Merge main into M13.3 (tracking completion)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

docs/architecture.md

@@ -300,6 +300,42 @@ demand (`self_at`) and `roto.propagate` fills in the background; `Session::edit`
 Key in step with the layer's source, masks and upstream effects. Freeze stores final 8-bit mattes
 in the effect.
 
+**Content-Aware Fill** (`raster::inpaint`, `engine::commands::content_fill`): the hole is
+where the layer's masked source (`Renderer::layer_input` with no effects) is under half
+opacity, grown by Alpha Expansion. Optical flow between neighbouring membrane-filled frames is
+*completed* inside the holes (membrane interpolation of the surrounding tiles' motion); each
+missing pixel follows it forwards and backwards to the nearest frame (or reference frame) where
+it is visible, the two candidates blended by temporal distance, with Lighting Correction adding
+the membrane of the border mismatch (Poisson-style). What no frame shows is synthesised by
+PatchMatch (Barnes et al. 2009) inside Wexler et al.'s (2007) multi-scale EM: Object refines each
+frame starting from the previous result warped along the flow; Surface carries one synthesised
+fill through the range; Edge Blend is the membrane fill alone. The frames are written as a PNG
+sequence and imported into a "Fill" layer above the source. It runs as a generic **background
+task** (`engine::jobs`: a closure with progress and cancel on a thread, inline on wasm32 or
+with `wait`, whose result is applied on the UI thread as one undo step); the Progress panel and
+`jobs.list` / `jobs.cancel` show these tasks together with the render queue and the analyses.
+
+**Scene Edit Detection** (`raster::cuts`) compares consecutive frames by colour-histogram
+distance and motion-compensated (block-flow warped) luma error, takes their geometric mean and
+marks frames that exceed a threshold and several times their neighbours' median; the cuts
+become layer markers, split layers, or split-and-precomposed layers. **Auto-trace**
+(`path::trace`) runs marching squares on the chosen channel, Douglas–Peucker simplification and
+cubic fitting with Corner Roundness blending chord and smooth tangents; holes become Subtract
+masks, and a work-area trace keys Mask Path per frame.
+
+**Align Video to Data** reads data footage (File ▸ Import of `.json`, `.csv`, `.tsv`, kept as
+text in the project). Supported data: JSON as an array of objects or an object holding one
+(`{"samples": [...]}`), CSV/TSV with a header row (quoted fields allowed). The time key is the
+`key` parameter or the first field named `time`, `timestamp`, `t`, `date`, `datetime`, `utc`,
+`gps_time`, `seconds` or `time_s`; values may be ISO 8601 date-times (with `Z` or `±hh:mm`),
+times of day (`hh:mm:ss.sss`), timecode (`hh:mm:ss:ff` at the footage rate), Unix seconds or
+Unix milliseconds. The layer's Start Time becomes `dataStart + videoStart − firstSample`
+(`videoStart` given, or the file's creation date on the desktop; reduced to a time of day when
+the data has no dates).
+
+**Lumetri Scopes** (`raster::scopes`) are density plots of the viewer frame (rendered over the
+comp background) with BT.601 / BT.709 / BT.2020 luma and colour-difference coefficients.
+
 ## 5. Expressions
 
 Expressions are JavaScript, run by boa, with After Effects' object model (`thisComp`, `thisLayer`,

docs/parity.md

@@ -22,9 +22,7 @@ Tracking 96%, Effects 85%, Web 65%.
 What is left, in priority order: the web app's depth (threads, storage, audio, non-blocking
 renders); stroke taper/wave and multi-segment dashes; variable mask feather points; camera iris/bokeh and focus-link commands; Render Queue field render, crop/resize and templates;
 approximated effects (Key Cleaner) and Liquify's viewer brush; OpenType features and variable
-font axes; AI/EPS/PDF import; more codecs; 59 preferences not yet wired; ScriptUI;
-Lumetri Scopes, Footage, Media Browser and Metadata panels; Content-Aware Fill; and
-the "better than After Effects" items (a plug-in API, branching history; GPU particles landed in M12.7).
+font axes; AI/EPS/PDF import; more codecs; 59 preferences not yet wired; ScriptUI; and the "better than After Effects" items (a plug-in API, branching history; GPU particles landed in M12.7).
 
 The sections below are the original audit (morning of 2 October, ≈ 64%) and its updates, kept for
 history.
@@ -165,6 +163,25 @@ style runs; the Layer Style dialog (LYR-9); Graph Editor snapping to markers and
 Reference Axes; Animate Text ▸ Variable Font Axes. Remaining in these rows: Lottie can't carry
 taper/wave, Advanced 3D's DOF has no iris shapes, and variable axes don't change advances.
 
+## Update: M13.6 panels and content tools
+
+The last placeholder panels became real panels with automation ids: **Lumetri Scopes**
+(waveform RGB / Luma / YC, vectorscope YUV / HLS, histogram, parade RGB / YUV; Rec. 601 / 709 /
+2020, 8-bit or float scale, clamp; `scopes.analyze`), **Footage** (double-click footage: its own
+time ruler, play, In/Out, Overlay Edit and Ripple Insert Edit; `footage.*`), **Media Browser**
+(desktop file system, favourites, thumbnails, import and drag to the Project panel or timeline;
+`mediaBrowser.*`), **Metadata** (codec, size, rate, duration, colour profile, file dates; item
+and project comments; `item.metadata`) and **Progress** (every render and analysis with
+progress and cancel; `jobs.*`). **Content-Aware Fill** (panel and Layer ▸ New ▸ Content-Aware
+Fill Layer): Object / Surface / Edge Blend, Work Area / Entire Duration, Alpha Expansion,
+Lighting Correction and a reference frame from a layer, rendered to a PNG sequence in a Fill
+layer above the source as a background job. **Scene Edit Detection** (markers, split, split and
+precompose), **Auto-trace** (alpha / RGB / luminance, current frame or work area, to the layer
+or a new layer) and **Align Video to Data** (JSON / CSV / TSV time keys) replace their disabled
+menu entries. Not yet: creating a reference frame by editing a still (After Effects hands it to
+Photoshop), Content-Aware Fill's learned model (ours is PatchMatch + flow propagation), and Media
+Browser in the web app.
+
 ## Highest-value gaps, in order
 
 1. ~~On-canvas text editing and per-character styles~~ (landed: M9.9–M9.10).