[ten9876]

Summary

Found while refreshing the GitHub wiki for v26.5.1 → v26.10.1: in-repo text that no longer matches the app. Each item was checked against current source before changing it.

In-app help (resources/help/)

• aethersdr-help.md: Title Bar section rewritten for the 52 px bar (feat(gui): unified 52 px title bar on Qt 6.12 window chrome — native captions on macOS/Windows, frameless on Linux (RFC #4764) #6198) — radio tabs with the heartbeat on the status dot, ☰ menu, + popover; no heartbeat lamp, lightbulb or Minimal Mode button. Tools → AetherTX… replaces Aetherial Audio; DIGI applet, overlay DSP button and status-bar applet ☰ removed. Default applet order now matches AppletPanel::kDefaultOrder (23 IDs, PHNE).
• understanding-noise-cancellation.md: BNR runs in-process on a local NVIDIA GPU (no Docker/container, not cloud), with a ~1 GB one-time download matching the Download button (docs/nvidia-bnr.md aligned too); NNR added as the seventh client method; RN2 Noise Floor slider; controls live in Settings → AetherRX… → AetherNR; unavailable methods are dimmed with a reason.
• configuring-aethersdr-controls.md: a single click on an off-screen indicator activates and recentres (Fix off-screen slice indicator centering #4795).

Docs

• README: APRS has a station table, not a map; the lightbulb line points to Help → Submit your Idea… 💡; TMate 2, Ulanzi Dial and AetherPad added to Tested Controller Devices.
• CONTRIBUTING: Help-menu idea route, per-launch logs under <config>/AetherSDR/logs/ (attach the newest; aethersdr.log links to it on Linux/macOS), Help → Support & Diagnostics… / File an Issue….
• BUILDING: QRhi failure shows the "Spectrum renderer unavailable" card (fix(gui): surface QRhi spectrum failures #4787).
• psk-reporter-city-lights.md: Warmth default is 0 throughout (kDefaultWarmth).
• rtty-decoder-implementation.md: the pane opens for RTTY only.
• DEVELOPER-GUIDE.md: theme tokens instead of hard-coded hex, per AGENTS.md.
• docs/agents/gui.md: naming examples use the CAT applet (CatControlApplet) and the current Support & Diagnostics log category labels.

UI strings

• Automation-bridge Access Token tooltip now agrees with docs/automation-bridge.md (export AETHER_MCP_TOKEN only in the launching shell, never in an MCP config). All four bridge tooltips on Radio Setup → Network link the hosted doc rather than a source-tree path.
• Ulanzi mapper fallback no longer says "Linux build only" (backends exist on all three platforms).
• File an Issue and Submit your Idea AI prompts fetch AGENTS.md instead of CLAUDE.md.

Packaging

• 60-hid-encoders.rules: TMate 2 (1721:0614, from kSupportedDevices in HidDeviceParser.cpp), and a note that AetherPad (2341:0266) is deliberately absent because it is the Arduino Giga R1's stock identity.

Testing

• AetherSDR + 7 related test targets built (lean build, Qt 6.12); radio_setup_label_theme_token_test, radio_setup_recording_mode_dim_test, radio_setup_max_power_field_test, radio_setup_region_field_test, hl2_pc_audio_lock_test, titlebar_headphone_mute_test, automation_menu_lookup_test pass.
• All Static-checks scripts pass, incl. gen_bridge_docs.py --check; colour ratchet +0; a11y warnings unchanged from base.
• No test or translation pins any changed string.
• Not run: the full suite or a live app launch.

🤖 Generated with Claude Code

[github-actions]

Accessibility check findings

This PR touches src/gui/ files with 9 accessibility issue(s).
Full details appear as inline annotations on the diff above.

Files scanned : 4
Files with findings: 1
Total findings : 9

See docs/a11y.md for the canonical
Qt patterns (setAccessibleName, QAccessibleValueChangeEvent,
QAccessibleInterface subclasses, etc.). Structural remediation
(keyboard navigation, custom QAccessibleInterface subclasses)
is tracked in #3288.

Findings are warnings only — the build is not blocked.

[aethersdr-agent]

Issue fit

No linked issue, and none is needed: this is a documentation-accuracy sweep, which GOVERNANCE.md does not route through an RFC. Reviewed against the PR body's own stated intent — "each item was checked against current source before changing it." I treated that as the load-bearing claim and spot-checked roughly two dozen of the individual assertions against the head checkout. Nearly all of them hold, several to the exact constant. No test is required: no behaviour changes, and the three source edits are string-only.

A docs PR's failure mode is a confidently-worded sentence that sends a user somewhere that does not exist, so I prioritised the claims with a reachable wrong answer (paths, defaults, VID/PIDs, enumerated lists) over the prose.

Scope

File(s)	What it changes	Claimed in body?	Verdict
resources/help/aethersdr-help.md	Title bar, Tools menu, overlay list, applet order, status bar	Yes	In scope
resources/help/understanding-noise-cancellation.md	BNR/NNR/RN2 rewrite, control locations	Yes	In scope
resources/help/configuring-aethersdr-controls.md	Off-screen indicator click behaviour	Yes	In scope
README.md, CONTRIBUTING.md	Help-menu routes, log paths, controller list	Yes	In scope
docs/BUILDING.md, docs/DEVELOPER-GUIDE.md, docs/agents/gui.md	QRhi card, theme tokens, CAT applet naming	Yes	In scope
docs/psk-reporter-city-lights.md, docs/rtty-decoder-implementation.md	Warmth default, RTTY-only pane	Yes	In scope
src/gui/{RadioSetupDialog,SupportDialog,TitleBar,UlanziDialMapperDialog}.cpp	Tooltip + AI-prompt + fallback strings	Yes	In scope
packaging/linux/60-hid-encoders.rules	TMate 2 udev rule	Yes	In scope

Everything in the diff is explained by the PR body. No CHANGELOG.md entry was added — correct. The three src/gui edits are string-only; I checked that no control flow, no settings key and no public surface moves with them. No new config key, capability or protocol verb, so no Constitution II/III or AGENTS.md persistence surface is touched.

Blockers

None.

The one candidate I expected to find was the new log path, and it survives. CONTRIBUTING.md sends reporters to ~/.config/AetherSDR/logs/ (and the macOS/Windows equivalents), and LogManager.cpp:323-324 builds exactly that:

    return QStandardPaths::writableLocation(QStandardPaths::GenericConfigLocation)
           + "/AetherSDR/logs/aethersdr.log";

GenericConfigLocation appends no org/app component, so there is no doubled AetherSDR/AetherSDR/ here — all three documented paths are right.

Nits

All non-blocking.

• resources/help/understanding-noise-cancellation.md:~255 — the BNR download size disagrees with the button it describes. The help text says the Download button "fetches NVIDIA's runtime (about 1.2 GB, once)". That number matches docs/nvidia-bnr.md:60, but the control the reader is looking at says otherwise — src/gui/AetherDspWidget.cpp:1639 sets the button text to "Download (~1 GB)" and :1783 the tooltip to "...(one-time, ~1 GB)". Since this PR's own governing rule is docs/agents/gui.md's "User-facing names match the on-screen labels," the help should probably read ~1 GB and docs/nvidia-bnr.md be reconciled separately.

• packaging/linux/60-hid-encoders.rules — AetherPad gains a README entry but no udev rule. kSupportedDevices (src/core/HidDeviceParser.cpp:7-24) has seven devices; after this PR the rules file has six. The missing one is the AetherPad at {0x2341, 0x0266}, which this same PR adds to README's Tested Controller Devices as working "over USB HID (hidapi builds)" — on Linux a non-root user cannot open it without a rule. I suspect this was a deliberate omission rather than an oversight, since 0x2341:0x0266 is Arduino's generic composite VID/PID and a blanket MODE="0666" on it would widen access to every stock Arduino Giga on the machine. Worth confirming either way; if it is deliberate, a comment in the rules file saying so would stop the next person filing it as a gap.

• CONTRIBUTING.md:~34 — the aethersdr.log pointer is a symlink with no platform guard. The new text says "aethersdr.log pointing at the newest" unconditionally. LogManager.cpp:255-257 implements it as QFile::link(candidate, symlink) with the link name dir + "/aethersdr.log". On Windows QFile::link produces a shortcut and wants a .lnk suffix, so the plain aethersdr.log likely never appears there — meaning the Windows half of that sentence may not describe what a Windows reporter actually sees. This is pre-existing product behaviour, not something the PR introduced; the PR only states it. Reasoned from code — I could not run it on Windows. Softening to "the newest file is the one to attach" would be true everywhere.

• src/gui/RadioSetupDialog.cpp:2009 — repo-relative doc path in an end-user tooltip. The tooltip now ends "See docs/automation-bridge.md.", which means nothing to someone running an installed binary with no source tree. The advice is correct and well-sourced — docs/automation-bridge.md:191-199 says exactly this, including "never in an MCP config file" — so only the pointer needs work; a URL would travel better.

What I tried to break

• Attacked the body's central claim ("checked against current source") on the most falsifiable items, and it held, repeatedly to the exact value: kDefaultWarmth = 0 (CityLightsShading.h:19); the 23-ID applet list matches AppletPanel::kDefaultOrder (AppletPanel.cpp:167-169) in both membership and order, PHNE included; TMate 2 0x1721:0x0614 matches HidDeviceParser.cpp:24; RN2 Noise Floor 0–50% matches Rn2SettingsModel::kMaxRxDryMix = 0.50f; the AetherNR method strip order (NR2, NR4, MNR, DFNR, RN2, BNR, NNR) matches kLabels at AetherDspWidget.cpp:253; "Six further WDSP tuning controls" is exactly the six in m_nnrAdvanced (:2187-2200).
• Hunted for an over-deletion in the rewritten lists, and found none. The overlay menu section dropped DSP and added Memory; SpectrumOverlayMenu.cpp:425-433 confirms the live button set is +RX, +TNF, Band, ANT, Display, Memory, DAX — and the help file's unchanged line 163 still carries +RX, so the list is complete rather than quietly truncated. Same check on the status bar: the ASR toggle is real (MainWindow.cpp:5778).
• Tried to falsify the RTTY "RTTY only" rewrite, which is the change most able to make a doc actively wrong. MainWindow_DigitalModes.cpp:367 gates on isRtty && RttyDecodeSettings::enabled(), and diglOffset no longer appears anywhere in the RTTY decoder or applet sources — so both the RTTY-only scope and the persisted-enable-flag caveat are accurate, and removing the DIGL/diglOffset paragraph was correct rather than a lost feature.
• Checked the claims that cite a UI affordance existing at all, since those are cheap to assert and easy to get wrong: "Spectrum renderer unavailable" card (SpectrumWidget.cpp:12891), Minimal Mode in View with Ctrl+Shift+M (MainWindow_Menus.cpp:1334), Reset Applet Order (:1311), Tools → AetherTX... (:1461), Help → Submit your Idea... 💡 / File an Issue... (:1699, :1705), the three-missed-beat alarm and right-click blink toggle and EXPERIMENTAL badge (TitleBar.cpp:298-341), NR2's Opus/SmartLink dim reason (AetherDspWidget.cpp:715, #1597) and NR4's Windows-LLVM tooltip (:279). All present.
• Verified the Ulanzi string change is justified rather than cosmetic. UlanziDialMacOSManager.cpp exists alongside the Windows manager, so dropping "(Linux build only)" from UlanziDialMapperDialog.cpp:254 is a fix, not a preference — the retained comment at :249-250 says "(Linux, Windows)" only about which thread the backend lives on, which is a different statement and still correct.
• Confirmed the &-mangling in the diff above is a harness artifact, not content. The ${PR_DIFF} fragments in "Support ${PR_DIFF} Diagnostics" and issues?state=open${PR_DIFF}per_page=100 are the task prompt eating &. In the head checkout the files correctly use a single & in prose, while the Qt menu action properly escapes it as "Support && Diagnostics..." (MainWindow_Menus.cpp:1721) — which is also what SupportDialog.cpp:36 titles the dialog. Nothing to fix.
• Ran the §6 preference check and came up empty. Every edit either corrects text to match shipped code or adds a device already in kSupportedDevices. No default value, keybinding, layout or confirmation changes — the one behavioural-sounding item, the token tooltip, tightens security advice to match an existing doc rather than changing what the code does.
• Could not verify, and am not reporting as findings: the NNR CPU figures ("~13% of one core", "about twice" for Premium, Raspberry Pi unsuitability) and "runs after the AGC" are measurements and architecture statements with no constant to grep; and I could not locate the mechanism that auto-disables client-side NR on CW/digital, so appending NNR to that pre-existing list is plausible but unconfirmed — VoiceModeGate.h turns out to gate DVK and Copy Assist, not NR. The macOS-only-rebuild caveat in BUILDING.md I also left unchecked. I have no build, no test run and no way to drive the GUI, so every observation above is reasoned from reading the head checkout, not reproduced at runtime.

Recommendation

Approve with nits. This is an unusually well-executed docs sweep — the enumerated lists match the code exactly rather than approximately, which is the part authors normally get wrong, and the three string edits are genuine corrections with citable authority behind each. The four nits are all one-liners and none of them misleads a user into a dead end; the BNR size and the AetherPad udev question are the two worth a reply before merge, and the AetherPad one may just need a confirming "deliberate" so it does not get re-filed later. Thanks for checking each item against source instead of against the old text — it made this reviewable.

---

_{🤖 aethersdr-agent · cost: $6.8521 · model: claude-opus-5}

[jensenpat]

Verdict

Approve. Docs and three UI strings match current source. No blocker.

Reviewed 3d367fe75d9fd5ae5febd1dd5f4832244a7e273e against origin/main (2a89515d6), which is ahead of the PR base. git merge-tree --write-tree applies cleanly. Required checks on this SHA are green: build, check-macos, check-windows, Static checks.

Issue fit: no linked issue. The body is a Principle VIII sweep so in-repo text matches the app. That is what the diff does.

Scope

Group	Files	What it does
In-app help	resources/help/aethersdr-help.md, understanding-noise-cancellation.md, configuring-aethersdr-controls.md	Title bar, menus, 23-ID applet order, overlay Memory, status-bar ASR, client NR (NNR, RN2 Noise Floor, in-process BNR), single-click off-screen recenter
Contributor docs	README.md, CONTRIBUTING.md, docs/BUILDING.md, docs/DEVELOPER-GUIDE.md, docs/psk-reporter-city-lights.md, docs/rtty-decoder-implementation.md, docs/agents/gui.md	Help-menu route, log dir, controller list, QRhi card, theme tokens, warmth 0, RTTY-only pane, CAT applet and Support category labels
UI copy	RadioSetupDialog.cpp, UlanziDialMapperDialog.cpp, SupportDialog.cpp, TitleBar.cpp	Token tooltip, Ulanzi fallback, AI prompts fetch AGENTS.md
Packaging	packaging/linux/60-hid-encoders.rules	TMate 2 1721:0614

No CHANGELOG entry. No default, binding, or protocol change. docs/agents/gui.md is infrastructure CODEOWNERS; the rest is reviewers.

Checked against source

• AppletPanel::kDefaultOrder is the same 23 IDs in the same order, including WFM and PHNE. Display names and tray abbreviations match makeEntry (PHN, WAV, CLK, Channel Strip, TGXL, PGXL).
• Overlay buttons are +RX, +TNF, Band, ANT, Display, Memory, DAX. No DSP row.
• Title bar kHeight is 52. The active tab dot is the discovery heartbeat; three missed beats alarm; right-click toggles blink; EXPERIMENTAL badge is present. Minimal Mode is View / Ctrl+Shift+M. Tools has AetherTX.... Help has Submit your Idea... and File an Issue....
• Off-screen left-click emits offScreenSliceCenterRequested (activate and center). Right-click opens the slice menu.
• kDefaultWarmth is 0. RTTY pane is isRtty && RttyDecodeSettings::enabled(); DIGL does not open it.
• Log categories are Discovery, Connection / Commands, Protocol / Status. Class is CatControlApplet.
• RN2 dry mix clamps to kMaxRxDryMix 0.50. AetherNR labels are NR2, NR4, MNR, DFNR, RN2, BNR, NNR. NNR Standard/Premium tooltips match the help (~13% of one core, about twice, not for a Pi). The code comment names six undocumented WDSP controls. Client NR disables itself when an audible slice is CW or digital (updateAetherDspModePolicy), NNR included.
• QRhi failure card title is Spectrum renderer unavailable. macOS text offers only the rebuild; other platforms also offer AETHER_NO_GPU=1.
• Token wording matches docs/automation-bridge.md (export AETHER_MCP_TOKEN in the launching shell, never an MCP config).
• TMate 2 {0x1721, 0x0614} is in kSupportedDevices. Ulanzi has macOS and Windows managers, so dropping "Linux build only" is correct.
• APRS roster is a station table (AprsStationList).

CI on this SHA is the verification for the string and udev edits. I did not rebuild locally; the changes are prose and literals, and Static checks already passed on 3d367fe.

Nits (non-blocking)

• Help says the BNR Download is "about 1.2 GB", which matches docs/nvidia-bnr.md. The button and tooltip say ~1 GB (AetherDspWidget.cpp). Rounding only; worth aligning later so the help matches the control.
• CONTRIBUTING.md says aethersdr.log points at the newest file on every platform. LogManager uses QFile::link, which on Windows wants a .lnk shortcut. The timestamped file is the reliable one everywhere.
• The Access Token tooltip ends with See docs/automation-bridge.md, a repo path an installed app does not have. The advice itself matches the doc.
• AetherPad (0x2341:0x0266) is in the README and not in the udev rules. That is the right omission: HidDeviceParser.cpp documents it as Arduino's generic composite identity, and a MODE="0666" rule would cover every stock Giga.

Follow-up (not a merge gate)

No new test pins the three changed strings. A small assertion on the token tooltip and the Ulanzi fallback would catch the next stale rewrite. Not required for this PR.

[ten9876]

Issue fit

No linked issue. Reviewed head 3d367fe75d9fd5ae5febd1dd5f4832244a7e273e against the body's stated documentation-accuracy intent and merge base df60f7579bd96a83772d3ad1a57aeec9c43ff4b6. The changes substantially achieve that intent: current menus, applet order, noise-reduction controls, RTTY gating, controller support, and reporting instructions replace stale descriptions. These are accuracy fixes with existing implementation authority, not a new feature or an unapproved UX preference; an RFC is not needed.

Scope

Files/group	Change	Claimed?	Verdict
resources/help/aethersdr-help.md	Current title bar, menus, overlay, 23 applet IDs and status bar	Yes	In scope
resources/help/understanding-noise-cancellation.md	Local BNR, NNR, RN2 slider, method availability and current control locations	Yes	In scope
resources/help/configuring-aethersdr-controls.md	Off-screen slice activation/recenter and context menu	Yes	In scope
README.md, CONTRIBUTING.md	APRS table, controller list, Help reporting routes and logs	Yes	In scope
docs/BUILDING.md, docs/DEVELOPER-GUIDE.md, docs/agents/gui.md	Renderer failure notice, theme-token guidance and current names	Yes	In scope
docs/psk-reporter-city-lights.md, docs/rtty-decoder-implementation.md	Warmth default and RTTY-only decoder/mark authority	Yes	In scope
Four src/gui/*.cpp files	Token advice, canonical AGENTS prompt URLs, Ulanzi fallback	Yes	In scope; strings only
packaging/linux/60-hid-encoders.rules	Existing TMate 2's 1721:0614 access rule	Yes	In scope

Everything in the diff is explained by the body. All eight commits were authored on October 7 and concern the stated sweep. No changed settings key, default, control flow, dependency, radio verb, capability field, architecture baseline or CHANGELOG entry.

Blockers

None.

Nits — explicitly non-blocking

1. CONTRIBUTING.md:36 — qualify the latest-log pointer claim for Windows. The new cross-platform paragraph promises aethersdr.log pointing at the newest launch. Both the Windows startup branch (src/main.cpp:650) and rotation path (src/core/LogManager.cpp:255) use QFile::link with the name aethersdr.log, rather than a .lnk name. Qt documents that a valid Windows shortcut requires a .lnk extension. The timestamped log is the dependable attachment path across platforms; qualify the pointer as Unix-only or simply direct reporters to the newest timestamped log. This is a documentation overstatement of existing code, not a new product regression. Native Windows behavior was not reproduced.

2. src/gui/RadioSetupDialog.cpp:2010 — give installed users a usable documentation destination. The updated Access Token tooltip was reproduced through Radio Setup's Network page and correctly says to keep the token out of MCP configs and shell history. Its final See docs/automation-bridge.md. points to a source-tree path an installed application does not provide. Use a hosted documentation URL, such as https://github.com/aethersdr/AetherSDR/blob/main/docs/automation-bridge.md, or omit the pointer. No change to credential handling is needed.

Empirical verification and attacks that held

• Preflight: no tests added or modified. Inspected all seven targets named in the body before execution. They use socket-free dispatch/state injection; the recording-mode test's demo backend is in-process, with no synthetic firmware peer. All seven passed: six in the restricted environment; radio_setup_recording_mode_dim_test initially timed out after PipeWire/PulseAudio permission errors and then passed in 0.76 seconds with normal local permissions. No claim that the full suite passed; no changed regression test needed mutation checking.

• Reused the existing lean Qt 6.12 shared-core build from the author's clean checkout, copying the app, core, tests and test registrations into the review scratch directory rather than rebuilding or writing to that checkout. Bridge build metadata stamps e00f790da; git diff e00f790da HEAD contains only docs/agents/gui.md and the udev rule. All compiled application sources, help resources and test sources are identical to reviewed head. This was an execution pass on those existing artifacts, not an independent fresh build.

• Drove identity pr6250-review on explicit socket /tmp/pr6250-demo-unrestricted/bridge.sock, fresh settings and AETHER_AUTOMATION_NO_TX=1. Connected only DEMO-0001. menu list, invoke, dumpTree, text helpBrowser, slice mode, get slice, and get dsp verified current Help/Settings routes, the updated 52 px/PHNE help resource, seven NR choices, MNR's visible disabled reason, RN2's 0–50 range/default 0, and the new token advice without reading a credential. Settled engine snapshots showed:

  {"sliceMode":"USB","NNR":{"available":true,"enabled":true},"active":"NNR"}
  {"sliceMode":"CW","NNR":{"available":true,"enabled":false},"active":"none"}
  {"sliceMode":"USB","NNR":{"available":true,"enabled":true},"active":"NNR"}

  These are extracted fields from get slice/get dsp, not a single bridge response. Final get radio reported serial=DEMO-0001, panCount=1, sliceCount=1, transmitting=false; closed the review-owned instance. The earlier restricted launch stalled before the bridge started and only that owned process was stopped.

• Independently checked the 23-ID ordering against AppletPanel::kDefaultOrder, TMate 2 VID/PID against HidDeviceParser, warmth against kDefaultWarmth, RTTY-only gating against refreshRttyDecodeState, single-click handling against the production off-screen event/wiring, the renderer-failure card and platform advice, platform Ulanzi implementations, and AGENTS prompt URLs. No new design, routing, or persistence behavior is hidden in the string edits.

• All four conformance gate outputs are identical between merge base and head. Colour ratchet: +0 across all three counts. Accessibility: 9 warnings on both sides, unchanged. Bridge-doc generation, test-registration, frozen CI test gate and whitespace checks passed. Automated code-review skill unavailable; verified the previous reviews' leads manually rather than adopting their verdicts.

Hardware convergence, native macOS/Windows GUI behavior, actual BNR download size/denoising quality, and NNR CPU/model-quality estimates were not measured. NNR figures match the existing UI guidance; that is source agreement, not independent performance proof. The demo cannot validate controller hardware or Flex-specific positive effects, and no live radio was contacted.

CI and recommendation

Approve with nits. Required checks are successful on the reviewed SHA, but their pull_request runs were created at 2026-10-07T16:27:40Z. Current main 4b9be492790102ae82070a0bae58795a9e445418 was committed at 2026-10-07T23:01:28Z, so the green checks describe an older merge result, not today's main. The source audit, seven focused test executions and isolated demo verification found no blocker; fix the two copy nits when convenient and leave final approval/merge to the maintainer.

[ten9876]

All review nits from @aethersdr-agent, @jensenpat and the maintainer review are addressed in 88e58b0b7:

1. BNR download size. resources/help/understanding-noise-cancellation.md:255 now says "about 1 GB", matching AetherDspWidget.cpp:1639 Download (~1 GB) and the :1783 tooltip. docs/nvidia-bnr.md:60 is aligned to ~1 GB as well (maintainer call), so every user-facing figure agrees.
2. aethersdr.log on Windows. CONTRIBUTING.md:35-36 now reads "Attach the newest one; on Linux and macOS aethersdr.log is a link to it." This matches src/main.cpp:650, where the Windows branch uses QFile::link, which produces a .lnk shortcut.
3. Repo path in the tooltip. RadioSetupDialog.cpp:2010 now links https://github.com/aethersdr/AetherSDR/blob/main/docs/automation-bridge.md. The three sibling tooltips on the same Network page (:2047, :2142, :2220) had the same source-tree pointer and are folded in (maintainer call). strings build/AetherSDR | grep -c blob/main/docs/automation-bridge.md → 4.
4. AetherPad udev. The omission is deliberate (maintainer call). packaging/linux/60-hid-encoders.rules now says why: 2341:0266 is the Arduino Giga R1's stock identity, so a rule would open every Giga on the machine.

Verification at 88e58b0b7

• AetherSDR was rebuilt (the RadioSetupDialog.cpp.o objects are newer than the source). radio_setup_label_theme_token_test, radio_setup_recording_mode_dim_test, radio_setup_max_power_field_test, radio_setup_region_field_test and automation_menu_lookup_test all pass (100% tests passed out of 5). No test pins the changed tooltip strings.
• All 18 static-check scripts plus gen_bridge_docs.py --check exit 0. check_a11y reports the same 9 pre-existing warnings, and git diff --check is clean.
• All 9 commits report verified=true, reason=valid.

Not done: the suggested string-pinning test for the tooltip and the Ulanzi fallback is not added; it was a non-gating suggestion.

[github-actions]

Accessibility check findings

This PR touches src/gui/ files with 9 accessibility issue(s).
Full details appear as inline annotations on the diff above.

Files scanned : 4
Files with findings: 1
Total findings : 9

See docs/a11y.md for the canonical
Qt patterns (setAccessibleName, QAccessibleValueChangeEvent,
QAccessibleInterface subclasses, etc.). Structural remediation
(keyboard navigation, custom QAccessibleInterface subclasses)
is tracked in #3288.

Findings are warnings only — the build is not blocked.