Content and structure are solid — the banners, AGENTS.md additions, CONTRIBUTING.md note, CODEOWNERS gating, and doc are all exactly the right response to the #3351 documentation gap.

One required fix before merge: the TU map in docs/architecture/mainwindow-decomposition.md is missing two of the nine sibling TUs. Both exist in CMakeLists.txt and have opening charters that make the entries easy to write:

A contributor looking at the table before adding DSP applet wiring or session-scoped code wouldn't find the right TU — which is exactly the gap this PR is closing.

Side note on #3407: the new "Wiring split mirrors the widget's lifecycle" convention in the doc clarifies our earlier comment. The RxApplet::wfmActivated connection in the constructor is correct per that rule (singleton widget, wired once at construction → stays in MainWindow.cpp). Only the activateWFM() / deactivateWFM() / reflectWfmButtons() bodies still need to move to _DigitalModes.cpp per the decision table.

Thanks for this, @ten9876 — surfacing the #3351 decomposition at the project level (point-of-use banners + canonical doc + AGENTS.md/CONTRIBUTING + the CODEOWNERS gap-close) is genuinely useful, and the layering across tools is well thought out. The banners are comments so there's no build risk, the AGENTS.md #adding-code-to-mainwindow anchor and the ../../CONTRIBUTING.md / ../../.github/CODEOWNERS relative links resolve correctly, the CODEOWNERS scoping-to-core-only is correct (there were no prior src/gui/ rules), and the "~440 lines" AGENTS.md figure now matches reality (it was stale at "~330").

One substantive issue, since this doc's whole job is to be the map agents consult before routing code:

The TU map is missing two sibling TUs, and the "8 today" count is off. The repo actually has 9 MainWindow_*.cpp siblings; the map in docs/architecture/mainwindow-decomposition.md lists 7. Missing:

• MainWindow_Session.cpp (Phase 2c) — wireDiscovery() / wireRadioModel() / wirePanLifecycle(); the seed of the future RadioSession aggregate (#3445).
• MainWindow_DspApplets.cpp (Phase 2d) — wireDspApplets() / wirePooDooTiles(); client-DSP applet + PooDoo tile wiring.

Both are real MainWindow:: members carved out under #3351, so they belong in the map. Concretely:

1. Add the two rows to the TU map table (the Phase column already fits — 2c/2d).
2. Update "there are 8 today" (in the When a new TU is warranted section) to 9.
3. Minor consequence worth a glance: the decision-guide row "Connecting a newly-created radio object (slice/pan/VFO/DSP widget) → MainWindow_Wiring.cpp" slightly overlaps these — pan-lifecycle wiring actually lives in MainWindow_Session.cpp and DSP-applet wiring in MainWindow_DspApplets.cpp. Not blocking, but a one-line note ("session/pan-lifecycle wiring → _Session; DSP-applet wiring → _DspApplets") would keep an agent from landing it in the wrong sibling.

An incomplete map is the one failure mode that undercuts the goal here — an agent looking for where session/discovery or DSP-applet wiring goes wouldn't find a home and would likely default back to MainWindow.cpp, exactly what this is meant to prevent. Worth a quick pass to enumerate the siblings straight from ls src/gui/MainWindow_*.cpp so the map can't drift. Everything else looks good.

_{🤖 aethersdr-agent · cost: $6.0137 · model: claude-opus-4-8}

Thanks for this, @ten9876 — this is a well-targeted bit of documentation hygiene. The "documented only inside each sibling's header comment, which you see after opening the wrong file" framing is exactly the gap, and surfacing it at the point of use is the right instinct.

I verified the structural claims against the tree and they hold up:

• TU map is accurate. All nine MainWindow_*.cpp siblings plus MainWindowHelpers.{h,cpp} and MainWindowShortcutState.h exist exactly as listed; the "9 siblings / 10 files include the header" counts are correct.
• The AGENTS.md line-count bump is real, not a typo. AGENTS.md on main is already 407 lines, so "~330" was stale; 407 + ~33 added ≈ "~440". Good catch correcting it.
• CODEOWNERS does what the PR claims. The two exact-path lines are appended last, so under last-match-wins they gate only MainWindow.{h,cpp} to @aethersdr/maintainers while the sibling TUs fall through to the * Tier-3 reviewer default — matching the stated "deliberately broad" intent. The inline comment warning against a future MainWindow* glob is a nice touch.

Two minor, non-blocking points:

1. The MainWindow.cpp banner's top rule line looks UTF-8-corrupted. In the diff, that first line's box-drawing run breaks with replacement characters (───���───) partway across, whereas the MainWindow.h banner is clean. It's harmless to the build (it's a comment), but a little ironic in a banner whose whole job is to look clean to everyone who opens the file — worth re-saving that line as UTF-8 to match MainWindow.h. Please double-check on your end, since this could in theory be a transport artifact rather than the committed bytes.

2. CODEOWNERS header block vs. the new lines. The carefully-written tier narrative at the top (lines 1–52) maps src/ → Tier 3 reviewers and frames Tier 1 as "governance, security, bot policy." The two new maintainer-gated src/gui/MainWindow.{h,cpp} lines are source code at the maintainer tier — a genuine exception that the top-of-file model doesn't mention. The inline comment above the lines explains why well; consider a one-liner in the header block too (or noting it's an intentional source-code exception), so a reader following the tier model isn't surprised to find src/ files at Tier 1.

Neither blocks merge. Nice work closing the ownership gap the split opened.

_{🤖 aethersdr-agent · cost: $2.4734 · model: claude-opus-4-8}

Both items from the CHANGES_REQUESTED are resolved. MainWindow_Session.cpp and MainWindow_DspApplets.cpp are now in the TU map with accurate charters drawn from their file headers, the sibling count is updated in the prose (9 TUs / 10 files), and the decision guide has two new disambiguation rows (_Session.cpp vs _Wiring.cpp for discovery/connection/pan-lifecycle; _DspApplets.cpp vs _Wiring.cpp for client-DSP applets). CONTRIBUTING.md and the AGENTS.md decision table are consistent with the updated map. Approved.