Move the fork's engineering notes out of the root and into docs/fork

Ten working notes had accumulated at the top level, where they outnumbered the
files a visitor actually wants and sat among upstream's own root documents with
nothing to say which was which. Git records them as renames, so the history of
each one follows it.

Their cross-references are relative and the whole set moved together, so those
still resolve; the two mentions from outside - .gitignore and two workflow
comments - are updated. docs/fork/README.md indexes them by subject, because a
folder of ten shouting filenames is not much better than a root of ten.

FORK.md stays at the root deliberately. It is the fork's front door, README
links it, and the release notes of every published release carry an absolute URL
to it - moving it would break those permanently.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

Author: nickolas122

.github/workflows/main.yml

@@ -245,7 +245,7 @@ jobs:
             # fails under mingw and Win32 is selected by accident. Asking for it
             # explicitly means a future toolchain that happens to pass the test
             # cannot silently switch backends and discard the patches applied above.
-            # See WINDOWS-WINRT-PHASE0.md.
+            # See docs/fork/WINDOWS-WINRT-PHASE0.md.
             qmake -r -- -feature-native-win32-bluetooth
             make -j8
             cd ../..
@@ -503,7 +503,7 @@ jobs:
   # Qt ships a working WinRT Bluetooth backend only for MSVC: QT_FEATURE_winrt_bt
   # is 1 in the msvc2022 package and -1 in every mingw one (6.8.2, 6.9.3, 6.10.3
   # and llvm-mingw alike), where you get qlowenergycontroller_dummy.cpp instead.
-  # WINDOWS-QT6-PHASE1.md has the measurements and why building it under mingw is
+  # docs/fork/WINDOWS-QT6-PHASE1.md has the measurements and why building it under mingw is
   # a fork of Qt rather than a build of it.
   #
   # window-build stays on mingw + Qt 5.15.2 until this one has been proven against

.gitignore

@@ -6,7 +6,7 @@ src/Makefile
 src/.qmake.stash
 
 # Building from the repository root - which is what the linux-x86-build job does,
-# and what BUILDING-ON-WINDOWS.md describes for a local build - runs qmake on the
+# and what docs/fork/BUILDING-ON-WINDOWS.md describes for a local build - runs qmake on the
 # subdirs project and generates these as well. Anchored to the root so the tracked
 # src/ios/adb/libzipfile/Makefile.{am,in} are untouched.
 /Makefile

FORK.md

@@ -115,6 +115,11 @@ shifting.
   longer clobbers the static library, qthttpserver builds without a native perl,
   `aqtinstall` is pinned, and SmtpClient is pinned to a commit that exists.
 
+## Why, in detail
+
+The working notes behind these changes — measurements, dead ends, and the things that
+turned out not to be the cause — are in [docs/fork/](docs/fork/).
+
 ## Versioning
 
 Releases are tagged `v<upstream base>-qz.<n>` — for example `v2.21.6-qz.1`, meaning the

AUTO-ERG-MODE.md docs/fork/AUTO-ERG-MODE.mdAUTO-ERG-MODE.md renamed to docs/fork/AUTO-ERG-MODE.md

@@ -0,0 +1,33 @@
+# Fork engineering notes
+
+Working notes written while making the changes listed in [FORK.md](../../FORK.md). They
+are a record of *why* — what was measured, what was ruled out, what turned out to be
+wrong — not user documentation. Upstream's docs are in the folder above.
+
+They are kept because the expensive part of this work was the evidence, and a note that
+says "this was tested and it is not the cause" is worth more later than the fix itself.
+
+## Windows Bluetooth: the move to WinRT
+
+| | |
+| --- | --- |
+| [WINDOWS-WINRT-PHASE0.md](WINDOWS-WINRT-PHASE0.md) | The case for moving off the Qt 5 Bluetooth path, and where the original argument for it did not survive contact with the source. |
+| [WINDOWS-WINRT-BACKEND.md](WINDOWS-WINRT-BACKEND.md) | The plan for the backend swap. |
+| [WINDOWS-BLE-HARDENING.md](WINDOWS-BLE-HARDENING.md) | Reconnect behaviour, service subscription, and the bond-cache problem that stopped the trainer working unpaired. |
+
+## Windows on Qt 6
+
+| | |
+| --- | --- |
+| [WINDOWS-QT6-PLAN.md](WINDOWS-QT6-PLAN.md) | The route to Qt 6 and the questions that decided its shape. |
+| [WINDOWS-QT6-PHASE1.md](WINDOWS-QT6-PHASE1.md) | Toolchain measurements — why mingw loses and MSVC 2022 wins. |
+| [WINDOWS-QT6-PHASE4.md](WINDOWS-QT6-PHASE4.md) | The result, including a prediction from Phase 0 that the bike disproved. |
+| [BUILDING-ON-WINDOWS.md](BUILDING-ON-WINDOWS.md) | How a Windows build of this fork is actually obtained, and the traps worth naming. |
+
+## Devices and training apps
+
+| | |
+| --- | --- |
+| [DIRCON-SERVER-REFACTOR.md](DIRCON-SERVER-REFACTOR.md) | Why the DIRCON endpoint had to outlive the bike, and the refactor that gave it process lifetime. |
+| [AUTO-ERG-MODE.md](AUTO-ERG-MODE.md) | The automatic ERG detector, parked, with an honest account of why. |
+| [PLANO-MEGAGYM-EXECUCAO.md](PLANO-MEGAGYM-EXECUCAO.md) | Notes in Portuguese, from making the bike report the effort the rider actually made. |