Reconcile the pending doc edits with the creator's actual ruling, and close the
revenue-share question that was blocking the monthly price.
Verified from the App Developer Agreement v8.10 PDF itself (downloaded and
text-extracted, not a search excerpt). Section 6(b) has only three tiers:
6(b)(i) 15% for Apps and their In-App Products *not listed in* 6(b)(iii);
6(b)(ii) 12% Games-only; 6(b)(iii) 30% for Xbox console apps/games, Xbox
non-subscription IAP, and Windows 8/Phone 8. LlamaCasty is a Windows PC App,
so 6(b)(i) governs BOTH tiers -- there is no subscription surcharge. The
agreement's own changelog (v8.0, Oct 26 2017) states it outright: "implement
the 85/15 revenue share for non-Game subscriptions."
=> $10/mo nets $8.50 (~$102/yr); $400 lifetime nets $340.
Also confirmed: the 15% applies after VAT/GST (Net Receipts definition),
payouts are monthly above a $50 threshold, and no better small-indie rate
exists in the standard terms.
Creator ruling recorded: $10/mo subscription or $400 lifetime, no annual, no
.99. This supersedes the 2026-09-21 one-time $29 -> $49 model.
Two obligations ADA 6(h) attaches to the recurring tier, recorded because they
change its risk profile and are the reason lifetime is the hedge: we must
fulfil the subscription for the entire period as marketed (on breach Microsoft
may refund the full amount plus taxes in its sole discretion), and raising the
price disables auto-renew -- so $10 is effectively locked for the product's life.
Stale facts fixed in the same change rather than appended:
- research-store-certification.md: IARC was 11.11.1/11.11.2 in one table and
10.11.1 in another; corrected to 10.11.x.
- research-store-certification.md: policy 10.8.1/10.8.2 still asserted "Polar
explicitly permitted" and "tick the third-party purchase box". Void now that
Store IAP is the route and Polar is being torn down.
- task-48: the working YouTube demo account (10.3.1) was accidentally dropped
from the certification list during the Store-IAP edit. Restored -- a reviewer
cannot use the app without one.
- MyMistakes.md: the verified-fact-vs-decided-outcome lesson now closes its
loop (the creator did rule the way the research pointed), and records the
over-correction that followed.
Docs-only. No code, no build, no tests.
The creator raised it as instinct: "who hasn't been screwed over cancelling a
sub early to be told that the extra six months remaining are your loss?" Checking
the mechanism confirms it, and worse than expected.
Microsoft's general position: "Digital goods like apps, add-on content,
subscriptions... aren't refundable unless the offer or applicable law states that
you're eligible for a refund." Pro-rated refunds on cancel exist only in Canada,
Denmark, France, Israel, Korea, Turkey (all lengths) and Finland, Germany,
Netherlands, Poland, Portugal (renewals only). Critically: "monthly subscriptions
and initial (pre-renewal) purchases aren't eligible for a prorated refund."
A first-year annual subscriber who cancels in month 2 loses months 3-12, in most
countries, and the developer cannot refund it. That is verbatim the trap the
creator named, so we must not build it. => monthly only. Max loss $10, max
grievance a rounding error, one fewer product to declare, less support surface.
The chargeback argument gets stronger, not weaker: a customer down $89 on an
annual cliff disputes through their bank, which is frozen funds and account risk
against a solo dev. Monthly bounds that at $10.
Also records two rulings:
- No .99 prices. "Let's stop with the x.99 stuff - I find that irritating. Just say:
ten bucks a month." The convention only makes a price look cheaper, which is
incoherent for a brand sold on honesty and no-dark-patterns.
- No feature gating, now argued as revenue rather than taste: the free tier's reach
is the funnel and the branding flash is an ad running inside other people's
content.
Pricing model itself stays OPEN in TASKS.md — $10/mo is the lean, and a lifetime
product (~$300, the hedge against the no-moat renewal problem) is undecided. The
Store revenue share is still unverified: confirm the net, not the list.
MyMistakes.md records the actual error: I had offered "just refund anyone who
asks" as a mitigation without checking who holds the authority to execute it.
Store refunds run through Microsoft, not the developer. The check that followed
is what produced this constraint.
MONETIZATION.md got the full analysis but is gitignored, so it stayed local.
The distribution answer existed only in conversation, so every session re-derived
it. It is now in the map, and the route decision is explicitly parked as the
creator's.
new TASKS/research-store-certification.md — Store Policies 7.20 + MSIX packaging:
which policies bind, which don't (and why), cert economics, camera/mic gating
layers, YouTube age + COPPA, the 11.12 UGC judgment call.
Two real defects surfaced, neither fixed (docs-only unit):
- FfmpegLocator downloads an unsigned exe from GitHub and runs it. That is
policy 10.2.2 (dynamic code inclusion) verbatim, and it is the root cause of
the 2026-09-01 404 — the pin aged out of BtbN's 14-day retention on the
creator's first real recording attempt. -> TASK 48 item 1, not
Store-conditional.
- Distribution.md:318 recommended a $400+/yr EV cert for a SmartScreen bypass
Microsoft removed in March 2024. Fixed; had it shipped it would have cost
$400+/yr to buy what $150 buys.
Also new: TASKS/task-48 (checklist, carved out of TASK 36 item 6) and
TASKS/task-49 (chat profanity filter, not blocked). ai.md gains the durable
invariants — full trust or recording breaks silently, chat is rendered never
stored — plus a correction to the FFmpeg locator section. MyMistakes.md records
the lesson: a policy citation is a claim about scope, not just text.
MARCOM.md got the privacy-copy guard but is gitignored by design, so that edit
stays local and did not travel here.
The creator reported that clicking Cancel in the post-recording save prompt still
saved the file under the default name: "cancel should cancel the save option,
discarding the recording" — Enter is what accepts the default.
The modal was always correct (Enter -> DialogResult=true, Cancel/Escape/X -> false).
The bug was in the caller: FinalizeRecordingAsync only overwrote `stem` when the
dialog returned true and then ran File.Move UNCONDITIONALLY, so a falsy result fell
straight through into the save. The nullable stem made "I don't want this" look
like "I accept the default".
Extracted the decision into internal MainViewModel.CompleteRecordingSave so it is
testable against real files without a frame pump:
creatorSaved == false (Cancel/Escape/X) -> delete the temp file, leave the
videos folder empty; a failed delete returns DiscardFailed and the toast
names the file + folder (a "discarded" recording still on disk is worse
than no report).
creatorSaved == true (Save, or Enter on the pre-filled default) -> File.Move.
A blank box still means "keep the auto name", but only on an explicit Save.
Test: ytLive.Tests/RecordingSaveDialogTests.cs — 5 facts over real temp files. The
headline asserts Cancel leaves the directory EMPTY, not merely "the stem is
unchanged"; a modal's own test cannot cover the decision it feeds.
Full suite 362/362 (the RealMouseDrag flake passed this run).
The credit existed only as a WPF BrandFlashLayer TextBlock in the preview at
25% opacity on a 300s timer. A viewer of the stream or the recording never saw
it, so "free tier shows branding" was not actually being delivered. The frame
pump has carried an unused per-frame flashFrame slot for exactly this.
Reverses the documented "Pre-GA posture" (ai.md Monetization), which kept the
flash preview-only so test VODs stayed clean. Creator ruling 2026-09-26: the
free tier has to be honest advertising, so it now reaches the broadcast.
One rendered VideoFrame feeds BOTH the frame pump and the preview, so the
creator's view cannot drift from what viewers get. Spec: 500ms in / 1000ms hold
/ 500ms out, first credit ~5s after go-live then every rand(30s)+30s, single
unwrapped line at a random spot inside the frame every time, neon white core
with a feathered red/blue halo.
Neon recipe is derivative work, per AGENTS.md: bright core + feathered
multi-radius halo with R and B split in opposite directions, from
https://nudaui.dev/components/neon-glow (layered text-shadow falloff),
https://help.maxon.net/rg/en-us/Content/html/Blurs-and-Glows-chromatic-glow.html
("with no displacement, the green channel is all but invisible behind your
white text" once R and B are split) and the OBS obs-stroke-glow-shadow
"feathered stroke with user-defined size and intensity". Neon blue is #4d8bff
rather than the theme's #0f3460, which renders near-black as a halo.
Also fixes a PRE-EXISTING bug found on the way: FramePump's fully-static
shortcut stretched the cached bake and returned it, silently discarding every
per-frame overlay - the social bar and the alert ticker were already lost
there. SceneCompositor.Overlay is now internal (it copies before blitting, so
the bake is never mutated) and the shortcut composites overlays first.
The credit is deliberately NOT folded into SceneGraph.GetBakedBase:
SceneRegion.Matches compares element ids only, so a credit in the cached bake
would freeze there and never expire. Per-frame only, and Epoch is stamped
every read because the 8MB master buffer is recycled - without that the
paste cache would freeze a stale credit.
BrandFlashEnabled is now derived !IsPremium and non-assignable, and the
presenter re-checks the licence on every read, so a key entered mid-credit
cuts the advertisement on the next tick instead of letting it finish.
Tests: 10 new facts. 357 total, 356 pass; the one failure is the known
environment-flaky RealMouseDrag layer test (needs an uncovered desktop
session). Adds RealAppHost.RunAsync - a frame-pump test must await inside the
collection's shared STA thread or the whole suite deadlocks silently.
NOTE: scope-check.sh flags Helpers/InstanceProfile.cs, AppLog.cs,
TokenStore.cs, WebView2Manager.cs and InstanceIsolationTests.cs as outside
this commit. That is a false positive: it uses `git diff HEAD`, which cannot
distinguish a planned second commit in the same session from an unrelated
edit. Those files are the dev multi-instance unit, committed immediately
after this one - as is the InstanceProfile paragraph in ai.md, which shares a
file with the Monetization section corrected here.
The creator confirmed the clip fix ("the video plays now"), then asked where the
scrolling text was — it was invisible, for a structural reason. AlertTickerFrame
existed only as a frame-pump callback blitted into the OUTPUT; PreviewPane.xaml had
no element for it, because the strip is master-width and global, not a Source, so it
cannot ride a per-element Image. Nothing was wrong in the renderer: there was no
consumer in the preview. Same class of defect as the missing IsAlertBox trigger, one
layer up (MyMistakes RULE 5/6).
- tickerPreviewSink on AlertOverlayLayer, published from RefreshAlertPreviews() so it
is always the UI thread; MainViewModel.AlertTicker writes it into one reused
WriteableBitmap bound to a new global AlertTickerElement, mirroring SocialBarElement.
- Source.AlertDisplayMethod + panel "Display" selector: TickerScroll / Flash / Solid.
Flash pulses 0.5s on / 0.5s off for the whole alert; Solid is centred and still.
- The marquee was also unreadable: a fixed 140px/s took ~17s per pass, so a 10s alert
showed the text once, entering from the right and never crossing. Paced in reads per
alert instead (TickerReadsPerAlert = 3 inside the alert's own length, speed derived
from it) — never px/s. Research (websearch: how do OBS/Streamlabs/StreamElements
alert boxes present announcement timing?) settled the unit: Streamlabs exposes "Alert
Duration: choose how long your alert stays on your stream" and "Text Delay", never a
scroll-speed slider (https://support.streamlabs.com/hc/en-us/articles/52499995174299-Setting-up-Your-Streamlabs-Alerts).
Run is phase-started half a frame in so the first frame isn't blank.
- Persistence: AlertDisplayMethod INTEGER NOT NULL DEFAULT 0 via the idempotent
table_info migration, appended LAST in the SELECT because the Source reader is
positional (GetInt32(32..34)) — a mid-list insert would silently shift a neighbour.
Tests: 15 new facts (suite 339/339). RealApp STA host: the pane draws the strip and
collapses at alert end; the layer publishes a real 1920x48 frame for all three methods
and nothing when the ticker is off; three passes counted in 10s by the pill's leading
edge resetting (a seamless marquee never blanks, so an empty frame cannot count a pass);
Solid byte-identical at every moment; Flash on for half of each second; the panel shows
and writes back the choice; the DB round-trips all three alert fields together.
Incidental finding: a bound ItemsSource ComboBox in LeftPanel.xaml broke
LayerReorderPersistenceTests.RealMouseDrag (that test injects PHYSICAL mouse input, so a
load-time re-measure moves the rows out from under the cursor). Rewritten as inline
ComboBoxItems, the shape the chat Font selector already uses in that panel. Recorded as
MyMistakes RULE (8).
Four builds (89/90 + two) burned proving the alert video was PERFECT: real h264
1280x720, 240 frames decoded, alert audio in the live mix, and the shipped asset
byte-identical (md5 0ee1f496…) to the creator's llamacasty-dancingLlama-thankyou.mp4
with every sampled frame full bright content. Build 90's new diagnostics then showed
the frame reaching BOTH consumers every second for the whole clip —
`Alert preview: frame=680x200 a255` and `Output resolver: frame 680x200
playing=True` — an opaque, correctly-sized frame, handed over and never seen.
Root cause: Controls/PreviewPane.xaml keeps the per-element Image Collapsed unless a
DataTrigger fires, and there was no IsAlertBox trigger (only IsImageSource / IsChatBox /
IsWebSource / IsWebcam). The alert box's Image was therefore Collapsed forever. Chat
boxes rendered because they HAVE a trigger — that asymmetry is the whole clue, and it
is why removing/re-adding the layer could never fix it.
Fix: an IsAlertBox DataTrigger beside the IsChatBox one.
Two Good Dog tests, because the existing fakes had been hiding this:
- AlertBoxPreviewVisibilityTests (red `Expected: Visible / Actual: Collapsed`, now
green) drives the real MainWindow + PreviewPane and asserts the bound Image is
Visible — the first alert test that crosses the XAML at all.
- AlertClipOutputTests is the first test in the repo to run a REAL codec: pinned ffmpeg
generates a clip, the real AlertClipDecoder + AlertOverlayLayer + SceneCompositor
composite it, and the box rect must be a colourful picture that DIFFERS from idle. It
passed while the feature was broken in the app — which is exactly why it was needed:
it exonerated decode+layer+compositor and pointed the hunt at the last hop.
Also keeps the two once-per-second alert diagnostics (resolver + preview) that settled
it, pending the creator's call on whether to keep them.
Notes: the pinned BtbN ffmpeg has no libx264, so the test clip is -c:v mpeg4.
SceneCompositor.Render fills its base with opaque black, so the test compares against an
idle render rather than counting non-zero bytes. MediaSource has the same missing trigger
in PreviewPane.xaml — left unfixed as out of scope, noted in HANDOFF.
Full suite 324/324. Lesson recorded in MyMistakes.md rule (5): a producer that hands over
a correct frame has still delivered nothing — when output arrives and the user still sees
nothing, stop auditing the producer and audit the consumer's VISIBILITY.
The 12:35 live session proved the decoder healthy (frames=240 audioChunks=156
failed=False both clips, alert audio in the mix at peakMix 0.277→0.733) yet the
creator still saw 'no video plays / you lost the video'. Root cause: pacing,
not decode. AlertClipDecoderFor built the decoder with no frameRateProbe, so
RunAsync computed frameDuration = TimeSpan.Zero and RunVideoAsync's pace step
was dead code — all 240 frames of the 10s clip dumped through the pipe in the
first ~1-2s (130MB as fast as ffmpeg read), then the box froze on the LAST
frame while audio chunk-paced its real-time ~7.8s. Reads exactly like a dead
decoder on screen. The media path already wired this seam
(MainViewModel.cs:308); the alert factory never supplied one.
Derivative fix (media/decoder plumbing, cited in broad consensus of players):
pass FfmpegFrameRateProbe into the per-play alert decoder exactly like
MediaVideoSource does. Good Dog:
AlertClipDecoderTests.AlertClipDecoder_PacesVideoFramesToTheProbedFrameRate —
real AlertClipDecoder via the fake-process/fake-probe/delay-recorder seam;
red on the old factory (zero pacing delays), green with the probe (one ~10ms
delay per frame at 100fps). Full suite 321/322, one known-env flake
(RealMouseDrag reorder). Docs in this commit: task-47 addendum,
MyMistakes.md (a decoder that drops data faster than wall-clock looks
identical to a dead one), HANDOFF rewrite.
Routing and ruling both good: 'procs in the chat windows but nothing in the alert box'
with zero exceptions — every failure stage was silent by design (Debug.WriteLine-only
preview catch, bare catch{} in RunAsync firing Completed regardless). A decode run that
starts but yields neither a frame nor an audio chunk left _clip != null with
_latestClipFrame == null, so RenderClipFrame returned null forever: the box stayed
transparent for the whole alert and the six-animation fallback (which only ran on
setup-time throws) never fired.
Fix: AlertOverlayLayer.Advance gives a started clip a NoFrameFallbackSeconds (1.0s)
grace — produce no frame AND no audio inside it and the clip is torn down, _elapsed
resets, and the SAME alert continues as the animation branch (never blank, never drained
early). Diagnostics stop the lying silence: BeginClip logs box id/path/size + setup
failures; AlertClipDecoder.RunAsync logs frame/chunk end-state and the exception it used
to swallow.
Good Dog test: SilentDecoder_FallsBackToTheAnimationAfterTheNoFrameGrace (transparent
inside grace, disposed past it, animation renders, drains to idle).
Reference: alert playback must degrade to its fallback on dead-air, never vanish —
same contract as every player/booth failure budget (e.g. OBS source fallbacks).
https://obsproject.com/forum/threads/source-visibility-fallback.187383/
Live test session proved the native alert box DID play (the alert ring was the
only source in the mix: micLevel/loopLevel 0.000 while peakMix went
0.375->0.733->0.891 after the sim injections) but the app crashed at
11:22:01.301 the moment a REAL message round-tripped through the poller:
System.ArgumentException: Must create DependencySource on same Thread as
the DependencyObject
at ...MS.Internal.Data.DataBindEngine.ProcessCrossThreadRequests()
OnMessageReceived ran RefreshAlertPreviews on the MTA poller thread and
stamped the WPF-bound VideoImageSource with a WriteableBitmap created there;
the binding engine's cross-thread re-bind killed the process. ChatOverlayLayer
already marshals this exact seam (Dispatcher.Invoke) - mirror it, guarded for
Application.Current null so the pure test seams still run inline.
Good Dog test:
AlertLayerVideoTests.OnMessageReceived_FromPollerThread_MarshalsPreviewWritesToTheUiThread
calls the seam from a raw MTA Thread while the RealApp loop runs, then asserts
on the UI thread that the preview bitmap's Dispatcher is the App's (red on the
old seam, green on the fix). WPF cross-thread rule recorded in MyMistakes.md.
Full suite 320/320, clean build 0 warnings, scope-check green.
TASK 43's alert box grows a real video celebration. Per-alert IAlertClipDecoder
(ffmpeg bgra + f32le pipes, real-time paced, disposed at drain) plays the shipped
Assets/alert-default.mp4 (stamped into the Asset table at startup) unless the
creator picks their own file — path reference only, never stored in the DB; the
six AlertRenderer animations stay the fallback. ~0.3s fade rides the alpha
envelope on straight-source copies (EOF freeze-frames then fades out); audio
forwards to a new AudioMixer alert ring (8s, 48k stereo) drained at unity — no
duck, creator ruling — scaled by volume × fade. An auto-composed marquee ticker
('Funder — Super Chat · $10.00', 140px/s) scrolls top-of-frame via a
FramePump._alertTicker seam through Render/CompositeLayers, mixed into the cache
signature (dynamic overlay, never baked). New Stream Alerts section in LeftPanel.
Derivative-work references (how OBS/Streamlabs alert boxes do per-alert video):
- https://support.streamlabs.com/hc/en-us/articles/217741147-Setting-Up-Your-Streamlabs-Alerts (custom image/video per alert type + variations)
- https://obsproject.com/kb/stream-tutorial-2-alerts (alert overlay as an on-screen zone)
- https://streamlabs.com/content-hub/widgets/alert-box (per-event alert playback)
Good Dog: AlertLayerVideoTests drives a fake IAlertClipDecoder through the whole
lifecycle in one pass (custom path wins, decoder spawns/disposes, fade envelope
0→127→255, audio volume×fade, ticker scrolls, EOF fade-drain to idle). It caught
the clip branch of Advance not clearing _current before AdvanceToNext — the layer
stayed IsPlaying after drain (MyMistakes post-mortem).
Full vstest 319/319; clean build 0 warnings; scope check green.
The liveChatId fix (TASK 44) worked on the next Test Stream, but every Mock
Chat Input send returned 'YouTube rejected the message (error 400)'. The
runtime log's body: 400 MISSING_REQUIRED_FIELD, domain
youtube.api.v3.LiveChatMessageInsertResponse.Error.
The liveChat/messages.insert snippet requires type ('textMessageEvent' or
'pollEvent') alongside liveChatId and textMessageDetails.messageText; the
TASK 41 body omitted it, and the Good Dog test asserted only liveChatId +
messageText were present — false-green while real YouTube rejected every
send. Fix body + assert type in the same test so the field can never drop
silently again.
Reference: https://developers.google.com/youtube/v3/live/docs/liveChatMessages/insert
Good Dog: ONE integration test (strengthened TestStream_DockTooling...).
Gate: clean build 0 warnings; full vstest 317/318 (the 1 failure is the
known environmental RealMouseDrag flake — passes 3/3 in isolation).
The open creator report ('I still cannot post a chat message in the TEST tab',
feedback 'Chat polling couldn't start — Mock Chat Input is disabled') was a single
cause: the liveChatId never resolved. Two YouTube API facts:
1. The id lives in snippet.liveChatId — contentDetails has no such property
(TASK 44 read part=contentDetails: could never resolve).
https://developers.google.com/youtube/v3/live/docs/liveBroadcasts
2. It only exists once the broadcast is LIVE — the official sample lists
broadcastStatus=active, and our fetch ran before the frame pump pushed RTMP
(enableAutoStart flips ready→live). A ready-state list legitimately returns
no id.
https://github.com/youtube/api-samples/blob/master/java/src/main/java/com/google/api/services/samples/youtube/cmdline/live/GetLiveChatId.java
Fix: GetBroadcastLiveChatIdAsync reads part=snippet and polls with a bounded
retry (10x/2s); PrepareAndStartLiveAsync starts the frame pump FIRST, then
resolves the id. Chat stays non-fatal. Docs ai.md/TASKS.md/HANDOFF.md +
MyMistakes recipe updated same commit. TASK 44.
Good Dog: ONE integration test (GetBroadcastLiveChatIdAsync_Polls_Snippet_...
) drives a broadcast that gains its id mid-retry and asserts every request used
part=snippet. Gate: clean build 0 warnings, vstest 318/318.
Creator: 'when I attempt to refresh my YPP page, I get an error about not being
able to reach YouTube. Seriously?' Real log: channels.list failed (403) x3.
Root cause #1 (the 403): channels.list?mine=true&part=statistics,auditDetails,
contentDetails returns 403 insufficientPermissions when the token lacks the
youtubepartner-channel-audit scope — which the auditDetails part ALONE requires,
per the docs ('A request that retrieves the auditDetails part ... must provide an
authorization token that contains the youtubepartner-channel-audit scope'). That
scope is MCN partner tooling with a 2-week token-revocation rule; the app must not
hold it. TASK-39's 'current scopes suffice, no re-consent' slice-1 claim was wrong
for this part; mock-fake tests never touched the real API, so it shipped green and
403'd every refresh since 2026-09-22.
https://developers.google.com/youtube/v3/docs/channels/list
Fix: part=statistics,contentDetails only; standing flags removed from
ChannelStatsService -> YppStatSnapshot surface -> YppTrackerViewModel -> drawer,
replaced by an honest deep-link row ('Channel standing isn't exposed to YouTube
apps — check the Earn page'). YppSnapshot standing columns stay (schema-stable,
always false). channels.list failures now log the response BODY — the bare code
could not name insufficientPermissions, which is what made this undiagnosable.
Root cause #2 (found by the new Good Dog, masked by the 403): statistics come back
as JSON STRINGS ('350'); raw GetInt64() throws. Tolerant ReadInt64 (ValueKind-first;
JsonElement.TryGetInt64 THROWS on strings — type-in, not try-type).
Good Dog: ChannelStatsServiceTests.CaptureCurrent_RequestsNoAuditDetails_AndStillParsesTheSnapshot
(URL asserts no auditDetails + snapshot parses); YppPullOutTests fixture updated.
Recipes for both 403/scope and statistics-strings entered in MyMistakes.md.
Also shipped in the same commit (shared PreviewPane.xaml + ai.md): the audio-sync
status dot removal from the #77 feedback round (creator: 'what is the point of the
status light? Lose it') — IntToSyncBrushConverter deleted with it.
verify.sh gate: 0 warnings, 316/316 pass, scope-check clean.
startup.log 2026-09-22 (user runs of the new build): ALL THREE session stops
logged 'Broadcast transition(complete) failed (403) invalidTransition → End
close-out: enableAutoStop will finish'. The old ai.md assumption ('invalidTransition
when autoStop already fired') was wrong — the blind complete POST races
YouTube/autoStop marking the broadcast complete (enableAutoStop=true is always
set, tests included), so a blind POST earned only the 403 + log noise. Per the
transition errors table, invalidTransition is a current-status problem and
complete is not gated on streamStatus (only testing/live are):
https://developers.google.com/youtube/v3/live/docs/liveBroadcasts/transition
Fix: EndBroadcastAsync pre-flights liveBroadcasts.list -> status.lifeCycleStatus
and skips the POST when the broadcast is already complete/revoked; an inconclusive
pre-check still posts (old behavior) rather than silently stranding a live
broadcast; still never throws (enableAutoStop finishes every skip).
Good Dog: EndBroadcast test rewritten to the one contract — live -> GET pre-check
+ POST transition (URL shape asserted); already complete -> zero transition POSTs;
inconclusive pre-check + 403 POST -> error string, never a throw. Full suite
315/315 (one-run audio/mouse timing flakes cleared), build 0 warnings.
Recipe recorded in MyMistakes.md; ai.md / Services index / TASK-9 note corrected.
startup.log 2026-09-22 (via the user's run of the new build): every session
start logged 'Stream health poll failed: The requested operation requires an
element of type String, but the target element has type Object'.
Root cause: GetStreamHealthAsync did status.healthStatus.GetString(), but the
real API nests it as status.healthStatus = {status, lastUpdateTimeSeconds,
configurationIssues[]} — an OBJECT, per the liveStreams reference
https://developers.google.com/youtube/v3/live/docs/liveStreams (checked the
docs before fixing, per derivative-work rule). Our flat-string fixture was the
wrong assumption since TASK 9; the report-by-exception health banner had been
silently dead.
Fix: read the nested healthStatus.status + nested configurationIssues[], tolerating
the legacy flat-string shape so nothing else breaks. Tests flipped to the real
object shape (the Good Dog for this change). Full suite 315/315, build 0 warnings.
Also recorded in MyMistakes.md (API-shape recipe) and ai.md/Services index. End-of-session
transition(complete) 403 invalidTransition (enableAutoStop fallback) observed on all three
2026-09-22 stops — separate racy-design issue, carried, not in this change.
Reorder: OnSceneElementsReordered now calls SaveLayoutNow() (was debounced
ScheduleSave) so a drag-drop persists instantly — one write stores the new
layer list (Source.SortOrder) and each layer's preview geometry (X/Y/W/H) in
the same rows. Reference: WPF ListBox drag-reorder requires an explicit persist
at drop; a debounce window lets a crash lose the drop.
Recording save dialog (RenameRecordingDialog.xaml): height 276→304 so the
Cancel/Save button row is no longer obscured; Save button named
SaveRecordingButton so the sizing test can assert it sits inside the client area.
Tests: deterministic seam test now asserts the dragged layer's geometry (preview
layout) is persisted with the new order; sizing test asserts 304 + textbox and
button-row bottoms within client area. 308/308 green.
A browser grabbing the C920 kills our reader startup: the camera stays in
MJPG 1920x1080@30 (another app's format) and our OBS-refusal to re-negotiate
under SharedReadOnly contention (`SetMediaStreamPropertiesAsync` throws
'file in use / CaptureMode is SharedReadOnly') leaves it there.
CreateFrameReaderAsync(.., Bgra8) + StartAsync then refuses with
OutputFormatNotSupported — the MJPG-active source only exposes NV12 at the
reader level (clue: startup.log 19:01/19:05 sessions, same hardware that
started YUY2 640x480 fine at 08:52). Fresh launches showed no webcam and
'Add Webcam' failed.
Reader creation is now a per-candidate ladder (ReaderSubtypeCandidates):
Bgra8 for uncompressed cameras (unchanged fast path); NV12 then the
source-default for MJPG cameras — converted in OnFrameArrived like any
non-BGRA frame. Each candidate is allocated AND started under its OWN
catch: WinRT answers an unsupported subtype with a throw (E_INVALIDARG),
not a status, so a single rejected format must degrade to the next
candidate instead of aborting acquisition (creator rule — see MyMistakes
WINRT resource-allocation recipe). Rejections are logged and collected
into the final error.
Good Dog: 4 unit tests lock the candidate ordering (MJPG never Bgra8,
case-insensitive, uncompressed keeps Bgra8 first, unknown/null -> Bgra8).
301/301 green, 0 warnings. [no push]
ty-1841: capture fixed (band ~20 updates/s, no tears) but FramePump stalls
on EVERY iteration (totalMs 21-44, render=full-render split=0 elements=6
dynamic=4) — the render ceiling, ~22-28 composites/s, caps the desktop in a
60fps file. SceneGraph can't help: the live backdrop is element 0 and cannot
be baked (a cached capture goes stale), so the cache lives at the pump.
RenderFull wraps both full-render call sites: BuildFullRenderSignature hashes
the full input identity (options rect, social bar, per-element layout/visual
bits + the frame each element would resolve through the SAME resolver seam,
using array identity + Epoch + CropBounds); unchanged identity reuses the last
composite with one Buffer.BlockCopy (~3ms) instead of a ~30ms re-composite.
Cache buffer is a separate long-lived array, written pre-burn/pre-recycle
(the caller burns the frame counter and recycles scratch AFTER render).
Engagement gated on the 1:1 config (the only deployed tier). Telemetry
surfaces `cache NR/WH` on the 5s stats line.
Same shape as OBS (sources cache their surface, the scene blits on update) —
docs.obsproject.com/backend-design, the pattern this repo cites since the
2026-09-04 paste-cache slice.
Good Dog: FullRenderCache_StaticInputs_RenderOnce_Then_Reuse_UntilInputChanges
(static scene renders ONCE + byte-identical reuse; new frame+Epoch invalidates).
297/297 green, 0 warnings, verify.sh scope-locked (FramePump.cs, FramePumpTests.cs
+ ai.md/HANDOFF/MyMistakes). No push — device re-verify next.
Measure take ty-1824 on the slice-16 build: the downscale fix worked (conv
~47ms, ring allocs 0) but desktop band was still 88% frozen at 6.8 updates/s.
Telemetry isolated the real wall — CreateCopyFromSurfaceAsync readback ~45ms
of each conversion, serialized one-in-flight => ~17/s capture cap. Docs fact:
pool-sized surfaces CLIP, not scale (Microsoft Learn), so readback stays
native; the lever is concurrency.
- MaxConcurrentConversions=3 with pool 2->5 buffers (in-flight frames fit)
- new MonotonicGate (Interlocked compare-exchange): stale OLDER completions
are dropped, never overwrite a newer LatestFrame (mirror of 1742 tear)
- FrameRingBuffer.Rent/ConsumeAllocations now lock; downscale row scratch is
per-conversion locals
- Good Dog test PublishGate_TryPublish_OnlyStrictlyNewerWins; 296/296 green,
0 warnings; docs cited Microsoft screen-capture page + libyuv fixed-point.
Local only, no push.
The 240Hz monitor delivery + one-in-flight conversions + naive double-per-pixel
DownscaleBgra (~150ms/frame under load) froze the desktop layer 90% of take
ty-1742 (6.1 fresh content updates/s, freeze runs to 2.8s; decoded raw-frame
audit). The render stat (33-36ms) was real but moot — the capture CONVERSION was
the wall, and the one torn frame was a ring slot rewritten under the consumer's
read. Reference: WGC delivers at DWM/monitor cadence
(https://learn.microsoft.com/en-us/windows/apps/develop/media-authoring-processing/screen-capture)
and libyuv row-simple/fixed-point scaling
(https://chromium.googlesource.com/libyuv/libyuv/) — the repo's own take-4 rule.
- DownscaleBgra: integer 8.8 fixed-point, shift-only-at-the-end (same two-stage
math as SceneCompositor.Bilinear). ~150ms -> ~5ms per 2.5K->1080p frame.
- 10ms MinConvertInterval: the ~4.2ms 240Hz tail stopped queuing ~150ms of
serialized conversion/s; capacity sits just above the 60/s the pump can use.
- FrameRingBuffer (depth 8, redLine 4): reuse-DISTANCE ring — a buffer is only
rewritten >=4 rents after its last hand-out else fresh-allocated, so a frame a
consumer still holds (session.LatestFrame survives conversions, dispatcher
preview lags) is never read-while-overwritten. Needs no consumer Release API.
- 2s startup.log telemetry: frames/s, conv avg/max ms, skip busy/cadence, ring
allocs — the device take is judgeable numerically.
Good Dog test: Ring_NoLap_ReusesOnlyAfterRedLineRents. 295/295 green, 0 warnings.
C4 (composite Epoch-cached downscale) deferred pending the device re-measure.
Local only, no push.
ty-1723/1726 device takes showed accelerated playback + audio tail cut-off:
a render overrun (~35ms vs the 16.6ms slot) SKIPPED the missed slots (slice 10's
freshness choice), so a 60fps-authoring pump wrote one frame per 35ms into a
60fps container — 1723: 697 frames/11.62s vs 11.84s audio; 1726: 163/2.72s vs
2.93s, video ending 0.21-0.24s early.
OBS never leaves a wall-time hole: the video thread emits one frame per tick and
a lagging producer DUPLICATES the newest frame ("lagged frames due to rendering
lag/stalls" — obs-output.c; "If the video frame queue is full, it will duplicate
the last frame" — docs.obsproject.com/backend-design). The pump's submit is now a
bounded catch-up over the missed slots (while now >= nextTick), fresh on the first,
repeated after — duration == wall, judder not fast-forward. Safe because Channel.
TryWrite never blocks (the take-9 smear was the blocking pipe-write; each emit is
nanoseconds). Burned frame index moved inside the loop: every emitted slot carries
its own +1 (also fixes the old unconditional pre-gate bump that gapped the judge
sequence on non-submitting fast-render iterations).
Good Dog test: Pump_Overrun_Renders_EmitsEverySlot_NotSkipped (60fps, 35ms render
cost, asserts >=0.65 of the wall slots emitted). 294/294 green, 0 warnings.
No push — web/A/V work is commit-local until greenlight.
The ~30Hz CapturePreviewAsync PNG poll capped real cadence at ~20Hz
(35–165ms full-HD encode+decode), so a 60fps widget still juddered at ~1/6
speed. Replaced polling with frame-driven capture of the composition
controller's root visual — the mechanism WebView2CompositionControl and
Flutter's webview_windows use (graphics_context.cc captures the root
surface_ visual via CreateGraphicsCaptureItemFromVisual; reference:
github.com/microsoft/Windows.UI.Composition.WinUI / flutter-internal
webview_windows). Frames now arrive at the renderer's own pace; capture
memory is epoch'd ring reuse + one crop-sized shared WriteableBitmap.
New Services/WebCaptureFrameSource.cs owns GraphicsCaptureItem + free-
threaded Direct3D11CaptureFramePool + session (Straight alpha readback,
per-frame FindContentBounds → CropBounds). WebView2Manager reworked around
per-session composition controllers + one UI-thread Compositor created via
the CoreMessaging CreateDispatcherQueueController P/Invoke (the 19041
projection lacks CreateOnCurrentThread); internal seam ctor
(Dispatcher, Func<string,IScreenCaptureSource>?) for hermetic tests.
CaptureScheduler.cs deleted; the three SetCaptureInterval cadence hooks
removed; InitWebView2() moved from MainWindow ctor to Loaded (a parent
HWND must exist for the composition controller); the hidden WebViewHostPanel
overlay deleted. TransparentBackgroundScript unchanged.
Tests: WebView2ManagerTests reworked — 4 control-size + scheduler tests
dropped, FindContentBounds tests moved to WebCaptureFrameSource, ONE
integration test (Frames_PublishCroppedPreview_And_CoalesceToLatest_CarryingCropBounds)
drives the seam with a FakeWebSource + background-STA DispatcherPump.
Suite 293/293, 0 warnings.
NOTE: composition path NOT yet verified on a device — the take is the
next step. Web work committed locally only (no push per standing rule).
Docs same-commit: ai.md Slice 14 + supersede marker on Slice 11, HANDOFF,
MyMistakes (CoreMessaging DQ + namespace-landmine recipe), TASK 17,
Controls/ViewModels/Services indexes.
Positive offsets still delay the whole mix via the delay line (lip-sync fix);
negative offsets now ARM once at StartLive and drop |N| ms off the pipe's write
head so audio events land earlier when audio runs BEHIND video. Slider relabeled
AUDIO SYNC, Min −500, locked while live/recording (IsEditMode). LayoutStore and
VM clamp to −500..500.
OBS reference for eat-the-head negative sync: https://obsproject.com/kb/obs-studio/buffering-time (negative sync values pull audio earlier by discarding buffered player audio).
Test: StartLive_NegativeOffset_AdvancesAudio_ByDroppingTheStreamHead (6x0.9 head
must be eaten before 0.2 bed reaches the wire).
AudioMixer.Start() begins mic/loopback capture at app startup to feed the
level meters, so both 2s ring buffers fill with pre-live audio. StartLive()
drained from the oldest tail sample, putting every recorded event ~2.2s late
in the audio track (confirmed by clap analysis + cross-correlation on two
takes: +2.11 to +2.22s).
Fix: AudioMixer.StartLive() now runs _micBuffer.Clear() + _loopbackBuffer.Clear()
immediately after the pipe starts, before the drain task runs. Recording now
begins at go-live; the <=10ms in-flight chunk evicted by Clear() is imperceptible.
Regression test: StartLive_DiscardsPreLiveBacklog_SoFirstAudioIsCurrent
saturates the loopback ring with stale 0.8 pre-live audio, then asserts the
wire carries fresh post-live 0.2 (max < 0.3).
Reference (external, per derivative-work rule): OBS 'Audio mixer' keeps its
buffers fed continuously and syncs the stream start timestamp at record time
rather than replaying pre-live capture; a go-live flush of the capture buffer
is the accepted pattern for live tools restarting a stream.
Docs: HANDOFF.md (fix shipped), MyMistakes.md (A/V sync measurement recipe).
- AudioSyncDelay.Configure reallocated/zeroed its buffer every ~10ms tick
(AudioMixer re-reads the UI setting each mix), so any non-zero sync offset
erased the just-written audio -> total silence. Now early-returns when the
delay samples are unchanged. Regression test proven both ways.
- SceneCompositor.BlitContentRaw defaulted cbW/cbH=0 when CropBounds is null
(regression from ed9d7c1) -> webcam blit to an empty rect = gray block.
Default to src.Width/Height. Regression test proven both ways.
- Truncated recordings: rawvideo mux stamps frames at declared 60fps by
arrival; a scene whose first layer is dynamic (hidden elements still count)
kills the bake cache -> full render ~35ms -> ~27fps submitted -> halved
file length. Static scenes bake once (246ms cold, then <1ms) -> 60fps,
full-length (probe + 13:53 take, 301/300 per 5s, 12.46s file from 12.3s
wall). FramePump.ProbeRender names the hot render path on slow frames.
The element-space raster builds on a TRANSPARENT base but BlitContentRaw's
partial-alpha branch applied the opaque-dst source-over blend: color premultiplied
by the sampled alpha, then alpha forced to 255. Pasting that raster saw a==255 and
straight-copied darkened ink over the scene — a recording box that the raw-bitmap
preview (correct alpha) never showed. Transparent margins and opaque content were
unaffected, which is why every take looped on the page/CSS while the capture was
transparent all along (15:51 dumps: alpha max 255, mean ~19, zero 57%).
BlitContentRaw now takes transparentDst; the raster call passes true and writes
straight color + straight alpha so the paste rows (BlendRowOpaque/Weighted) do the
real source-over onto the opaque master. Master paths byte-identical.
ONE integration test PasteCache_SemiTransparentLayer_RevealsBackdrop_NotOpaqueInk:
50%-blue over red reads (127,0,128) fixed vs (0,0,128) buggy — proven both ways
(verified by stashing the fix: fails before, passes after). Clean build, 0 warnings;
22/23 compositor-class tests pass, the sole failure the documented pre-existing
Composite_FullScene_MasterPixels pixel (1380,700).
Alpha-compositing model: standard source-over with producer-cached surfaces, the
OBS/libyuv paste model already cited in ai.md/MyMistakes (rawvideo recipe, row-blit
BLEND_NONE / straight-alpha branches); full story in MyMistakes (RESOLVED entry).
Per the good-dog rule: one integration test, memory updates (MyMistakes/ai.md/
HANDOFF) in the same commit.
The 15:34 take's box interior is the widget's OWN full-canvas opaque paint — a
black void with wide content strips at the top/bottom and a right-edge bar —
arriving AFTER the first-second blank-transparent dumps. A background-color-only
!important wipe (b4bba4b) can't touch it because CSS gradients/backdrops are
background-IMAGE, not background-color.
OBS's fix for this exact symptom is background-image:none (obsproject/obs-studio#6659:
"set the CSS for html and body to background: none !important"). Injection now wipes
background-image:none!important on html,body,html * alongside the color wipe; HTML
overlay art (<img>/DOM/CSS shapes) survives — that is browser-source semantics.
Also re-arms WidgetDumpRemaining=5 ~30s after nav so the next take dumps the real
document INSIDE the recording window (the previous dumps proved blank at +1s and
missed the later backdrop paint).
9/9 WebView2Manager tests, 0 warnings.
The real-widget dumps (12:54/13:50) proved the OLD inline
element.style.background='transparent' injection holds only while the page has
nothing to paint: the capture was alpha-transparent, yet the recording showed a
black opaque box over the whole element rect (1231,679 705x396) once the widget
connected and repainted a container background-COLOR — CapturePreviewAsync always
honors page CSS (MicrosoftEdge/WebView2Feedback specs/BackgroundColor.md), so any
page-painted background wins over DefaultBackgroundColor.
This is the OBS-solved class (all web-uri resources paint their own background):
browser sources use a Custom CSS override, and the decade-validated formula for
arbitrary pages is a pre-parse <style> with 'background-color: transparent
!important' — https://obsproject.com/forum/threads/translucent-transparent-browser-source.59549/
('body { background-color: rgba(0,0,0,0) !important }') plus the div-level variant
for stubborn widgets (woahtech.com OBS custom-CSS guide).
Injection is now an idempotent pre-parse style element wiping background-color on
html,body,html * with !important (outranks every page rule, runs before page parse
via AddScriptToExecuteOnDocumentCreatedAsync). Only background-COLOR is targeted —
background images and widget art survive. Regression test asserts the element-wide
!important form and that the losing inline form is gone.
9/9 WebView2Manager tests, 0 warnings.
The one-shot dump + alpha log was bound to the FIRST capture ever = the initial
about:blank placeholder (alpha=0, rgb=0, 5/5 sessions) — a blind instrument. The
pre-parse transparency injection (1a39b09) is intact but was NEVER verified
(MyMistakes '? VERIFY'), and the 2026-09-10 take still shows a black, opaque
block. Stop guessing: NavigationCompleted for a non-about:blank document now
arms WidgetDumpRemaining=5; the next 5 post-paint captures dump
%TEMP%\ytLive-web-<id>-w1..5.png + shared AlphaStats(min/max/mean/zero%) +
FindContentBounds to startup.log.
The stats line alone names the branch: zero%≈opaque ⇒ the injection did not hold
for this widget's CSS (fix: !important stylesheet / chroma-key); large zero% +
tight bounds ⇒ capture IS transparent and the black lives in the compositor
blend. No new test: WebView2 runtime is not instantiable in the suite and the
re-point is log-only; 8/8 WebView2 tests still pass.
The recording is 60fps but WebView2 capture was a blind 100ms DispatcherTimer = 10Hz;
each captured web frame repeated ~6x into the file caps web animation at the capture
rate, not the page's (user: 'the animation appears to be too slow').
- New CaptureScheduler (Services/CaptureScheduler.cs): per-session dispatcher timer
that DROPS a tick while a capture is in flight (latest-wins, never queues) — the
guard that makes a higher cadence safe: concurrent full-HD PNG CapturePreviewAsync
calls (~10-30ms each, slow per WebView2Feedback#20) would stack CPU and publish
stale-after-fresh. Effective cadence = max(interval, capture duration).
- Cadence: SetCaptureInterval(33) on record/stream start, (200) idle — applied via
MainViewModel.Streaming.Operations.cs.
- De-throttle the hidden page: shared CoreWebView2Environment created BEFORE
EnsureCoreWebView2Async with --disable-backgrounding-occluded-windows
--disable-renderer-backgrounding --disable-features=CalculateNativeWinOcclusion.
Off-screen WebView2 is a hidden page when the host window is unfocused/covered and
Chromium then parks rAF and clamps timers to ~1s (WebView2Feedback#1172/#3070,
Chrome-88 timer-throttling blog).
- Telemetry: first 30 captures per session log elapsed ms (PNG encode + decode) to
startup.log — that decides whether ~30Hz stays or drops to ~20Hz; the FramePump
drops frames (never time-lapses, slice 10) if UI-thread GC churn starves it.
- ONE test: CaptureScheduler_Drops_Ticks_While_Capture_InFlight_And_Resumes
(deterministic TCS-driven, no WebView2 runtime). Suite 290/291 — sole failure the
pre-existing compositor pixel test.
- Docs same-commit: ai.md slice 11, MyMistakes.md, HANDOFF.
References: https://github.com/MicrosoftEdge/WebView2Feedback/issues/1172https://github.com/MicrosoftEdge/WebView2Feedback/issues/3070https://github.com/MicrosoftEdge/WebView2Feedback/issues/20https://developer.chrome.com/blog/timer-throttling-in-chrome-88
slice 9 made the DURATION right but content still hiccuped; aggregates (301/300, uniform
file PTS) could not see it. Measured root cause: FfmpegEncoder.SubmitFrameAsync BLOCKED
on WriteAsync(8.3MB)+FlushAsync when ffmpeg lagged the pipe, and the burst while-loop
re-wrote that same stale composite per crossed slot — frozen runs.
OBS shape (derivative, wrapped pre-1.0): the encoder queue in libobs/obs-encoder.c —
encoder thread never couples back into the video thread; overflow = dropped data, never
a frozen producer. https://github.com/obsproject/obs-studio/blob/master/libobs/obs-encoder.c
- FfmpegEncoder: SubmitFrameAsync is now an enqueue (ArrayPool copy) into a bounded
Channel (cap 120) drained by its own task; drop-newest + count when full;
StopAsync flushes the queue then EOF (TryComplete). IFfmpegEncoder.DroppedFrames.
- FramePump: ONE fresh composite per iteration (burst loop deleted); worst-submit stat,
stall logger (>2x interval names the stage), dropped/stalls in stats.
- Burned-in 6-digit dot-matrix frame counter (white box, bottom-right) on every composite
— the clock-independent judge replacing the WSL ticker: +1/frame, jumps = counted drops.
- ONE new test Backpressure_QueueOverflow_DropsFrames_AndNeverBlocks (slow-sink fake:
submit never blocks, drops counted, stop flushes exactly submitted-minus-dropped).
- Full suite 290 tests, 289 pass — sole failure the pre-existing compositor pixel test.
- Docs same-commit: ai.md slice 10 (+ encoder/stop-note corrections), MyMistakes point 8,
HANDOFF.
Audio untouched (queued follow-up); web overlay still frozen pending timing closure.
Root cause (measured, not guessed): the take-3 'rebase on overrun' reset
nextTick to wall-now, erasing every missed slot. Pump delivered 215-219/300
per 5s (~43fps) but rawvideo carries no timestamps — ffmpeg muxes by frame
count at -framerate 60, so every recording played fast with honest-looking
stats. A second pacer, ffmpeg -re, throttled the demux separately
('Resumed reading ... after a lag' 0.79s->4.82s).
Fix follows libobs video-io.c (https://github.com/obsproject/obs-studio)
- the video thread never resets its deadline; one frame per interval slot,
a late render repeats content (judder), never skips time:
1. FramePump: count-based emission, while(now>=nextTick){Submit; nextTick+=I}
2. FfmpegArgs: -re removed - the pump is the pacer
Muxed duration is now frame-count/fps == wall time by construction. Covers
game background + webcam (one shared pump). 0 warnings; suite 288/289 -
the one failure (Composite_FullScene_MasterPixels 1380,700) also fails with
this change stashed: pre-existing, untouched, recorded as follow-up.
The widget page paints html/body opaque, so CapturePreviewAsync output had
zero alpha-0 margins — every take since bccdb48 built the crop on 'the
capture is transparent' and never verified it. WebView2 spec is explicit:
'WebView will always honor a webpage's background content' and
DefaultBackgroundColor only shows through pages with no background style
(MicrosoftEdge/WebView2Feedback specs/BackgroundColor.md). The late
ExecuteScriptAsync injection (NavStarting/NavCompleted) ran after page CSS
and lost the fight.
Fix via CoreWebView2.AddScriptToExecuteOnDocumentCreatedAsync — the
documented pre-parse hook ('before the HTML document has been parsed and
before any other script included by the HTML document is run'), the same
mechanism OBS user.css uses. Nav handlers kept as post-load re-assertion.
Restores the take-23/24 stride-correct crop path (my take-25 removal was
wrong — it regressed the bounding box into a solid black box).
Adds one-shot diagnostics: raw capture PNG -> %TEMP%/ytLive-web-<id>.png
plus alpha min/max/mean/%zero and FindContentBounds result in startup.log,
so the next run proves the capture is transparent instead of guessing.
Build 0 warnings, 289/289 tests pass. MyMistakes.md records the full chain.
User tested: 'didn't work at all, and broke additional crap.' Second failed remedy for
web-source-over-webcam dark composite -> AGENTS.md spin guard: no third guess; next attempt
requires external research (OBS/CEV browser-source transparency in recorded output, cited URL)
plus the blank-page falsifier. HANDOFF OPEN list updated: research-first step added, #11 noted
landed as d35823a.
take-19-2/take-20 'webcam black square under the web widget': the capture was
alpha-cropped (FindContentBounds) and that CROP was fed to the compositor, whose
UniformToFill zoomed the opaque content box to cover the whole element rect the
moment the widget drew any content (idle transparent page = full-frame crop = the
correct viewport, hence take-1-good/take-2-bad). OBS model verified: the page is a
fixed 1920x1080 canvas and the element rect is a viewport onto it — measure with the
alpha crop (preview/selection), hand the composite the FULL canvas on an 8-deep ring
+ Epoch (same identity discipline as camera/screen). Regression test:
Composite_FullCanvasWebSource_TransparentMarginsRevealWebcam (the reveal contract:
margin pixel = webcam, badge pixel = widget).
Roll-forward of today-slices 6fd1d9c onto the slice-8 base, two-loop hunk dropped.
Release counter (per-build GUID read as noise; +1 per commit from git rev-list,
baseline 241 -> #13, generated by GenerateBuildStamp; GUID demotes to startup.log).
Tests pinned to Label/#N >= 13; wordmark display test asserts the Label.
Flash fix (take 14 finding): consumer holds must never outlive depth x source
period — 4 slots at high refresh lap ~27ms vs a <=50ms compositor read, so a
recycled slot flashed its new frame over the lagged old one. All shared rings 4->8
(OBS/overlay precedent for ring discipline).
Camera producer now rotates an 8-deep ring + Epoch instead of a fresh ~3.7MB
array per device frame (110-220MB/s LOH churn); WebView2 capture reuses a canvas
scratch + 8-deep output ring + a reused WriteableBitmap instead of two fresh
arrays + a fresh bitmap per 10Hz tick. Paste cache stays identity-keyed (Epoch).
Take 10 (59a02a5b, slice 6) finally produced a self-contradicting stat: render
22.4ms + submit 2.5 against a 16.7ms deadline, yet avg wait 10ms — a rebasing
pacer CANNOT sleep after a blown deadline. The wait was queue time: StartAsync
fires from a UI command handler, and async continuations re-capture the current
SynchronizationContext — the 'WPF-free, hermetic' frame pump had been rendering
ON THE DISPATCHER behind the live preview the entire starvation saga. OBS keeps
obs_graphics_thread/video_thread off-UI for exactly this reason (dedicated
threads; see docs.obsproject.com/backend-design 'Libobs Threads').
- FramePump: _pumpTask = Task.Run(() => PumpAsync(...)) — null context inside,
every continuation stays on the pool.
- Audited, not ignored, what that exposes: StaticPixelCache.Get now locks (pool
miss-decodes raced UI callers); ChatOverlayLayer.RenderFrame checks its cache
off-thread but marshals the rare raster MISS to the dispatcher (DrawingVisual
+ RenderTargetBitmap are UI-thread objects) and re-validates there; pump
events already marshal in the VM.
- GCLatencyMode.SustainedLowLatency for the pump's life (restored in finally).
- Stats gained 'worst render Xms' — bimodal averages hid per-tick spikes.
- Webcam routes through the paste cache (the IsOpaque bypass re-sampled ~156k
px every tick even between identical device frames).
ONE integration test: Pump_Produces_OffTheStartingContext — an inline-pumping
SynchronizationContext makes the old construction run the resolver on the
starting thread by capture; the loop must never. 70/70 per-class green, clean
build 0 warnings. Docs same commit (ai.md slice 7, TASKS take-11 gate,
MyMistakes #6, HANDOFF). take 11: ~300/300 + honest wait -> saga closed,
Unit B (two-line top bar spec, fully captured) starts.
Take 9's numbers were decisive: paste cache moved work to ~25ms/frame but the
period stayed ~37ms. The missing ~12ms per tick is Task.Delay rounding every
sub-tick request up to the Windows system-clock tick (~15.6ms default —
documented: learn.microsoft.com/en-us/dotnet/api/system.threading.tasks.task.delay).
A frame finishing 3ms early requested 3ms and slept 15.6. Producer capped at
~27fps no matter how fast the compositor got — which is why two real render
fixes read as 'zero change' in playback. Game-loop/OBS canon for this
(stackoverflow.com/questions/5441464; learn.microsoft.com/en-us/windows/win32/
api/timeapi/nf-timeapi-timebeginperiod): raise the timer resolution for the
session, sleep only the bulk of the remainder, and SPIN the last ~2ms across
the deadline.
- FramePump: timeBeginPeriod(1) on entering the pump loop, timeEndPeriod(1) in
the finally; pacing = bulk _pacingDelay(ahead - 2ms) + bounded Thread.SpinWait
tail; blown deadlines rebase unchanged (never burst).
- Stats now report avg wait: render+submit+wait must equal the period — the
accounting is closed, no stage can hide in an unmeasured gap again.
- Webcam dropped its IsOpaque paste-cache bypass: it re-sampled ~156k px every
tick even between identical device frames; cached paste beats the sampler on
hits, costs the same on misses.
52/52 per-class green (pacing + pixel suites unchanged — output byte-stable),
clean build 0 warnings. Docs same commit (ai.md slice 6, TASKS.md take-10
gate, MyMistakes #3 + renumber, HANDOFF). User's top-bar spec remains next in
queue (Unit B) — re-sent many times, captured, no open questions.
The stamped build settled what slices 3-4 could not: chat cache works (resolve
~0.0ms) but render stayed 26-27ms -> 124-135/300. The cost was the compositor
re-rasterizing EVERY layer every tick: this Live scene re-samples chat (159k) +
web widget (271k) + image (95k) + cam (156k) ~ 680k px @ ~38ns — for layers
whose pixels do not change between chat/web/cam updates.
OBS shape: cache the surface, paste per tick. BlitCachedLayer rasterizes a
non-opaque layer ONCE into an element-space, transparent-based frame keyed by
(source-array identity, src W/H, ceil'd dst rect, round, mirror), then pastes:
integer position, row alpha-blend, opacity applied at paste. Producers hand out
fresh immutable arrays -> array-identity keys cannot serve stale content; dict
bounded (48, clears whole). Drag/opacity live in paste params, not keys, so
editing stops triggering resamples too. Opaque backdrop keeps the memcpy path;
the webcam keeps the direct path via its IsOpaque flag (revisit if take 9 is
borderline).
ONE integration test: PasteCache_RepeatRender_IsByteIdentical_And_ContentChange-
Propagates (byte-exact raster-vs-paste incl. round-clip margins, new-array
propagation); existing pixel suite guards sampler semantics. 85/85 across
compositor/pump/chat/capture/session classes, clean build 0 warnings. Also:
BuildStampTests.cs was written last commit but never staged — its own scope-check
slip, added here (the run had used the on-disk file; tracked now).
Docs same commit: ai.md slice 5 + stale 'general path only 130k' claim corrected,
TASKS.md take-9 gate, MyMistakes recipe (prove the stage; a fix that doesn't move
the stat wasn't the bottleneck), HANDOFF. take 9 expectation: 300/300, render
<= ~8ms -> saga closes, Unit B starts.
Take 6 measured render 35-41ms — WORSE than take 5's 25.5 — and the run could
not be attributed to a binary: exe mtime != build contents (incremental builds
serve stale exes; a source edit without rebuild is a silent old binary). Three
takes of a perf saga had been judged against builds nobody could prove.
- ytLive.csproj GenerateBuildStamp target: fresh GUID per compile (writes
obj/BuildStamp.g.cs -> Helpers/BuildStamp.Id/BuiltLocal). Deliberately defeats
incremental lies: every 'dotnet build' recompiles the app project.
- Wordmark shows the id as a superscript (TopBar.xaml, x:Static, 9px grey
BaselineAlignment=Superscript); startup.log records 'Build <id> (compiled
<time>)' so every take is cross-readable with the visible UI.
- FramePump stats split the tick: 'avg render Xms (resolve Y), avg submit Z' —
the resolver is timed separately (wrapper resolver on per-tick paths; bake
keeps the raw one) so take 7 names the hot half of 'render' with data.
- Fixed a latent transition-clock bug found on the way: lastTick now restarts
every frame (the branch rework had restarted it only during transitions,
letting a transition begun after idle complete instantly on its first Tick).
ONE integration test family: BuildStampTests (unit: shape) +
BuildStampDisplayTests (RealApp, namescoped FindName on TopBar proves the
wordmark SHOWS the id). 47/47 per-class green, clean build 0 warnings. Docs
same commit. User's top-bar/session spec (re-sent twice) + settled Q&A
decisions folded into HANDOFF Unit B — next work unit after take 7 verdict.
Take 5: render 58.9 -> 25.5ms (138/300, still ~2.2x). The blits were fixed; the
resolver was not: ResolveOutputFrame -> RenderChatBox ran a FULL WPF raster
(FormattedText + RenderTargetBitmap + CopyPixels + channel swap) EVERY tick
whenever the chat buffer was non-empty — and the buffer survives sessions, so
even a signed-out record-only take paid it. Established answer (OBS text
sources): re-render on change, blit the cache every tick.
ChatOverlayLayer: content version bumped from Messages.CollectionChanged
(covers adds, the 500-cap removal, the fade Clear from any caller) + a config
key (size + all Chat* appearance props); RenderFrame returns the cached
VideoFrame by identity until either changes (compositor only reads cached
frames). Conservative ordering (version latched BEFORE render) makes a
mid-render message re-render next tick, never serve stale.
ONE integration test: ChatOverlayLayerCacheTests (RealApp, real renderer):
Same() for unchanged inputs, NotSame() on message/config change, null on
empty. Full regression green (62 across touched classes), clean build 0
warnings. Accepted cost pending take 6: one ~15-25ms tick per arriving
message; if live-chat bursts sag n/300, next slice = debounced off-tick
re-render. Docs same commit: ai.md pipeline section, TASKS.md TASK 18,
MyMistakes recipe (raster-on-change + session-surviving-buffer trap),
HANDOFF (take 6 -> then Unit B, spec unchanged).
Two defects made the producer 17x slow (37s record -> 2.1s/127-frame file,
rawvideo stamps by arrival): FramePump slept the FULL interval after each
render (period = render+submit+interval) and SceneCompositor did per-pixel
float sampling + Math.Round blends over all 2.07M master pixels, scanning the
whole destination per overlay (258ms avg render vs 1.5ms submit).
Both solutions are established, not invented — researched before coding per
the derivative-work rule:
- deadline pacing: OBS libobs/media-io/video-io.c video_thread (nextTick +=
intervalTicks, sleep only the remainder, rebase on overrun, never burst)
- row blits: libyuv pattern (BSD-3, chromium.googlesource.com/libyuv/libyuv)
— 1:1 aligned identity fast path, per-pixel alpha branch, integer
fixed-point blend, overlay clipped to the intersection rect, skip the dead
black pre-fill when the backdrop covers
ONE integration test: Pump_Paces_To_The_Deadline_Compensating_Render_Cost
(lands after a fake-seam lesson: pacing fakes must await, not complete
synchronously, or the pump loop runs inline on StartAsync and hangs vstest).
Clean build 0 warnings; FramePumpTests 10/10, SceneCompositor/SceneGraph/
SocialBar/StretchMath 20/20. Docs same commit: ai.md pipeline section,
TASKS.md TASK 18 (webcam-in-output + rename modal verified from take 3),
MyMistakes recipe, HANDOFF rewritten. Take 4 pending on the user's machine.
- Modes: creator ruling 2026-09-01 — the VOD is the copy; dual encode degrades both
outputs on mediocre hardware ('we're not them'). ai.md 'cheap on NVENC' claim
retired; TASK 18 three-modes -> two; pills-become-radios noted as PENDING code.
- Design Principle gains 'we're not them' (hardware realism) as its fifth bullet.
- TASK 34 scope: countdown+notify live scheduling is the whole story — 'going live is
our schedule'. Scheduled playout discussed & DECLINED; premiere/upload + playout +
simultaneous-record-stream added to the closed out-of-product table.
- TASK 22: SYNC slider provenance restored — creator-requested (OBS delay-filter fix,
native). Placement question stays open; capability does not.
- MyMistakes: the provenance rule (a fact without its who/why becomes a future argument).
Docs only — zero code in this commit, per creator instruction.
gramps