the row now says rule 6).
  • C2a answer (2026-10-07, R2 fix pass): done in home-apps.md: §7.1 opens with the default (hugging ornament, laser Sort & Filter as the toolbar's trailing group, gamepad X/Y legends) and labels the five slots the libFixedSlots variant (off, S28), slot 1's laser text marked "(libFixedSlots only)"; HA-11 and §8's Cancel row say 60 with the E-MENU (Cancel) line; §0.5's §1.7 row and §15 #18 (the "#18" of the request) say R2-3 resolved the conflict; §0.5's §1.10 and §1.12 rows follow R2-2 and R2-5. xc-home-search.md (rev. 4) §3 follows R2-2 too.
• REQ Coordinator->C4a: PLAN R2-5, R2-8.
  • controls.md §18.2 C9: score tooltips by the amended IN-7 criterion (PLAN §2.3 P3), not "label at 900 ms; gone ≤ 250 ms". This answers P3's FYI REQ to you.
  • §8.2, C-D16 and the E-MENU row (line 838): Cancel 60 with its own E-MENU line; the grid-width clause ≥ 280.
  • controls-pickers.html, -timezone.html and -steamvr.html draw Cancel at 56: re-render at 60 when you next touch them.
• REQ Coordinator->P4: PLAN §1.18 A14 (round R2). DESIGN2:
  • §3.2 bottom-ornament row: the quiet legend's dim band on window routes (R2-1).
  • §3.7 menu row and the layout-by-count table: 6 compact, 7–10 grid, 11–14 grid with the row area scrolling inside the slab; "No menu scrolls inside the 520 px box" → "no menu of up to 10 items scrolls; the box is y 108–616"; Cancel 60 with its own E-MENU line; E-MENU widths ≥ 320 in one column, ≥ 280 in the grid (R2-5).
  • §3.7 Sheet and Alert rows: the placement numbers of PLAN §1.12 (R2-6).
  • §3.8 depth table "Source card while its menu is open": 0 with a CSS glow in both profiles (R2-3).
  • §3.8 Dimming: replace "in native mode it would dim only the hidden real panel" with SG-5's measurement (R2-4).
  • Nothing to change for R2-7: line 685 already names --lgs-white-glow, and line 1027 records .32.
  • P4 (2026-10-07 11:40, maintenance session): done, tagged [A14] in place (11 tags) and registered in DESIGN2 §0.3 (A1–A14, with sources). §3.2: the quiet legend's dim band on window routes with its values and flag (R2-1), the ornament hugs its members on library routes too and the laser Sort & Filter pill is the toolbar's trailing group (R2-2, also §7.3). §3.7: menu row with the 108–616 box and Cancel 60 with its own E-MENU line; the count table 6 compact, 7–10 grid, 11–14 grid with the row area scrolling inside the slab; "no menu of up to 10 items scrolls; the box is y 108–616"; value menus follow the table, "up to 8" is placement; E-MENU ≥ 320 in one column, ≥ 280 in the grid (R2-5); Alert centred at (640, 328) / (640, 360) and Sheet top 108, ≤ 488 / 552 tall, outside click and default focus as recorded deviations (R2-6). §3.8: source card 0 with a CSS glow in both profiles (R2-3); Dimming rewritten from SG-5's measurement, with the surface dim and CC-A's window.dim 0.6 alone (R2-4). FD-7 re-checked (grep -o "\[A<n>[],]"): A1 11, A2 4, A3 14, A4 6, A5 13, A6 3, A7 2, A8 3, A9 1, A10 1, A11 1, A12 3, A13 3, A14 11; wp/P4.md.
• REQ Coordinator->P3: PLAN R2-8. P3-D4 is decided.
  • contracts/interaction.md §0 and §4.1: drop "pending the coordinator" and cite PLAN R2-8; IN-7's pass criterion is the amended one.
  • tipQuick stays off (S29) and stays as a test switch.
  • Record IN-7 against the amended criterion in wp/P3.md. Your measurements (leave start 210–230 ms; label 0.9 at about 1050 ms; label gone at 400–420; node hidden at 560–610) already meet it.
  • Your REQ P3->V1 can drop the tipQuick part.
  • P3 (2026-10-07 14:45 EDT, maintenance session 4): done. contracts/interaction.md §0, §4.1 (and §8) cite R2-8 with the amended criterion, nothing "pending"; tipQuick stays off as a test switch; REQ P3->V1 now asks only for wp.p3. IN-7 recorded against the amended criterion in wp/P3.md: PASS (s3k, 08:17, both inputs; re-run this session on the busy device: laser meets every clause on its own samples, the pad materialize starts 802-803 ms after focus measured directly, its runner samples lag by the sampler's own frame gap; the runner now scores R2-8 itself). Noted there: the first tooltip shown in a window costs about 30 ms once (node and style created), not changed.
• REQ Coordinator->V1: PLAN R2-1, R2-2, R2-6, R2-8, R2-9.
  • device/defaults.json: "quietBacking": "window" (C1a's open REQ). "actionsLive": true only at release, under §4.6 item 9 (P2's open REQ). Do not set tipQuick, libFixedSlots or modalOutsideGuard (S28 to S30 are off).
  • docs/NATIVE.md: point its glassd section to glassd-material.md (GM v2 materials, not the v1 table; P9's note in GM §7) when you do the §6 update.
• REQ Coordinator->V2: PLAN R2-2, R2-6, R2-9, R2-11, R2-13. In REPORT.md and conformance.md:
  • score P-23 at 612 (§4.4);
  • list P-65 and P-67 as recorded deviations (§4.4);
  • use the amended PLAN-2c-1 (R2-2) and IN-7 (R2-8);
  • use R2-13's statistic for G-PERF (CSS-only verdicts from native=off runs only);
  • run §4.6 item 9's shipped-state check (P2's open REQ to you) before marking any T3 launch path kept (§4.5 item 4).
• REQ Coordinator->C5a: PLAN R2-5, R2-10.
  • Keep GP §3.4's Cancel at 60: C1c's REQ asking you for 56 is superseded.
  • Achievements stay window with the quiet legend on the dim band, as built (R2-10; C1a's open REQ about .gp-foot.quiet).
  • GP §8 Q-B (the binding UI's Back text run) was decided at plan time by S4 (E-BACK, PLAN §1.16; S4's source column lists GP Q-B): close it in GP.
  • C5a (R2 fix 1, 2026-10-07): done, all three. (1) GP §3.4 already said 60; the GP mockups drew C1c's 56: game-pages.css .gp-menu .cancel is now 60 tall, ≥ 192 wide, 4 px clear, and game-pages-manage, -manage-sub, -playfrom were re-rendered (viewed); C5a-D7 in wp/C5a.md says 60. Nothing in C5a's live files draws a menu (C1c's slab). (2) R2-10 check run: glass.py js with --flags wp.c5a,wp.c1a,wp.p3 (19:33, native off) reads %{BasicUiRoot}[data-lgs-glass] = window on /library/app/620980/achievements/my/individual (and hero on the game page); the quiet legend is C1a's. The achievements fallback was also rebuilt this pass (pinned row at y 116, 1136 px column, Steam's row slots: GP §4.10, C5a-D10). (3) Q-B is closed in GP §8 ("Closed at plan time … decided by S4 as E-BACK").
• REQ Coordinator->C6a: PLAN R2-5. Value menus follow the count table: 6 options compact, 7–8 the grid. "Up to 8: anchored" is placement; settings routes with more than 8 keep your list page. SET's Cancel at 60 stands; C1c's REQ asking you for 56 is superseded. C6a (R2 fix 1, 2026-10-07): adopted. Value menus are C1c's slabs by the count table (6 compact, 7–8 grid, anchored); a settings value menu with more than 8 options takes the list page, which is not built yet (SET §10 fallback: C1c's scrolling column, logged in C6a's Status); Cancel stays at 60 (C1c's 56 superseded). SET's text already says 60.
• REQ Coordinator->C7: PLAN R2-5, R2-14.
  • The friend menu's Cancel is 60, not 56 (C1c's REQ is superseded on that point).
  • SM-D2's /chat circle box is now PLAN's rule, (sidebar width − 94, 14), and applies only while searchCircle is on. The People test row in SM §8.2 still expects "(362, 14)" and a 456 px sidebar: update it when you next touch SM.
• REQ Coordinator->C3b: PLAN R2-4, R2-5.
  • CC-A in native mode uses the report's window {dim: 0.6} alone, with no surface dim (they compound to 0.36). P7's open REQ says the same.
  • concepts/control-center.md §3's "Power alert … t1 tint" row: no t1; it is the CSS scrim (§1.8).
  • control-center-power*.html: Cancel 60 when you next re-render (C1c's REQ about these mockups stands otherwise).

Second pass (R2-15, R2-16), 2026-10-07 13:10:

• REQ Coordinator->P10: PLAN R2-15 (PLAN §1.16 E-GRID and E-GRID (labels)).
  • (1) E-GRID (labels) in gates' AUD. A text record inside an E-GRID cell no longer takes the cell's result (snap matches it to the cell through exemptMatch(..., el, 'aud') today). Judge it by §1.16's label line on the themed snapshot:
    • (a) its rect lies inside the cell's rect ± 1; for the attended cell (.gpfocus in pad mode, :hover or .lgs-dwell in laser mode) inside the popup's width (x 0–300) instead;
    • (b) ≤ 2 line boxes: distinct tops (± 2 px) of Range.getClientRects() over its text, counted inside its visible rect;
    • (c) no glyph cut: on the run's element and on each ancestor up to the cell whose overflow is not visible (or that has a clip-path), scrollWidth > clientWidth + 1 needs computed text-overflow: ellipsis on that box, and scrollHeight > clientHeight + 1 needs -webkit-line-clamp 1 or 2 on it; on the run's element word-break and overflow-wrap are normal and hyphens is not auto;
    • (d) in audDiff: the themed record's text equals the stock record's, or it ends in "…" and the cell's aria-label equals the stock text.
    • Report a pass under exempt as {id: "E-GRID (labels)", waived: "SHRUNK", why} with the clause results. A failing clause keeps the SHRUNK, with "(E-GRID label criterion not met: …)". The cell line is unchanged.
  • (2) tools/p2/test_gates_page.py: four label cases: an ellipsized label (waived), a label cut without an ellipsis (SHRUNK kept), a three-line label (kept), and a T3-style "…" label whose cell has no aria-label (kept).
  • (3) contracts/lab.md §6: E-GRID is in PLAN §1.16 [R2-15]. Drop "not yet in PLAN §1.16" and add the label line to the table.
  • (4) Minor, seen while reading §6: the E-SEG check accepts ≥ 120 wide for every segment. PLAN §1.16, CTL §8.3 and CTL §18.1 say ≥ 140, and 120 only for a compact segment. CTL §8.3's regular label is 22 px, so check ≥ 140 when the segment's label computes to 22 px or more, and ≥ 120 below that (C4a's --lgs-seg-min-w is already 140).
  • P10 (2026-10-07 session 5): done, all four. (1) AUD no longer lets a text run take its cell's result: the cell line applies only to the cell (exemptMatch host), and each text run inside an E-GRID cell is judged by L.gates.labelFacts / labelVerdict on clauses (a)–(d) as written (attended = .gpfocus, :hover or P3's .lgs-dwell, then inside x 0–popup width; line boxes = distinct getClientRects() tops ± 2 px inside the visible rect; the cut test box by box up to the cell, with -webkit-line-clamp read on a vertical -webkit-box; word-break/overflow-wrap/hyphens on the run; (d) the stock text, or "…" plus the cell's aria-label). A pass is {id: "E-GRID (labels)", waived: "SHRUNK", why: "(1) … (2) … (3) … (4) …"}; a failing clause keeps the SHRUNK with "(E-GRID label criterion not met: …)". Also from C2b's review R2: on a line-clamped box an overflow across is a cut whatever text-overflow computes (Chromium draws no ellipsis there). (2) tools/p2/test_gates_page.py cases 12a–12f: your four (ellipsized waived; cut without an ellipsis kept; three lines kept; T3 "…" without aria-label kept, with it waived) plus a single long word in a clamped box (kept), a one-line nowrap ellipsis (waived), the attended plate inside the popup (waived) against the same plate on a resting cell (kept) and a short label (waived); case 0 checks lab/exemptions.json against §1.16. (3) contracts/lab.md §6: E-GRID and E-GRID (labels) are §1.16 rows [R2-15], "not yet in PLAN" is gone, the label line is in the table; _pending (ids §1.16 lacks: reported, never waived) is empty. (4) E-SEG: ≥ 140, or ≥ 120 when the segment's label computes under 22 px (case 14: 130 px segments fail with a 22 px label, pass with 20 px). Live (gates barpopup --pre <open "+"> --only aud --mode laser, T1, 16:07): 24 cells waived: SHRUNK by the cell line, 6 labels by the label line (e.g. "Frametop Input Settings": "(1) inside its cell; (2) 2 line(s); (3) no glyph cut; (4) its text is the stock name"), and 5 SHRUNK kept: qBittorrent, Chromium, LXTerminal, RenderDoc "cut across at the run without an ellipsis (a line-clamped box draws none across)" and Hide/Show Screens: C2b's M4, reported as it should be. contracts/lab.md §4 AUD, §6, §10.
• REQ Coordinator->C2b: PLAN R2-15.
  • C2b (2026-10-07 fix pass R2, session 2; evidence in wp/C2b.md): done, all three. (1) No glyph is cut and no label uses -webkit-line-clamp (Chromium draws no "…" across inside a clamp, and live its own "…" overflowed the box: "Remote..", "System.."). The CSS-only floor draws one line ending in "…"; with the runtime, names that fit wrap to two lines, and T3 draws a longer name as two explicit lines (breaks between words or after "/"), the second in its own box ending in "…" at a word's end. The withdrawn "not built" deviation is gone from the Status. (2) PLAN-2b-4 PASS in T1 (CSS only) and T3, both modes: every shortened cell shows its stock name whole, inside x 6–294, the labels it covers at .22, the previous cell back at rest. One limit: in the CSS-only floor the laser plate shows on P3's 80 ms dwell, not after 0.4 s (CSS cannot delay a layout switch; with wp.c2b the 0.4 s step comes from P3's rt.attend). (3) PLAN-2b-1 PASS in T1, T3 and T3 + toggle, both modes (and High Contrast, Reduce Motion): AUD 0 issues, only E-GRID and E-GRID (labels) waivers, each why recorded, none "not met".
  • (1) Labels cut no glyph (§1.16 E-GRID (labels); your review R2's M4). A name that does not fit ends in "…", drawn by the box that clips it: for example text-overflow: ellipsis on the clamped label, or T2's data-lgs-fit with white-space: nowrap for a single long word, as your reviewer suggests. Lines break only between words or after "/", at most two lines, inside the cell. Your log's "not built: the word-aware …" is not accepted as a deviation; please withdraw it.
  • (2) PLAN-2b-4 (your card): the full name on attention, in T3 and without T3 (R23's fallback), in both input modes. Without T3, the attended cell's own label shows the whole name in the plate's look, in CSS: on .gpfocus in gamepad mode, and in laser mode on :hover after the 0.4 s dwell (a transition-delay, for example), both keyed on the input mode (P-01, P-02). Keep it inside x 0–300, left-aligned in column 1 and right-aligned in column 4 as HA §4's plate, with the labels it overlaps at ≤ .22.
  • (3) Re-run PLAN-2b-1 in T1 and T3, in both modes, and record each E-GRID waiver's why (cell line or label line) once P10's label line is live. Until then the evidence names R2-15 and says the label runs were judged by hand (scrollWidth, text-overflow, line count).
  • Your Status's "READY" waits on (1) to (3) and on the review R2 findings.
• REQ Coordinator->C2a: PLAN R2-15. home-apps.md:
  • §4, Labels row: add "Without T3 (R23's fallback), the attended cell's own label shows the whole name in the plate's look, in CSS (PLAN-2b-4)".
  • AT-12: add "no label cuts a glyph: a name that does not fit ends in '…' (PLAN §1.16 E-GRID (labels))".
  • §0.5, the §1.16 row: E-GRID and E-GRID (labels) [R2-15].
  • C2a answer (2026-10-07, R2 fix pass): done, all three, in home-apps.md: the §4 Labels row (the T1 fallback's full name in the plate's look, keyed on .gpfocus / :hover after the 0.4 s dwell by input mode, PLAN-2b-4; no cut glyph, the clipping box draws the "…"), AT-12's pass column, and §0.5's §1.16 row (E-GRID, E-GRID (labels) [R2-15]).
• REQ Coordinator->V2: PLAN R2-15. Score PLAN-2b-1 with §1.16's E-GRID lines as the only waivers; a waiver whose why does not name its clause results is a FAIL. In the function ledger (§4.5), HA §12.7 row P2 ("Launch a program") counts as kept only when PLAN-2b-4 passes too, in T3 and in the T1 fallback, in both modes.
• REQ Coordinator->C1b: FYI, PLAN R2-16. theme/layers/21-search.json is yours (§2.6; your card's Owns and a new T4 line). The decision asks for no change of its own. The T4 line restates §1.7 rule 6 ("modal": true on the sheet rule) and §1.8 (hole.fill follows the scrim), which your review R2 lists as m1 and m2.
  • C1b (2026-10-07, R2 fix pass): done. theme/layers/21-search.json: both rules "modal": true, "hole": {"fill": "scrim"}; the search-sheet rule pops C1b's glass with "hitTest": false (it is pointer-events: none inside an inert layer, so the reporter's centre hit test always called it covered: review M6). In native mode, once the pop is acknowledged, 21-search.css drops the CSS tint, frost and shadow and cuts the page picture and scrim out under the sheet, so the crop carries only the sheet's content over glassd's slab. Native evidence in the C1b log.
• REQ Coordinator->C6b: FYI, your "Observation for the coordinator" is closed, and no change in your files is needed. SteamVR's stored Display Brightness never changed. ~/.config/openvr/config/steamvr.vrsettings (rewritten by SteamVR at 12:45:19) holds analogGain 0.13320851755, the same value as steamvr.vrsettings.pre-glass-shell (0.13320851318, saved 2026-10-06 21:27) up to float noise. So at 12:45, after your 12:33 shot, SteamVR's live value was the pre-Glass-Shell one. It is also consistent with the 40 % you read at 11:52 to 12:29 (0.1332^(1/2.2) = 0.40, if the slider maps gain with a 2.2 gamma [inferred]). The 120 % is most likely a reading of the page (for example the slider before SteamVR filled in its value), not a changed setting. Please mark the observation closed when you next touch your log.

Requests to the coordinator, handled

Found at the start of round R2 with grep -n "REQ [A-Za-z0-9-]*->[Cc]oordinator" docs/phase2/wp/*.md, 2026-10-07. Each is marked [x] in its log, with the answer on the line under it.

Log	Line (at the time)	Request	Answer
C1a.md	94	Quiet legend, slots sum, WN O1, O2, O4	R2-1, R2-2, R2-10, R2-11, R2-3
C1c.md	44	REQ 6a–c: Cancel, P-65/P-67, P-35	R2-5, R2-6
C1c.md	45	E-MENU width, D28, Cancel's own line	R2-5
C2c.md	49	REQ-7: source card	R2-3
C2c.md	50	REQ-8: library ornament, PLAN-2c-1	R2-2
C4a.md	63	White-fill glow token	R2-7
P2.md	98	actionsLive S-row and §4.6 item	R2-9
P3.md	35	IN-7 (P3-D4)	R2-8
P7.md	76–80	§1.8 after SG-5	R2-4
P10.md	72 (second pass, 13:10)	E-GRID in §1.16	R2-15

Other items seen (not requests to the coordinator)

• theme/50-appdetails.css:268 (C5a): laser hover on Play uses a lighter literal glow (0 0 18px 2px white .22). It is a hover look, weaker than focus by design; focus uses the token. No action.
• device/rt/03-react-lab.js:30 (P2): a lab fixture uses the old glow literal. Lab only, never shipped. No action.
• GM §7 (P9's notes): the reporter's wide hdr-search slab is superseded under wp.c1a (C1a's layers/20-shell.json), so nothing in the toolbar row pops. The NATIVE.md pointer is filed with V1 above.

Second pass (2026-10-07 13:10):

• C6b's "Observation for the coordinator" (SteamVR Display Brightness read 40 % at 11:52–12:29 and 120 % at 12:33). Checked read-only over SSH. ~/.config/openvr/config/steamvr.vrsettings (mtime 12:45:19) has analogGain 0.13320851755. The pre-Glass-Shell copy steamvr.vrsettings.pre-glass-shell (2026-10-06 21:27) has 0.13320851318. The other differences are lastAccessedExternalOverlayKey and the VRWebHelper debugger block, which LAB.md documents. No setting value changed, so nothing needs restoring. Answered as REQ Coordinator->C6b (FYI).
• C1b review R2, the findings that say "with the coordinator". Only m3 needed the coordinator (R2-16). M4 (AT-9c: B returns to the tab row, not the poster), M8 (the item menu and S-A to S-C rows) and m8 (AUD GONE for the virtualised reflow) are C1b's to fix first. If C1b wants a deviation or an exemption instead, it files REQ C1b->Coordinator with the evidence. Noted for that ruling: VP P-21 (a must) asks for focus back on the sheet's source after B, so any deviation from AT-9c must still meet P-21.
• P6 review R1 (f), "for the coordinator": on AllGames the search field overlapped the tab row in CSS-only mode, from in-progress C1a/C1b CSS. That is superseded by R2-2's toolbar rules and C2c's REQ (the ≥ 20 px gap), and the C1a/C1b/C2c gates on AllGames judge it. No separate action.
• E-SEG drift (gate vs PLAN): see REQ Coordinator->P10 (R2-15) item 4.

Requests (user decision, 2026-10-07)

• REQ Coordinator->V1: USER DECISION: ship native glass on by default ("native": on in device/defaults.json) even if the performance budget (glassd <= 2.5 ms, G-PERF, AT-0c cost) is not met. Performance misses are recorded in REPORT and V1.md as known issues, not gate failures. Only a functional failure blocks native: input does not pass through to Steam (AT-0a), doubled or misaligned content (AT-0c alignment, AT-0d), or no clean return to CSS-only (AT-0f). In that case, record it and still prefer native if the failure is cosmetic.