Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@pio-scripts/load_usermods.py`:
- Around line 57-93: _predict_dep_name currently ignores symlink:// and file://
(and other non-hosted URL schemes) and thus returns None for those entries;
update it so that after the existing hosted-Git branch (the
_URL_SCHEME_RE.match(entry) check) you still parse any scheme with
urlparse(entry) and extract the final path component (Path(parsed.path).name),
stripping trailing '/' and a '.git' suffix and returning None if empty; retain
the special-case branch that handles github.com/gitlab.com/bitbucket.com by
using the second path segment when available, and ensure the SSH branch
(_SSH_URL_RE) and '@' / plain registry branches keep their current behavior
(still strip fragments, versions, and '.git' as currently done).

In `@pio-scripts/validate_modules.py`:
- Around line 42-50: The current subprocess.run call invoking nm (in the block
that assigns result = subprocess.run(...) and nm_output = result.stdout) only
handles exceptions, so if nm exits with a non-zero returncode we may parse empty
stdout and erroneously fail; update the logic to check result.returncode after
the subprocess.run call and treat non-zero as a tool failure: log a warning
including returncode and stderr, then return the conservative pass set (same as
in the exception branch) instead of proceeding to parse nm_output; ensure you
reference result.returncode and result.stderr when building the warning message
and preserve the existing conservative-return behavior used for
subprocess.TimeoutExpired/FileNotFoundError/OSError.

In `@usermods/PS_Comet/PS_Comet.cpp`:
- Line 4: The global nextCometCreationTime must be moved into segment-local
effect state so each segment running mode_pscomet() has its own throttle; stop
using the translation-unit global nextCometCreationTime and instead store the
deadline in SEGENV.step (or in the per-segment extra bytes after PSdataEnd) and
initialize it in the effect initialization path, then reference and update that
per-segment field inside mode_pscomet() wherever nextCometCreationTime is
currently read or written so reinitialization and separate segment instances do
not share the same timer.

In `@usermods/TetrisAI_v2/tetrisbag.h`:
- Around line 89-92: The loop in queuePiece() shifts elements then assigns to
piecesQueue[piecesQueue.size() - 1], which underflows when piecesQueue is empty;
add an early guard at the start of queuePiece() (e.g., if (piecesQueue.empty())
return;) or otherwise ensure queueLength > 0 before running the loop and the
final assignment so you never compute size()-1 on an empty vector; update the
function that contains the for-loop and the call to getNextPiece() to use this
guard.

In `@usermods/user_fx/user_fx.cpp`:
- Around line 1131-1147: The 4-bit hash in SEGENV.aux1 is too small and causes
frequent collisions; compute and keep a larger fingerprint (e.g., full 16-bit
textHash) and persist it in a wider field (use SEGENV.data or another available
SEGENV field) instead of truncating to 4 bits. Change the code that reads
lastHashBits from SEGENV.aux1 to read the full stored hash from SEGENV.data (or
the chosen field), compute a 16-bit textHash in the loop (keep textHash as
uint16_t), set currentHash = textHash (or a simple reduced-but-still-large slice
like top 12 bits), compare that against the stored value when computing
textChanged, and ensure when rebuilding you update the stored hash in
SEGENV.data (and stop packing the hash into SEGENV.aux1). Update references
around textHash, currentHashBits/lastHashBits, needsRebuild and where
SEGENV.aux1 was previously written so the new larger hash is saved and loaded
consistently (affects mode_morsecode / SEGMENT.name change detection).

In `@usermods/usermod_v2_auto_save/usermod_v2_auto_save.cpp`:
- Around line 132-133: When the early-return path triggered by
autoSaveIgnorePresets && currentPreset>0 runs, it must cancel any pending
autosave and resync the tracked state before returning; update the block around
the condition that checks autoSaveAfterSec, enabled, autoSaveIgnorePresets,
currentPreset, strip.isUpdating(), now and lastRun to clear the pending
autoSaveAfter timer/flag and refresh the known* tracking variables (e.g., known
color/state variables the autosaver uses) prior to returning so a
previously-armed autosave cannot later persist the preset state unexpectedly.

In `@wled00/bus_manager.cpp`:
- Around line 1190-1199: The current code inserts BusPlaceholder instances into
BusManager::busses which preserves the original bus type and causes unsafe
downcasts (e.g., static_cast<BusDigital*>(bus.get())->isI2S()) and premature
loop termination on checks like bus->isOk(); fix by either (A) never pushing
BusPlaceholder into BusManager::busses (create placeholders in a separate
container or only keep real Bus* in busses) or (B) add a reliable discriminator
on Bus (e.g., enum BusKind or bool isPlaceholder()) and update every downcast
and loop to check that discriminator before casting or treating the bus as real;
audit and guard all uses such as BusPlaceholder, BusManager::busses,
Bus::isVirtual, Bus::isDigital, the static_cast<BusDigital*> sites, the
virtual-cast to BusNetwork, and FX_fcn.cpp checks of bus->isOk() so placeholders
are skipped or handled safely.

In `@wled00/bus_wrapper.h`:
- Around line 1389-1398: The current logic in the I2S assignment block
(_i2sChannelsAssigned, offset) force-coerces later bus pixel type t to the first
lane's type via _parallelBusItype and flips _useParallelI2S, which silently
creates mismatched pixel-type buses; instead detect when a later request's t
differs from _parallelBusItype and reject/abort creation (or fall back to RMT)
rather than overwriting t: update the branch that handles offset==1 to compare t
vs _parallelBusItype and, on mismatch, return an error/false or switch to the
RMT path, keeping _useParallelI2S only for compatible types and preserving the
original t when compatible.

In `@wled00/cfg.cpp`:
- Around line 110-119: The current code always writes default/empty
WPA-Enterprise values into multiWiFi[n], causing silent downgrades when the JSON
omits those keys; change deserialize logic so encType, enterpriseAnonIdentity
and enterpriseIdentity are only overwritten when the corresponding JSON keys are
present/non-empty—use the existing wifi JSON checks (e.g., test wifi["e_ident"],
wifi["e_anon_ident"], and wifi[F("enc_type")] or CJSON result) and only call
multiWiFi[n].encryptionType = encType and strlcpy(...) when a value was actually
provided by getStringFromJson/CJSON, otherwise leave
multiWiFi[n].enterpriseAnonIdentity, multiWiFi[n].enterpriseIdentity and
multiWiFi[n].encryptionType unchanged.

In `@wled00/data/cpal/cpal.htm`:
- Around line 117-121: When toggling preview via the gId('chkPreview') change
handler, the current logic calls applyLED() or sends
requestJson({seg:{frz:false}}), which only unfreezes currently selected segments
and leaves the previously previewed main segment frozen; modify the handler and
related preview code to cache the previewed segment IDs (e.g., store in a
variable like previewedSegIds when applyLED() is used for previewing) and when
turning preview off (or on page leave) call requestJson for each cached id with
explicit objects like {seg:{id:<cachedId>, frz:false}} (or a single request with
an array of ids if the API supports it), then clear the cache; apply the same
change at the other occurrence noted (around lines 693–695) to ensure previewed
segments are explicitly unfrozen.

In `@wled00/data/edit.htm`:
- Around line 111-136: The timeout branch in loadFiles currently calls cb() as
if the script loaded, which can start dependent work prematurely; change the
timeout handler (tm) inside loadFiles so it invokes the script's error path
instead of cb() — e.g., call the same onerror logic (or invoke s.onerror()) so
the retry/failure logic runs (and eventually calls cb() only when appropriate).
Update references in loadFiles (tm, s.onload, s.onerror, cb) so the timeout
triggers the error/retry flow and thus prevents loadResources/S from being
called prematurely.

In `@wled00/data/index.js`:
- Around line 46-62: The current loadIro IIFE sets l.src = 'iro.js' before
onLoad populates location info, causing file: mode to try a local file and retry
indefinitely; change the flow so the page initializes first (wait for onLoad to
run or check d.readyState) and only then call a refactored loadIro that builds
l.src from the resolved location/IP (use the populated loc/locip or a canonical
absolute URL) and creates cpick; remove the tight infinite retry by adding a
graceful fallback (stop after a few retries or detect file: scheme and skip
loading the remote script), and ensure errors call a single failure handler
instead of continuously retrying so the IP prompt can appear when appropriate.

In `@wled00/data/settings_leds.htm`:
- Around line 147-150: The current save path contains a hard-stop when bquot >
100 (the block that alerts "Too many LEDs! Can't handle that!" and returns
false), which prevents saving oversized but recoverable configs; remove that
early return and instead only show warnings (preserve the existing alerts for
bquot > 80 and the additional messages referencing maxM) so the form can proceed
and allow finalizeInit() logic to trim bus lengths; ensure no other code paths
rely on that return to avoid proceeding with completely invalid data.

In `@wled00/dmx_input.cpp`:
- Around line 151-158: The early return on invalid inputPortNum leaks previously
allocated GPIO pins; either validate inputPortNum before any pin allocation
(move the inputPortNum check above the pin-allocation code) or, if you keep the
current order, free/deallocate the allocated pins before the early return (call
the existing pin cleanup routine or add a deallocatePins()/releasePins() call)
so that when the block around this->inputPortNum sets default and returns the
pins are not left reserved.

In `@wled00/dmx_output.cpp`:
- Around line 82-100: The DMX API calls use the v4.1.0 signatures: replace the
current three-arg dmx_driver_install call in DMXOutput::init with the four-arg
form expected by esp_dmx (pass dmxPort, config (not &config), a personalities
pointer (use nullptr or an empty array) and a personality count (0) to satisfy
the new signature), and change the dmx_send call in DMXOutput::update to the
single-argument form dmx_send(dmxPort); keep using dmx_write(dmxPort, dmxdata,
DMX_PACKET_SIZE) and dmx_set_pin(dmxPort, outputPin, -1, -1) as before but
ensure the types match the updated API.

In `@wled00/FX_fcn.cpp`:
- Around line 1165-1188: The code counts i2sBusCount from raw
BusConfig.driverType before fallbacks and then later double-counts bus memory
(memUsage + getBusSize), causing use_placeholder to trigger incorrectly; fix by
(1) moving the I2S detection to use the resolved runtime driver (call
BusManager::getI(...) or resolve the bus via BusManager before counting) and
derive useParallelI2S from that resolved driver instead of BusConfig.driverType,
and (2) remove the duplicate size addition (do not add getBusSize() on top of
BusConfig::memUsage() or adjust memUsage so it only counts each bus once) so
that mem and use_placeholder are computed from the same, non-duplicated values
(refer to busConfigs, i2sBusCount, Bus::isDigital, Bus::is2Pin, driverType,
BusManager::getI, useParallelI2S, BusConfig::memUsage, getBusSize,
use_placeholder).

In `@wled00/wled_metadata.cpp`:
- Around line 131-137: The bufToString template leaves sbuf uninitialized past
real_len causing nondeterministic reads; in the bufToString(const char
(&buf)[len]) function, after memcpy(..., real_len) set sbuf[real_len] = '\0'
(instead of sbuf[len]) so the copied string is properly null-terminated and
String(sbuf) only reads the intended bytes.
- Around line 56-66: The metadata struct WLED_BUILD_DESCRIPTION is emitting a
hard-coded version of 1 which prevents shouldAllowOTA() from checking
safe_update_version; change the second field (currently
"/*WLED_CUSTOM_DESC_VERSION*/ 1") to the actual v2 metadata version (e.g.,
replace with WLED_CUSTOM_DESC_VERSION or the literal 2) so binaries produced
include desc_version > 1 and trigger the safe-update gate (update the
initializer in WLED_BUILD_DESCRIPTION).

---

Outside diff comments:
In `@wled00/json.cpp`:
- Around line 813-817: Preserve info.arch as a stable family identifier and add
a separate key for the detailed chip model: always set root[F("arch")] to the
family string (e.g., "esp32" for ESP32 targets) instead of conditionally using
ESP.getChipModel(); then publish the full chip model under a new key such as
root[F("chip")] using ESP.getChipModel() (or equivalent) so clients can still do
exact comparisons on arch while accessing detailed model via the new key.

---

Minor comments:
In `@usermods/pixels_dice_tray/BLE_REQUIREMENT.md`:
- Around line 17-20: The cp command in BLE_REQUIREMENT.md incorrectly uses
../../../ which overshoots the repo root; update the sample copy command so it
references the actual sample file path and copies it into the intended
destination (replace the existing cp line that mentions ../../../ with a command
that reads the sample from
usermods/pixels_dice_tray/platformio_override.ini.sample and copies it to
platformio_override.ini) so readers running the example from the repo root or
from within usermods/pixels_dice_tray get the correct file.

In `@usermods/user_fx/README.md`:
- Around line 269-273: The README example incorrectly references SEGMENT.call;
update the snippet and text to use SEGENV.call (the call counter lives on the
segment environment) so replace any mention of SEGMENT.call with SEGENV.call and
ensure the explanatory sentence refers to SEGENV.call incrementing on each
function call.

In `@wled00/data/common.js`:
- Around line 140-160: uploadFile is marked async but currently returns
immediately; convert it into a proper promise-returning async function by
returning a new Promise that resolves/rejects when the XMLHttpRequest completes:
create and return new Promise((resolve,reject)=>{ set
req.addEventListener('load', ...) to call resolve(this.status < 400) (and still
call callback if provided), set req.addEventListener('error', ...) to call
reject(false) or resolve(false) consistently (and call callback), and remove the
implicit immediate completion—ensure fileObj.value = '' runs after
resolution/rejection; keep existing await usage (e.g., await file.text()) so
uploadFile remains async but now correctly represents upload completion.

In `@wled00/data/index.css`:
- Line 3: The src declaration uses an unquoted data URL in the url(...) call
which triggers Stylelint's function-url-quotes rule; fix it by updating the src:
url(...) token (the data:font/woff2... entry) to use quoted URL syntax (i.e.,
url("...")) so the entire data URL argument is wrapped in quotes while leaving
the rest of the `@font-face` declaration unchanged.

In `@wled00/data/settings_sec.htm`:
- Line 69: The closing </span> in the OTA section is orphaned (see the HTML
fragment "Disabling this option will make your device less
secure.</i><br></span>"); remove the stray </span> so the markup properly closes
only the <i> tag and the <br>, or if intended, add the corresponding opening
<span> with the needed attributes—prefer removing the stray </span> to restore
valid HTML in settings_sec.htm.

In `@wled00/ota_update.cpp`:
- Around line 403-442: calculateBootloaderSHA256 currently ignores
esp_flash_read failures and BootloaderImageSizer parse errors and always sets
bootloaderSHA256CacheValid = true; change it to track any failure (e.g., set a
bool readFailed if esp_flash_read(...) != ESP_OK, and a bool parseFailed if the
sizer detects a corrupt/impossible size such that totalSize == 0 or totalSize >
BOOTLOADER_SIZE or other sizer-specific error), and if either flag is true or
totalHashLen == 0 then do not mark the cache valid: clear/zero
bootloaderSHA256Cache, ensure mbedtls_sha256_free(&ctx) is called, and leave
bootloaderSHA256CacheValid = false (return early). Update
calculateBootloaderSHA256 to only set bootloaderSHA256CacheValid = true when no
read/parsing failures occurred and totalHashLen > 0.

---

Duplicate comments:
In `@wled00/dmx_output.cpp`:
- Around line 95-96: The write method currently narrows the DMX address and can
wrap addresses >255 and lacks bounds checking; change DMXOutput::write signature
to accept a wider integer type (e.g., int or size_t) to match how DMXAddr is
computed in handleDMXOutput(), update the declaration in wled00/dmx_output.h
accordingly, and inside DMXOutput::write validate the incoming address against
the dmxdata buffer size (use the buffer length or a constant like
DMX_UNIVERSE_SIZE) before indexing to avoid out-of-bounds writes.
- Around line 72-75: The allocation-failure log is currently passing an enum to
a "%s" format which is undefined behavior; update the second DEBUG_PRINTF call
to pass the string-returning helper instead by replacing
PinManager::getPinOwner(outputPin) with PinManager::getPinOwnerName(outputPin)
(keep the "%s" format), or alternatively use the new getPinOwnerName() in any
combined message so the log receives a C-string; modify the DEBUG_PRINTF
invocation in the DMXOutput allocation failure branch accordingly.

In `@wled00/set.cpp`:
- Around line 476-478: The code only updates dmxOutputPin but does not
reinitialize the DMX driver, so change the handler that sets dmxOutputPin (where
request->arg(F("IDMO")).toInt() is used) to also reconfigure the DMX output:
stop/disable the current DMX instance, release the old pin, and initialize/start
the DMX driver on the new pin. Locate the DMX init/teardown logic (e.g.
functions used elsewhere like startDmxOutput()/stopDmxOutput(),
dmx.begin()/dmx.stop(), or the DMX driver init in setup) and invoke the teardown
then startup sequence after assigning dmxOutputPin; if such helper functions
don’t exist, add a small reinit routine that calls the driver’s stop and
begin/init with the new dmxOutputPin so the change takes effect immediately
without reboot.

---

Nitpick comments:
In `@CONTRIBUTING.md`:
- Around line 61-66: The blockquote starting with "[!CAUTION]" contains an extra
blank line before "Do not use "force-push" while your PR is open!" which
triggers markdownlint MD028; remove the empty line so the blockquote content is
contiguous (i.e., ensure "[!CAUTION]" is immediately followed by the warning
text) to fix the formatting and prevent inconsistent Markdown rendering.
- Around line 17-21: The blockquote lines starting with ">   " (the line
containing "**The easiest way to start your first PR**" and the following lines
including the image tag) have extra spaces after the ">" which triggers
markdownlint MD027; update those blockquote lines to use a single space after
">" (e.g., change ">   **The easiest..." to "> **The easiest..." and similarly
remove the extra spaces before the image tag and subsequent lines) so the
blockquote follows standard markdown spacing.

In `@usermods/pov_display/pov_display.cpp`:
- Line 38: Remove the redundant trailing `return;` from the end of the void
function in pov_display.cpp (the final statement shown as `return;` in the
diff); simply delete that line so the function returns implicitly rather than
explicitly.

In `@usermods/TetrisAI_v2/TetrisAI_v2.cpp`:
- Line 8: The global variable noFlashOnClear should be made file-local to avoid
exporting an unnecessary symbol: change its declaration in TetrisAI_v2.cpp to
have internal linkage (e.g., prepend static: "static bool noFlashOnClear =
false;") or move it into the TetrisAIUsermod class as a private member and
update all uses accordingly; update any references in the file to use the new
member name if moved into the class.

In `@usermods/usermod_v2_animartrix/usermod_v2_animartrix.cpp`:
- Around line 88-296: This block of mode_* wrapper functions (e.g.,
mode_Module_Experiment10, mode_Zoom, mode_RGB_Blobs, mode_Rotating_Blob) are
only used in this translation unit and should be file-local to avoid exporting
globals; make them static (or enclose the entire block in an anonymous
namespace) and reformat the block to the project's 2-space C++ style while
preserving the existing calls like anim.initEffect() and anim.<EffectName>().

In `@usermods/usermod_v2_auto_save/usermod_v2_auto_save.cpp`:
- Line 132: Replace the magic number 240 in the autosave throttle check with a
named constant (e.g., AUTO_SAVE_UPDATE_THROTTLE_SEC) to make the meaning
explicit; declare the constant at the top of usermod_v2_auto_save.cpp (or
alongside other constants) and update the condition that uses strip.isUpdating()
&& now - lastRun < 240 to use the new constant (refer to autoSaveAfterSec,
enabled, autoSaveIgnorePresets, currentPreset, strip.isUpdating(), lastRun, now)
so the throttle interval is self-documenting and easy to change.

In `@wled00/button.cpp`:
- Around line 379-382: The relay enable assumes pinMode was already set
elsewhere; make the relay setup self-contained by calling pinMode(rlyPin,
OUTPUT) before digitalWrite(rlyPin, rlyMde) in the block that checks rlyPin>=0
(so the relay is reliably driven even if handleOnOff(true) or beginStrip() order
changes), then keep the existing delay(RELAY_DELAY) afterwards; reference
rlyPin, rlyMde, RELAY_DELAY, digitalWrite(), and the related
beginStrip()/handleOnOff setup when updating the code.

In `@wled00/data/icons-ui/style.css`:
- Around line 4-6: The `@font-face` src list in style.css (the lines listing
url('fonts/wled122.ttf...'), url('fonts/wled122.woff...'),
url('fonts/wled122.svg...')) no longer includes a woff2 entry; confirm this
removal was intentional and, if not, restore the woff2 source (e.g., add a
url('fonts/wled122.woff2?...') format('woff2') entry) within the same `@font-face`
src order and ensure the referenced woff2 file exists in the fonts assets and
any cache-busting query string matches your build artifacts.

In `@wled00/data/settings_pininfo.htm`:
- Around line 16-30: The 250ms polling in S() (pinsTimer = setInterval(loadPins,
250)) is too aggressive for resource-constrained devices; change the interval
used by setInterval to a longer value (e.g., 500–1000 ms) or make it
configurable, and keep the call site referencing pinsTimer and loadPins so the
existing start/stop logic remains intact.