feat: add vse-cut-list example witnessing the 4.5-to-5.x sequencer API rename

[TMHSDigital]

Summary

Ships examples/vse-cut-list/ — the gallery's first sequencer example and the ROADMAP pool's "VSE sequences-to-strips witness" (pool entry removed, two newly identified candidates restocked). It witnesses the 4.5 LTS → 5.x sequencer API rename, the single most common way AI-generated VSE code dies on a modern Blender: every tutorial-era snippet does sequence_editor.sequences.new_effect(..., frame_end=...), and on 5.x that path is gone twice over.

API contracts witnessed

• Accessor rename — 5.x: sequence_editor.sequences raises AttributeError, only .strips exists. 4.5: both accessors exist and see the same strips (transition bridge asserted).
• Creation signature — strips.new_effect(...) ends a strip with frame_end= on 4.5, length= on 5.x; the wrong kwarg raises TypeError on each side (asserted, not assumed). 5.x also rejects the removed TRANSFORM effect type; per-strip strip.transform is the cross-version placement API.
• Frame reads — closed-form end-exclusive spans on canonical accessors: frame_final_* on 4.5; left_handle/right_handle/duration on 5.x, where frame_final_* is deprecated (removal announced for 6.0) and asserted still equal, with the RNA is_deprecated flag checked per side.
• Effect wiring — input_1/input_2 (not input1) on both versions, authored order, cross clamped to exactly the source overlap.
• Scene strip — 4-arg new_scene(...) on both; default span == source scene frame range; sources a separate Stage scene.
• Round-trip — spans, channels, colors, wiring, text, and pixel-unit transforms (tol 1e-3) survive save_as_mainfile → open_mainfile.
• Compositing (--check-pixels) — 96×54 render: cell centers carry strip colors (tol 0.1), cross cell is a true mid blend (strictly between sources per channel), margin matches neither cross source.

Hazards discovered while authoring (witnessed in code + README)

• GAMMA_CROSS asked to outlast the input overlap is silently clamped (request 9 over (25,33), get (25,33)).
• A scene strip pointing at its own scene renders transparent (alpha 0) — feedback loop; first-draft "studio" was the viewer showing alpha as grey.
• Effect strips consume their inputs — inputs never composite on their own channel, and the effect's transform applies over theirs; consumption requires the effect above the inputs (GC below T1/T2 let T2 paint the whole wall teal).
• An empty bpy_prop_collection is falsy — se.strips or se.sequences silently picks the legacy accessor on an empty timeline.

Falsification evidence (break-the-contract probes, all on the shipped script)

Probe	Result
Legacy se.sequences.new_effect(frame_end=...) on 5.1.1	AttributeError: 'SequenceEditor' object has no attribute 'sequences'
new_effect(frame_end=9) on 5.1.1 (guarded assert)	TypeError ... expected (name, type, channel, frame_start, length, input1, input2)
Expect GC span (25,34)	exit 5, measured GC span (25, 33, 8) != closed form (25, 34, 9) (also on 4.5.11)
Swap GC inputs	exit 7, measured inputs=(T2, T1), expected T1 -> T2
Corrupt text strip before save	exit 10 via reloaded re-assert (exit 9 inner)

Measured check values (CI logs carry these)

• 5.1.1: accessor: strips only (.sequences removed); end kwarg: length; spans A(1,33) B(25,57) C(1,57) GC(25,33) SC(1,251) on left_handle/right_handle/duration; GC input_1=T1 input_2=T2 clamped; transforms tol 1e-03; all survive save/reload
• 4.5.11: identical values on frame_final_*, .sequences bridged to .strips, frame_end
• Pixels (identical on 5.1.1/4.5.11, Cycles and EEVEE): TL(0.851,0.102,0.22) TR(0.059,0.749,0.8) BL(0.949,0.62,0.102) BR(blend)(0.341,0.349,0.463) margin(0.318,0.314,0.322)
• Hero parity: 5.1 vs 4.5 render mean|diff| 0.00002, mean luminance 0.3767 vs 0.3768

Visuals

The still is the sequencer's own output at frame 29 (mid cross): crimson/teal/amber program cells + the 50/50 cross-blend cell over the Stage scene strip (dark studio, warm wedge pool), caption carrying the closed form. An off-by-one in end-exclusive span math drops its cell to the dark stage.

Contact sheet (candidate leftmost, then the pinned calibration set): docs/gallery/contact-sheets/vse-cut-list-contact-sheet.webp

Per-criterion verdicts vs armature-bend / damped-track-aim / bmesh-gear:

• Stage darkness: candidate left-band 0.173 (family 0.122–0.202) ✓; top-band 0.309 vs 0.279/0.052/0.189 — brightest top but inside the armature-bend envelope ✓
• Wedge warmth: warm pool upper-right, consistent with family ✓
• Subject fill: mosaic 74% width / 72% height (spec 70–90%) ✓
• Saturation: Standard transform, authored strip colors read exact ✓
• Thumbnail legibility: 4 blocks + dark stage read instantly ✓
• Mean luminance: candidate 0.372 vs 0.279 / 0.172 / 0.305 — highest (saturated cells cover ~53% of frame), not sortable as the odd one out in the lineup

Test plan

• Local binaries: C:\Program Files\Blender Foundation\Blender 5.1\blender.exe (5.1.1, self-reported) and .scratch/blender-4.5.11-windows-x64/blender.exe (4.5.11 LTS, self-reported, official download).
• Both run: check-only (exit 0), --check-pixels --engine cycles (exit 0), --output hero render (EEVEE on 5.1.1 GPU host; parity render on 4.5.11).
• CI: blender-smoke 4.5/5.1 run both the check and the pixel witness (Cycles CPU).

Proven by live run vs inspection only

• Live run (both binaries): all checks, falsification probes, pixel witness, hero renders, hero parity diff, webp asset generation, contact-sheet composite + luminance stats, gallery regeneration.
• Inspection only: generated HTML alt/witnesses text (read post-generation), README/manifest/count wiring (also re-verified locally with the CI jobs' logic), landing-page template having no hardcoded counts.