desktop-v2 on the v3 path — operator-visible parity evidence¶
Ticket: MOR-1372 (v3-rework S12, the closing slice of MOR-1263) ·
Verified at: main 7612a4f8 ("feat(MOR-1370): scopeControls zone ownership on desktop-v2") ·
Kind: evidence artefact. It records what an operator gains, loses and keeps when desktop-v2
renders the semantic surface family instead of the legacy panel family. It defines no behaviour and
changes no production file.
Feeds: MOR-1090 (visual/pixel gate) and MOR-1099 (cutover review).
Sibling of: docs/validation/workspace-v1-verification.md (same directory, same "verified, not
asserted" contract). Its two ruling siblings are docs/plans/2026-08-06-settings-modal-boundary.md
(S10) and docs/plans/2026-08-06-panel-order-workspace-boundary.md (S11); where they ruled, this
document counts.
0. What this is, and the one rule it follows¶
The zone-ownership wave (S6-pre … S6b-2) moved ten legacy panel families off desktop-v2 and put
fourteen semantic surfaces in their place, one zone each. Every slice landed with its own tests and
its own independent verification. This document is the single place where the operator-visible
consequences of all of them are stated together, so that the cutover decision is made against a
list rather than against a feeling.
The rule: every row carries an evidence pointer, and a row whose evidence contradicted the planning seed says so out loud. Six rows did. They are collected in §3 and are the most important part of this document, because each of them was, until now, a fact the program believed and could have shipped on.
Verdict vocabulary (fixed, five values):
| verdict | means |
|---|---|
parity |
the operator sees the same fact, by the same or better means |
improved |
the semantic surface is strictly better on this axis |
recorded lack |
something the legacy family showed is not shown; ticketed or owner-batched |
changed derivation |
the fact is still shown but computed or presented differently |
by design |
a deliberate boundary, stated so it does not get re-litigated as a defect |
1. Harness blindness — stated once, and only once¶
Four zeroes are not four confirmations.
S6a, S7, S8, S9 and S6b-2 each reported "zero movement" from the browser capture harness. Every one
of those zeroes is structurally empty, for the same reason, and this section is the only place that
reason is recorded. Do not repeat it per row, and do not read any harness number in §7 as
desktop-v2 parity evidence.
The harness (frontend/fixtures/) mounts exactly two components — frontend/fixtures/main.ts:112:
RadioLayout is never mounted. ReferenceLayout.svelte:32 mounts <SemanticRadioSurfaces
strips="single" /> bare, and says why in its own header (:7-14): pulling in RadioLayout "would
pull in that whole chrome tree… for zero additional coverage". That judgement was correct when the
semantic surfaces had no chrome-side counterpart. It stopped being correct the moment the wave's
whole subject became the chrome-side suppression arm.
Consequences, all verified at 7612a4f8:
- No legacy twin is ever rendered by the harness, so no suppression can be observed.
desktopV2Layoutis never resolved. MOR-1355 (fe01a1f1) did land and does supply aSurfacePlan— but opt-in, for exactly one fixture (catalog.ts:814-818,topology-2-main-sub--planned), and it resolvesdualReceiverCockpitLayout, neverdesktopV2Layout(main.ts:104-109;capture.mjs:542-548says so: "the--referencefamily … has no plan-ful twin in this slice"). The seed's wording "supplies noSurfacePlan" is now out of date; the structural conclusion it supported is not.- The harness's own build-identity ledger proves the blindness.
manifest.json'sbuildIdentity.productionSourceDigestsfingerprints fifteen files. It includespresentation/layouts/dual-receiver-cockpit.ts. It does not includepresentation/layouts/desktop-declarations.tsorcomponents-v2/layout/RadioLayout.svelte— the two files the entire wave edited. A harness that does not hash the changed file cannot report on it.
What is therefore NOT proven by any harness run in this program: that desktop-v2 renders the
zone-owned surfaces in the intended order, that the legacy twins are gone from its DOM, or that the
composite looks acceptable. Those claims rest on jsdom component tests
(semantic-desktop-migration.component.test.ts is the load-bearing one) plus MOR-1090's pixel gate.
Widening the harness to mount RadioLayout under desktopV2Layout is the obvious follow-up and is
not in this slice.
One further evidence-shape caveat, from the S6b-2 verification, that belongs here rather than in a
row: the scope-toolbar suppression is proven as a two-link chain, not end to end. The
manifest → prop wire is asserted at RadioLayout against a stub, and the prop → DOM behaviour is
asserted on SpectrumToolbar in isolation. Nothing mounts the real toolbar under the real manifest.
2. The parity matrix¶
Twenty-three rows plus three sub-rows (P1r, P13c, P13r — the last two added by verify-mor-1372
when the MOR-1358 row was split; see §3.5). "Evidence" is a test name, a file:line, or a verify
report; "disposition" is what
happens next. Rows P10, P11 and P12 are safety adjudications — see §6.
| # | family / row | verdict | evidence | disposition |
|---|---|---|---|---|
| P1 | Meters peak-hold (MOR-1282). The seed recorded this as a lack. It is not one at 7612a4f8: MetersSurface enables peak-hold per meter (MetersSurface.svelte:57-64, showPeak true for Po/SWR/ALC/Id, mirroring the dock's own PeakKey set) and delegates the ballistics to BarGauge, which holds the state (BarGauge.svelte:63,76), decays over the shared PEAK_DECAY_MS = 1500 (meter-utils.ts:361) and resets on ondblclick (BarGauge.svelte:120) |
parity |
MetersSurface.test.ts:467 "BarGauge peak channel (MOR-1282)"; :493 "MetersSurface and MetersDockPanel are on the same peak-hold channel"; BarGauge.peakhold.svelte.test.ts:57 |
closed. See §3.1 for the correction, and the residual below |
| P1r | Peak-marker relevance gate (residual of P1, new here). The dock hides the marker on an irrelevant tile (MetersDockPanel.svelte:422, && tile.relevant); BarGauge has no relevance gate, so an RX-dimmed Po tile shows a peak marker on the semantic surface and none in the dock |
recorded lack (cosmetic) |
MetersDockPanel.svelte:422 vs MetersSurface.svelte:145-148,168; no test either side |
owner batch — cosmetic, sub-tile |
| P2 | SWR/ALC fault border (MOR-1345). The dock draws a fault border for SWR > 2.0 and ALC ≥ redline during TX; nothing in the semantic path reproduces it. Note the seed's framing is wrong: fault is not a tile-component prop, it is a field on MetersDockPanel's own local Tile type — MeterPanel and DockMeterPanel never had it |
recorded lack |
legacy: MetersDockPanel.svelte:75,272,286,400,495 + meter-utils.ts:327,332, pinned by MetersDockPanel.isolated.test.ts:433. Semantic: absent — no test pins presence or deliberate absence |
MOR-1345 |
| P3 | Whole-group-absent meters (S5-N1). A radio reporting no meter fields renders nothing on the semantic surface (MetersSurface.svelte:112, {#if meters} around the whole section). The seed said the legacy dock showed ? placeholders. It did not — it rendered a titled empty shell (MetersDockPanel.svelte:385-392 are unconditional; tiles are push-gated at :243-323). The ? is the semantic surface's per-field behaviour (MetersSurface.svelte:132,150) |
changed derivation |
MetersSurface.test.ts:127 "renders NOTHING at all when the view model carries no meters group"; semantic-meters-wiring.component.test.ts:244 |
accepted, direction corrected — §3.3 |
| P4 | SUB tuning on dual-receiver radios (MOR-1335). isActive is a single radio-wide fact (radio-view-model-adapter.ts:1200-1203,1255), so at most one VFO tile per radio is tunable. VfoSurface.svelte:226-228 gates the per-digit control on it, and the handler re-guards at :231. With state.active === 'MAIN' both SUB tiles are read-only. The legacy VfoPanel mounted one tunable widget per receiver |
recorded lack |
VfoSurface.svelte:208-214 (the gap is self-documented in-file); VfoSurface.test.ts:794,843; radio-view-model-adapter.test.ts:175 |
MOR-1335. The gate is deliberate — it closed a wrong-VFO dispatch hazard (MOR-1322 B1) — so the fix must add a receiver axis, not remove the gate |
| P5 | Translation delta. Corrected: the retired legacy family carries 4 real t() call sites across 2 of its 11 components, and the semantic family carries 18, all in VfoSurface. Three locale keys lose their only rendering surface on desktop-v2. See §5 for the full count and the grep artefact that produced "42" |
recorded lack (small) + improved (VFO) |
§5, with the counting method | MOR-1373 — retargeted; it is a 3-key gap plus a 13-surface hardcoding policy question, not a 42-string regression |
| P6 | Audio-link-lost readout (new in this document). RxAudioPanel.svelte:53-54 renders an output indicator that reads core.overlay.audioLinkLost when live monitoring is selected and the browser audio transport is down, and formatMonitorStatus(monitorMode) otherwise. RxAudioSurface has no equivalent element (data-testid sweep: monitor, af, focus, split, mod-input — no transport-health readout), and a repo-wide grep finds no other consumer of that key. On desktop-v2 the operator no longer sees that the audio link dropped |
recorded lack |
RxAudioPanel.svelte:18-20,53-54; RxAudioSurface.svelte:113-205; grep -rn audioLinkLost src/ → 1 consumer + 3 locale files |
needs a ticket. Arguably out of the facts-only doctrine (transport health is a client fact, not a radio fact) — which explains the omission but does not restore the readout |
| P7 | Deck relocation. Ten legacy panel families are gone from the sidebars and fourteen semantic surfaces now render inside .receiver-deck. See §4 for the corrected geometry and the surviving sidebar contents |
changed derivation |
§4 | accepted with consequence under the MOR-1263 owner decision; recorded for MOR-1099 |
| P8 | SPLIT is now disabled on unknown split status. The settings modal's SPLIT <HardwareButton> was never disabled (RadioLayout.svelte:425-434; active collapses unknown into the grey/false branch). VfoSurface's replacement is disabled={viewModel.split.status === 'unknown'} (:388) with a handler re-guard (:254-257) and aria-checked="mixed" (:386). On a radio that never reports split status the operator could press it before and cannot now |
by design (fail-closed) |
VfoSurface.test.ts:200 "unknown split renders an explicit 'unknown' tri-state" (the disabled assertion at :212 was added because mutation M14 deleted it and nothing failed); :319 "MUTATION KILL: clicking a disabled (unknown) split toggle never emits the intent" |
none — the wave's fail-closed doctrine, recorded as its own row per S10 §6 so it is never absorbed into "improvement" |
| P9 | Scan type / resume render as raw integers (MOR-1308 N4). RitXitScanSurface.svelte:173,176 print String(value) via textOf (:56-57), so 0x22 reads 34 and a resume mode reads 1. ScanPanel.svelte:17-23,37-40 carried label tables (MEM, SEL, 5s, 10s, 15s) rendered at :88,:123 |
recorded lack |
file:line above; no test pins the rendering — RitXitScanSurface.test.ts:376-393 pins only structural presence |
owner batch item. Deliberate and self-documented (RitXitScanSurface.svelte:39-43: the label tables are UI-only in v2) |
| P10 | deriveBand reads state.active with no freshness gate (MOR-1307 N4). radio-view-model-adapter.ts:705-707 reads state?.active === 'SUB' raw; the sibling activeReceiver derivation at :1200-1203 requires seen() (observed and fresh and available, :81-85). With active unobserved or stale, onSub silently defaults to MAIN and the adapter publishes currentBandTx: 'allowed' while simultaneously emitting { field: 'activeReceiver', code: 'field-not-observed' } (:1277-1278) |
recorded lack — safety-adjacent |
file:line above; no test exercises the ungated path — band-adapter.test.ts:348 inherits a fresh fieldStatus |
MOR-1356 (sole source: this document and verify-mor-1307) |
| P11 | RF_MUST_BE_IDLE is an allow-list over 3 of 7 reasons (MOR-1309 N1). AntennaSurface.svelte:75-77 lists tx-busy, radio-transmitting, rf-state-unknown; the KeyBlockedReason union (rx-tx-surface.ts:37-39) has seven members. The filter at :114-115 therefore ignores tx-target-unknown, tx-permit-denied, tx-permit-unknown and tx-fault. Adding an eighth member produces a silently non-blocking reason with no type error |
recorded lack — safety-adjacent |
file:line above; AntennaSurface.test.ts:303 "does not block on a permit or target reason alone" pins the narrowing as intended |
MOR-1361 (sole source). The narrowing is reasoned (:70-74); the residual is structural — no exhaustiveness check binds new union members to a decision |
| P12 | Break-in posture readout is not permit-gated (MOR-1310 6(c)). CwKeyerSurface.svelte:181 renders the posture <output> under the structural gate only, while the three break-in choices at :173 carry disabled={!usable(cw.breakIn) \|\| !permitAllowed}. Result: "break-in ARMED — the key transmits" beside three disabled buttons. Companion: unread level thumbs park at min (:196), so an unread keyer speed sits at 6 WPM |
by design |
CwKeyerSurface.test.ts:345 "renders armed-and-not-permitted differently from off-and-not-permitted"; :435 "renders an unread %s as unknown and parks the thumb at min" |
none — correct and deliberate. The radio can still key from its own front panel (:177-180); the readout reports the radio, the controls report the permit |
| P13 | Unknown announced as not-pressed (MOR-1358) — TOGGLES ONLY, and they are done. The defect is a two-state toggle emitting aria-pressed="false" over an unread reading, which claims "this control is OFF" about a fact the radio never reported. Every such toggle in the family now emits undefined: TxAuxSurface:113, DspSurface:124, RitXitScanSurface:138,144,167, RfFrontEndSurface:183, CwKeyerSurface:208,247, ScopeControlsSurface:176, AntennaSurface:181 (valueOf(ant.rxAnt) — the same shape under a different name, missed by every pressedOf grep) |
parity — closed |
shared helper + unit pin at semantic/pressed-of.ts / __tests__/pressed-of.test.ts; DOM omission pins at DspSurface.test.ts, RitXitScanSurface.test.ts ×3, CwKeyerSurface.test.ts ×2, AntennaSurface.test.ts:221; mutation-killed 8× (verify-MOR-1358 §4) |
MOR-1358 built, NOT YET LANDED at 0a1ecf60 (branch codex/mor-1358-shared-pressedof). Land it, then this row is closed. ScopeControlsSurface + AntennaSurface remain on local copies → MOR-1383, widened to two surfaces |
| P13c | Choice groups are a DIFFERENT question, and false there is correct — see §3.5. Sixteen mutually-exclusive options emit a definite boolean: six aria-pressed (FilterSurface.svelte:114,129,160, BandSurface.svelte:187, DspSurface.svelte:165,177) and ten aria-checked on role="radio" (RfFrontEndSurface.svelte:130,151 PREAMP/ATT, CwKeyerSurface.svelte:171,229, ScopeControlsSurface.svelte:120,133, RxAudioSurface.svelte:124,157,174, AntennaSurface.svelte:169). "This option is not the current selection" is true even when the selection is unknown; the unknown is carried by disabled + the adjacent <output> rendering ?/—. Two groups stay live under an unread reading — AntennaSurface:169 (disabled = TX-block only) and BandSurface:187 (disabled={!receiverKnown}) |
by design, with a residual |
FilterSurface.test.ts:188,415 and BandSurface.test.ts:334 assert 'false' and disabled — they pin intended behaviour, not a bug |
not MOR-1358. If the residual is worth a ticket it is "an unread choice group should disable", scoped to the two live groups. Do not omit aria-checked (required on role="radio") and do not use 'mixed' (unsupported there) |
| P13r | Filter-selection highlight is unpinned (MOR-1304 N1 / mutant MF6). isSelected(modeFilter.currentFilter, index + 1) (FilterSurface.svelte:129) is off-by-one-able: no test asserts which filter reads pressed, so index for index + 1 survives. This is a coverage gap, not an aria-semantics one — the seed's row 11 cited it as MOR-1358 evidence and it was lost when the two were merged |
recorded lack (test coverage) |
verify-mor-1304 N1/MF6; FilterSurface.test.ts has click tests at :222,278 but no positive selection pin |
needs a one-line pin, or fold into MOR-1358's landing |
| P14 | The workspace still cannot hide a zoned optional surface (MOR-1337), and this wave enlarges the residual by ten. zoneOwning() returns null both when no manifest zone declares a surface and when a zone declared it and the workspace subtracted it (the plan handed in is post-subtraction, resolution.ts:165); zoned() renders null bare rather than withholding (SemanticRadioSurfaces.svelte:652-660). Subtraction costs the zone wrapper, never the readout — and never resurrects the legacy twin, which gates on the manifest (RadioLayout.svelte:140-142) |
recorded lack |
self-documented at SemanticRadioSurfaces.svelte:129-136; tests pin the wrapper appearing (semantic-desktop-migration.component.test.ts:1273-1283) and none asserts a hide |
MOR-1337. The fix needs SurfacePlan to carry declaration and visibility separately |
| P15 | sdr-test double-presentation, final position. sdr-test declares only main: ['vfo','rxTx'] (declarations.ts:21), so all ten declared-channel legacy panels keep rendering and the twelve undeclared semantic surfaces render bare. Exception the seed missed: TxPanel does not survive there — hideTxPanel derives from declared.has('vfo'), which is true on sdr-test, so the TX panel is suppressed on the separate R9 channel |
recorded lack — out of scope |
semantic-desktop-migration.component.test.ts:1049-1062 asserts all ten twins render on sdr-test; RadioLayout.svelte:90,112 for the TX exception |
needs a ticket. Per S10 §9(2), widen to cover the settings-modal family too: rows 1-6 of the S10 table are the same class; its rows 8-10 were never suppressed on any manifest and must not be counted in |
| P16 | Shells share components but not the suppression channel — three instances. (a) LcdLayout.svelte:55 mounts <StatusBar /> with no declared prop, so StatusBar.svelte:271's scope-indicator suppression can never fire under the three LCD skins; (b) MobileRadioLayout.svelte:493,636 mount <SpectrumPanel /> without hideScopeControls and :689 mounts <BandSelector /> without hamBands — the file imports no declaredSurfaces/getLayout at all; (c) both sidebars use one un-namespaced storage key each (LeftSidebar.svelte:52 rigplane:panel-order, RightSidebar.svelte:38 rigplane:right-panel-order), shared across every skin that mounts them |
recorded lack — structural |
file:line above; RadioLayout.svelte:311,363,460 for the contrast; S11 doc for (c) |
MOR-1380. One pattern, three instances — a component that reads a suppression channel must be given it by every shell that mounts it |
| P17 | known-defaults now permanently mis-describes desktop-v2 (S11). drag-reorder.svelte.ts:48-49 records every default as "presented" on every load, regardless of the declared render gate, and :67-78 treats a known-but-absent id as "deliberately removed by the user" |
changed derivation |
drag-reorder.svelte.ts:44,48-49,67-78; S11 doc §1.2 |
accepted divergence. Precision note: the ten ids are recorded as known, not as removed — they are still in the stored order. The hazard fires only if a zone declaration is later withdrawn, and is recoverable via "Reset panel order" |
| P18 | scopeControls.fixedEdge — a registered leaf with no fact-layer home. The wire produces it (state_schema.py:167, runtime_helpers.py:204, lib/types/state.ts:108) and the field-status fixture carries an entry for it, but no adapter or view model emits it and validateScopeControls's exactKeys (radio-view-model.ts:1513-1515) lists twelve leaves without it |
by design |
exclusion reasoned in-file at radio-view-model.ts:763-774 ("a composite validator with zero consumers is exactly the speculative surface 'no speculative keys' forbids"); restated at ScopeControlsSurface.svelte:29 |
gap-register entry. The wire field remains; the fact layer declines it deliberately |
| P19 | Status-bar scope indicator. Retired on desktop-v2 (StatusBar.svelte:271, !declared.has('scopeDisplay')). It carried a translated tooltip — title={t('core.statusbar.indicator.scope', …)} → "Scope WebSocket: {state}" — and a coloured dot (:183-194, green/yellow/red/dim). ScopeDisplaySurface.svelte:44-55 reproduces the tone taxonomy but has no title (only aria-label="Scope status", :82), and emits tone as a bare data-tone attribute with an explicitly colourless stylesheet (:100-107) |
recorded lack (tooltip) + changed derivation (colour → design language) |
file:line above | owner batch. The colour loss is the DL contract working as intended; the tooltip loss is not, and the readout is richer (SRC / health / HW on-off) than the dot it replaced |
| P20 | Spectrum-toolbar boundary. Twelve fact-backed leaves retire under hideScopeControls (MODE, EDGE 1-4, SPAN, SPEED, HOLD, REF, DUAL + MAIN/SUB, the mobile REF row, and the settings-gear popover). Nine client-side view options render unconditionally: AVG, PEAK, BRT (×2 rows), colour scheme, BANDS, layers/region, EiBi, fullscreen — plus VIEW ON/OFF. The STEP tuning-step group (:184-199) also survives unconditionally but is not a view option and was never in the ruling's scope |
by design |
SpectrumToolbar.svelte:40-73 states the rule; retiring set at :213-284,323-330,393-401, surviving set at :180-212,293-347,348-406; SpectrumToolbar.component.test.ts pins both directions |
none. Correction: the S10 ruling names eight (enableAvg/enablePeakHold/brtLevel/colorScheme/fullscreen/showBandPlan/hiddenLayers/showEiBi, S10 §:93-94); there are nine — VIEW ON/OFF (:205-212) is not in the S10/S6b list. Do not count STEP into either number |
| P21 | 11B derivation gaps — closed by verification. The scope stepper clamps and the EDGE/SPAN predicate domain were unpinned; an unclamped set_scope_span {span: 8} could reach the wire. ScopeControlsSurface.svelte:40,103,106,109 now routes every stepper through clampSpan/clampSpeed/clampRef and the predicates through isSpanApplicable/isEdgeApplicable |
parity — closed |
verify-mor-1311 F1/F2; code as cited |
closed. Kept in the matrix as evidence of method: a mutation battery, not a review, found it |
| P22 | Counting corrections that entered the record. Broadcast presets are 16, not 17 (broadcast-presets.ts:11-16 6 LW/MW + :21-30 10 SW; the 17th hit was the interface's own name: field at :2). Declared-retired panel ids on desktop-v2 are 10, not 11 (band still renders — it is retired by prop, hamBands={!declared.has('band')}, never by mount) |
parity — bookkeeping |
BandSelector.svelte:18-19 states the 16 in-file; semantic-desktop-migration.component.test.ts:1028-1029 is the authoritative ten-id literal |
recorded; both corrections already propagated into the S8 and S11 artefacts |
| P23 | No drag-reorder for zone-owned surfaces. zoneOrder/visibleSurfaces are workspace-settings-only; SemanticRadioSurfaces renders zone-owned surfaces in fixed source order with no drag handle. The operator loses the sidebars' reordering affordance for every surface a zone retires |
recorded lack |
S11 doc §1.4 / §4 | deferred, unticketed. Route through MOR-1162 if picked up — see §9 |
3. Where the evidence contradicted the seed¶
Six corrections. Each was a belief the program held going into this slice.
3.1 MOR-1282 peak-hold had already landed (P1)¶
Both seed matrices list it as a recorded lack. It is parity at 7612a4f8. The behaviour did not
get re-implemented in MetersSurface — it moved down into the shared BarGauge, which both the
semantic surface and the legacy dock now sit on, and MetersSurface.test.ts:493 asserts exactly
that ("…are on the same peak-hold channel"). Copying the seed row unread would have shipped a false
regression claim into the cutover review.
3.2 "42 t() calls" is a grep artefact (P5, §5)¶
The number is a raw substring count of t(. It counts getShortcutHint(, parseInt(, setTimeout(,
repeat(, onBandSelect(, getBoundingClientRect( and friends. Nine of the eleven "retired
components carrying t() calls" do not import t at all. Full method and the corrected numbers in §5.
3.3 The ? placeholder belongs to the new surface, not the old one (P3)¶
S5-N1 was recorded as "the dock showed ? placeholders, the surface shows nothing". The dock has no
? anywhere in its markup; its policy is omission (MetersDockPanel.svelte:234-235), and with zero
meter fields it renders a titled, empty STATION METERS shell. The ? is MetersSurface's own
per-field honest-unknown rule (:21-22,132,150). The honest statement of the row is therefore:
whole-group absent → the dock showed empty chrome, the surface shows nothing; per-field absent →
the dock omitted the tile, the surface shows ?. The direction of the change is the opposite of
the one recorded on one axis and better than recorded on the other.
3.4 The receiver deck does not scroll (P7, §4)¶
Both seeds describe a "280px overflow:auto receiver deck". The 280px is real, but it is a
grid-row height (RadioLayout.svelte:509-511), and .receiver-deck is
overflow: hidden (:558). overflow-y: auto at :581 belongs to .content-left, .content-right
— the sidebars. This matters: the accepted consequence is not "ten surfaces you can scroll through",
it is "ten surfaces in a fixed 280px box that clips".
3.5 The seed's MOR-1358 row conflated three different things (P13 / P13c / P13r)¶
Corrected by the independent MOR-1358 verification (verify-mor-1358, which adjudicated this row
against the built fix). The seed cited verify-mor-1304 N1 and verify-mor-1306 N1 under one
heading; those two reports are about different defects, and a third population was then swept in.
- Toggles emitting
aria-pressed="false"over an unread reading — the real MOR-1358. A toggle has "unread" and "off" as distinguishable states, so"false"there is a fabricated positive claim. This is fixed. All eleven toggles in the family emitundefined; the rule now lives in one shared helper with a unit pin plus DOM omission pins on four surfaces, and it survives an eight-mutant battery. NoteAntennaSurface.svelte:181(valueOf(ant.rxAnt)) is an eleventh toggle that everypressedOfgrep in this program missed — it was already correct. - Choice groups are not the same defect. For a mutually-exclusive group,
falseon a non-selected option is a true statement — "this option is not the current selection" holds whether or not the selection is known — and the unknown is communicated bydisabledplus the adjacent<output>rendering?/—. Two further points make the "fix it the same way" reading wrong on the merits:aria-checkedis a required property ofrole="radio", so omitting it is invalid ARIA rather than honest; andaria-checked="mixed"is not a supported value onrole="radio". The in-repo exemplar the seed pointed at —VfoSurface.svelte:244-247'striState— sits onrole="switch"(:385,:397), a different role with a different contract, so it is not transferable to the ten radio sites. The three tests asserting'false'assertdisabledalongside it: they pin intended behaviour, and MOR-1358's built fix deliberately and correctly excludesFilterSurfacefor exactly this reason. - The filter-selection off-by-one (MOR-1304 N1 / MF6) is a coverage gap, not an aria-semantics one, and it is still open. It is now P13r rather than being absorbed.
Count correction: the aria-checked population named here is ten sites, not nine (2 + 2 + 2
+ 3 + 1); the family carries twelve aria-checked bindings in total, the other two being
VfoSurface's tri-state switches.
3.6 The sidebars keep band, not tx (P7, §4)¶
The planning brief said the sidebars are left holding audio-scope, memory and tx. tx is
suppressed on desktop-v2 — hideTxPanel derives from declared.has('vfo')
(RadioLayout.svelte:90,112) and is true. The third survivor is band, in its broadcast half only.
The later handoff had this right; the brief did not.
4. Deck relocation — accepted, with the consequence written down¶
desktopV2Layout declares 14 zones, one surface each (desktop-declarations.ts:36-131), which
is the entire SEMANTIC_SURFACE_NAMES vocabulary (contract.ts:29-32). In manifest terms the zone
literally named receiver-deck owns exactly one surface, vfo (:37); in DOM terms the
<section class="receiver-deck"> hosts all fourteen, because it mounts <SemanticRadioSurfaces />
(RadioLayout.svelte:314-316). Both statements are true and they are routinely confused — the count
that matters to an operator is fourteen surfaces in one box.
Geometry, corrected (§3.4): .radio-layout.semantic-deck sets grid-template-rows: 28px 280px
minmax(0,1fr) auto (RadioLayout.svelte:509-511, restated for ≥1680px at :515-517; the
non-semantic baseline is 200px at :507), and .receiver-deck is overflow: hidden (:558).
What is left in the sidebars, verified gate by gate:
| sidebar | survivor | why it survives |
|---|---|---|
| left | BAND (broadcast half only) | LeftSidebar.svelte:107-113 has no mount gate; the HAM half retires through the hamBands={!declared.has('band')} prop (:111), because a mount gate "would orphan the presets" (:99-106) |
| right | AUDIO SCOPE | RightSidebar.svelte:51 — capability-gated on hasAudioFft() only, never on declared |
| right | MEMORY | RightSidebar.svelte:75 — ungated; not a semantic family |
Everything else is gone: RF FRONT END, MODE, FILTER, AGC, RIT/XIT, ANTENNA, SCAN, RX AUDIO, DSP, CW
(the ten of RETIRED_ON_DESKTOP_V2, semantic-desktop-migration.component.test.ts:1028-1029), plus
TX on the separate R9 channel. The live-DOM inventory test at :717-726 asserts the resulting
desktop-v2 panel id list is ['band','desktop-language','desktop-vfo-ops','desktop-workspace',
'memory','memory'].
Disposition: accepted with consequence, under the MOR-1263 owner decision. It is recorded here — not resolved here — so that MOR-1099 reviews a stated trade rather than discovering it. The trade is: fourteen control-bearing surfaces stacked in a clipped 280px band, two near-empty sidebars, and no placement mechanism in this program. Placement is MOR-1162's problem; P23 is its companion row.
5. The translation delta, counted properly¶
Method. A real t() call site is t( not preceded by an identifier character or a ., in a
file that imports t from $lib/i18n. The seed's number is grep -o 't(' | wc -l. Both were run
over the same eleven files; the difference is the whole finding.
| component | seed (raw t() |
real call sites | imports t? |
|---|---|---|---|
ModePanel.svelte |
9 | 3 | yes |
FilterPanel.svelte |
8 | 0 | no |
DspPanel.svelte |
9 | 0 | no |
BandSelector.svelte |
6 | 0 | no |
RfFrontEnd.svelte |
3 | 0 | no |
RitXitPanel.svelte |
3 | 0 | no |
RxAudioPanel.svelte |
3 | 1 | yes |
ScanPanel.svelte |
1 | 0 | no |
AgcPanel / AntennaPanel / CwPanel |
0 | 0 | no |
| total | 42 | 4 | 2 of 11 |
The 38 phantom hits are ordinary identifiers ending in t: getShortcutHint(, onBandSelect(,
bandShortcut(, setTimeout(, clearTimeout(, toggleNrShort(, getBoundingClientRect(,
repeat(, gradient(, effect(.
The semantic family does not carry zero. VfoSurface.svelte imports t (:34) and has 18
real call sites over 14 distinct core.vfo.* keys, all 14 translated in ru-RU and ja-JP.
(Re-counted independently in verify-mor-1372: 21 raw t( hits, 18 real, 14 keys.) The other
thirteen of the fourteen semantic surfaces hardcode English.
What an operator on ru-RU/ja-JP actually loses on desktop-v2: three keys, all currently
translated in both locales.
| key | en-US | where it was | replacement |
|---|---|---|---|
core.modePanel.modInputLabel |
MOD IN |
ModePanel.svelte:103 |
RxAudioSurface.svelte:188 renders MOD: — hardcoded |
core.modePanel.modInputAria |
Modulation input source |
ModePanel.svelte:107,108 |
none (no aria-label on the semantic row) |
core.overlay.audioLinkLost |
Audio link lost — reconnecting… |
RxAudioPanel.svelte:54 |
none — see P6 |
npm run i18n:check → OK (3 locale file(s) checked, 204 source keys), with ru-RU/ja-JP each
missing the same 19 core.settings.workspace.* keys (informational; missing keys fall back to
English).
Retarget MOR-1373 accordingly. It is not "42 strings regressed". It is (a) three keys with no
translated home, one of which has no home at all, and (b) a policy question: thirteen semantic
surfaces hardcode English while VfoSurface proves the family can be translated. (b) is the
expensive half and the one that needs an owner decision.
6. Safety adjudications¶
P10, P11 and P12 are the rows that change what an operator may believe about whether the radio can transmit. This document's author must not be their reviewer (S12 S-class rule). For the independent reviewer, the three questions are:
- P10 — is "band-scoped TX permit computed under an unobserved active receiver, silently
defaulting to MAIN" acceptable until MOR-1356 lands? Note the second-order finding: the freshness
asymmetry is doubled, because
topFieldAvailable→isFieldAvailable(lib/state/field-status.ts:63-68) checks onlyavailability, dropping theobserved/freshlegsseen()requires. No test exercises the ungated path. - P11 — is an allow-list of blockers the right shape at all? The four omitted reasons are
omitted deliberately and reasonably, but nothing binds a future eighth
KeyBlockedReasonto a decision. MOR-1361 should be scoped as "add an exhaustiveness gate", not "add the four". - P12 — confirm the ruling stands: posture reports the radio, the controls report the permit, and showing "ARMED" next to disabled buttons is correct because the operator's radio can still key from its own front panel.
P9 (raw scan integers) is the fourth carry-forward and is an owner batch item, not a ticket: it is deliberate, self-documented, and a label-table port is a v2-UI concern the facts-only doctrine deliberately declined.
7. Harness evidence¶
Two full capture.mjs runs at 7612a4f8, same worktree, same port 5199, out-dirs distinct.
Read §1 first: this measures the harness's determinism, not desktop-v2 parity.
(There is no npm script; capture.mjs starts its own vite server from vite.fixtures.config.ts and
closes it, so no external dev server is needed.)
Claim, in the MOR-1351 shape: 61 captures, 0 invalid, 1830/1830 assertions, zero ok-flips.
Beyond the required claim, the two manifest.json files are identical except generatedAt — a
key-by-key diff of every top-level field, every capture record (valid, assertionsPassed,
assertionsTotal, consoleErrors, focusPath, tokens, paint) and every individual assertion
object produced zero differences. The explicit list of assertions whose value moved is
therefore empty.
The expected large movement the S12 brief predicted — 33 zone-less controls becoming zoned,
zonelessControls pins flipping, antenna ports going [disabled] with tuner-not-ready — did not
occur, and could not have. MOR-1355's plan-ful fixture is a single opt-in capture
(topology-2-main-sub--planned--desktop) resolving the cockpit manifest, and it was already the
baseline shape before these two runs. The prediction assumed a harness that resolves
desktopV2Layout; §1 explains why none exists.
Environment note for whoever re-runs this: one of three attempts died mid-run — once on a
Chromium crash during a screenshot (capture.mjs:440) and once on SIGKILL (exit 137). Neither is a
determinism signal; both are resource pressure. Re-run before drawing any conclusion from a partial
manifest.
Manifest metadata worth carrying: ticket is still MOR-1070, harness.productionFilesChanged is
0, and harness.stubbedSeams is four entries ($lib/runtime, the tx-controller app-host, the
mod-input TX guard, the command bus).
8. Closure statements¶
8.1 MOR-1317 — closed program-wide¶
Verified directly rather than inferred from an empty ledger (verify-mor-1370 §3, re-checked here):
| assertion | result |
|---|---|
declaredSurfaces(desktopV2Layout) ⊇ every SEMANTIC_SURFACE_NAMES member |
zero missing |
owned.size === SEMANTIC_SURFACE_NAMES.length (14) |
✔ |
desktopV2Layout.zones.length === 14 |
✔ — one zone per surface |
zoneOwning('scopeControls') === 'scope-controls' on a default workspace |
✔ |
| the same on a pre-S6b-2 stored workspace that never heard of the zone id | ✔ (the migration case) |
the cockpit does not own scopeControls |
✔ — manifest untouched |
zone-ownership-coverage.test.ts remains a live guard rather than a tautology: with zero excused
entries the partition still fails the moment a fifteenth name is appended to
SEMANTIC_SURFACE_NAMES.
8.2 MOR-1263 §3 — the planner's resolved-item list¶
| item | disposition at S12 |
|---|---|
| S6 split into S6a + S6b-1/S6b-2 | done as planned |
| Spectrum-toolbar boundary (client-side view options stay legacy) | done — P20, with the nine-not-eight correction |
| MOR-1355 sequenced immediately before S12 | landed (fe01a1f1); §1(2) records what it does and does not reach |
| MOR-1317 closes at S9 (sidebars) and S6b-2 (scope) | closed — §8.1 |
| MOR-1339 encoded as §1.5 in every zone brief | held; the exactly-one-instance pins are live (semantic-desktop-migration.component.test.ts:1359,1432; cw-keyer-zone-lifecycle.component.test.ts:207) |
| S10 first, S6b seventh | held |
| 3-file guardrail waived once, explicitly, in S6-pre | held — one waiver, named at landing |
8.3 This slice¶
0 production LOC. One new documentation file; no test, no source, no locale file changed.
9. Open items handed to the owner / MOR-1099¶
- Two unticketed deferrals must be filed or explicitly confirmed as deferred before MOR-1263 closes. (a) MOR-1162 routing for zone drag-reorder (P23 / S11 §1.4); (b) the MOR-1294 confirmation named as S8's landing gate. Both currently read "route it if picked up", which is not a disposition.
- P6 (audio-link-lost) has no ticket. It was found by this slice.
- P15 needs a ticket, and its language needs widening past meters to the settings-modal family (S10 §9(2)), excluding that table's rows 8-10, which were never suppressed on any manifest.
- MOR-1373 needs retargeting from "42 untranslated strings" to §5's actual shape.
- MOR-1380 — file the cross-skin channel pattern as one structural ticket with three instances (P16), not three pieces of trivia.
- MOR-1358 must land before this row closes. The fix is built and independently verified at
0a1ecf60but is not onmain(P13). MOR-1383 must be widened to two surfaces (ScopeControlsSurfaceandAntennaSurface, which carries the same shape under the namevalueOf). P13r — the unpinned filter-selection highlight — needs a one-line test pin. - Harness widening — mounting
RadioLayoutunderdesktopV2Layoutinfixtures/main.tsis the only thing that would turn §7's zeroes intodesktop-v2evidence. Not in this slice; MOR-1090's pixel gate is currently the only visual authority over the reworked skin.
10. Environment¶
Node v26.4.0, Playwright 1.58.2, Svelte 5.55.8, jsdom via vitest 4.1.0.
- This Node/jsdom pairing needs both
--experimental-webstorageand--localstorage-file. Without them a large class of pre-existing tests fails on a missinglocalStorageglobal, which reads as a mass regression. Always use a fresh dedicated file per run. - Three standing flakes, each reproduced on unmodified
mainand none caused by this wave:mod-input-tx-guard.isolated > arms the guard at startTx and still delegates to runtime(green when run isolated);semantic-lcd-migration > "scope" resolves to the migrated LCD entrypoint; and an intermittentsemantic-desktop-migrationinventory row. - Gates run for this document (fresh dedicated localStorage file):
npx vitest run src/semantic src/components-v2/layout/__tests__/semantic-desktop-migration.component.test.ts src/components-v2/wiring→ 62 files, 1647 tests, 0 failures;npm run i18n:check→OK (204 source keys).