FORGED BY SIN/ HOURGLASS EVIDENCE

Reproducibility and claim boundaries

Evidence, not a tutorial-shaped guess.

This page records what was downloaded, how every frame was retained and audited, how the capture mechanism was investigated, what was executed in Blender, which claims were corrected, and what remains unknown.

Source acquisition

Exact public media, hashed before interpretation.

Main timelapseMedia 2059266635484053504
1280×720 · H.264 High · 15 fps
295.466667 s · no audio
4,432 decoded frames
Main SHA-256498cba9c63fd358f9ab4c63209bd8c03e15b643fdec66a529daa5f9928b98957
Companion resultMedia 2056054367879417856
720×720 · H.264 High · variable timing
3.96 s · no audio
119 decoded frames
Companion SHA-256978b6b2d60e6c1d3afab5a7aa5dddea767fdc77b8683736639f1e0a6269f61b1

The formats were explicitly selected with yt-dlp. FFmpeg extraction used passthrough timing: no `-r`, fps filter, interpolation, decimation, `mpdecimate`, or scene-selection filter. Per-frame FFprobe PTS is authoritative, especially for the variable-timing companion.

Capture forensics

The clip is timed screenshots, not an input recording.

Main frame F4431 visibly shows “ChronoLapse by Keeyai” with Time Between Captures set to 1. At the public Chronolapse repository’s inspected commit, its screenshot routine copies a desktop bitmap with `wx.ScreenDC` and `MemoryDC.Blit`; it does not query or explicitly draw the operating-system cursor.

Directly observedF4431: ChronoLapse capture UI
Time Between Captures: 1
Blender brush circles remain visible
Source-code inspectionChronolapse commit
6e013ef4b19bd16d1b46f36b024f520900aebc6a
ScreenDC → MemoryDC.Blit
Strong inference≈1 screenshot per work second
15 screenshots per output second
≈74 minutes represented by 4:55
Not recoverableOS pointer path · clicks · keys
tablet pressure · movement between captures
exact Chronolapse binary/commit used

Application-drawn overlays are different evidence: when Blender’s brush or resize circle appears, the ledger records every recoverable sampled frame with its screen-space center and radius. An inferred operation is never relabelled as a recorded click.

Original-resolution sequential audit

Every decoded frame is represented in the UI ledger.

Each range auditor inspected every assigned original PNG sequentially at 1280×720 for the main recording and 720×720 for the companion. Contact sheets and OCR navigated the corpus but did not substitute for native-frame inspection. The final strict compiler rejected missing or overlapping frames, invalid PTS, unmeasured visible cursors, and visible brush samples without radii.

Main coverage verified4,432 original-resolution PNGs
F0000 through F4431
0 gaps · 0 overlaps
Companion coverage verified119 original-resolution PNGs
C0000 through C0118
0 gaps · 0 overlaps
Full forensic layerui-events.json + nativeAuditReceipts
control + before/after + mode + object
per-frame visible cursor samples
Teaching layer33 confidence-labelled macro transitions
12 limited credited source excerpts
raw video and exhaustive frames stay private

Measured review throughput

Automation found attention targets; people still verified the pixels.

The navigation pass used FFprobe timestamps, exact PNG hashes, OpenCV luma and edge deltas, 320-pixel SSIM, perceptual hashes, contact sheets, and crop-specific OCR. None of those outputs counted as visual confirmation. The acceptance receipts were created only after an auditor opened every cited native PNG at original detail and compared its UI state with adjacent frames.

Ordinary 25-frame batch132.1 seconds measured wall time
11.4 frames per minute
review plus forensic note
Ambiguous 25-frame batch317.0 seconds measured wall time
4.7 frames per minute
undo/redo plus reference-board analysis
Dense-menu 25-frame batch525.4 seconds measured wall time
2.9 frames per minute
small text plus state disambiguation
Measurement boundaryOnly F2675 through F2749 were stopwatch-instrumented
no retrospective full-audit duration is claimed

Caught claim errors

Full resolution changed the report.

  1. The glass mask dialog at F3333 is 128×128. The nearby 1024 belongs to the separate V.G. bake panel. Every reconstruction artifact was rebuilt after this correction.
  2. The compositor node at F4010 is named Pixelate, not Dilate.
  3. F4007–F4015 is the Compositor, not the Shader Editor. Exact hourglass execution remains unobserved; later creator replies confirm this reusable group provides final hard pixelization/antialias removal across models.
  4. F4214–F4267 remaps the falling stream’s UVs onto the sand image; it does not paint a vertical stream highlight.
  5. The sand dialog starts 128×128 Linear at F3595; image data later reads 64×64 at F3650. The transition itself is unknown.

Tool execution

The exact Blender and Ucupaint pair was executed in isolation.

Blender 5.1.0 and Ucupaint 2.4.5 are directly visible in the recording. Both official archives were hashed; the add-on was installed into an isolated Blender user environment rather than the global profile.

Blender archiveblender-5.1.0-windows-x64.zip
SHA-256 BC184226962904E3F26C5809BA7EC86BBEB670060825B046930B0C02A30A4EEE
Executed BlenderBlender 5.1.0
build hash adfe2921d5f3
official checksum matched
Ucupaint archiveucupaint-2.4.5.zip
2,098,075 bytes
Archive SHA-25602418d70bd5c7c76d6c74e48e37ad553f493701530a305a0b3667167968dd1c3
Real operators executedwm.y_quick_ypaint_node_setup
wm.y_new_layer · wm.y_new_layer_mask
wm.y_open_existing_data_to_layer
Mask contract verified128×128 · Closest · white · Multiply
fresh reopen retains live Ucupaint stacks

Primary process replay receipt

Twenty-five real `.blend` checkpoints, not a screenshot slideshow.

The replay rebuilds the visible process from default cube through final parenting. Every checkpoint embeds its source-frame range, evidence class, and replay operation. It is an evidence-linked reconstruction—not SquareAnon’s original file, topology, or hidden brush event log.

Ordered scenes25 checkpoint `.blend` files
25 fixed-camera checkpoint renders
default cube → final presentation
Final scene16 mesh objects
732 base vertices
598 base polygons
Real authored data128² MainAtlas · 128² GlassMask
64² SandPaint
three live Ucupaint material stacks
Fresh-reopen verification640×640 RGBA
decoded pixel SHA-256
FF7AEE5453C5FF6D4D2B7192A5E7F00024523F1140D7E8C9B16DEF2F1DDE31BB
Final `.blend` SHA-256E7EF8CE2F9940A4A7EFAE58710D2E06C7D8C4B1B15015E80376B84B8711DEF71
Checkpoint archive SHA-256498D94DE354C17CC2B1B0B294EC556025DEA188A20F12ED4F31B794D0E960C26

Secondary channel-isolation receipt

A controlled ablation proof, not the creator’s file.

Scene22 mesh objects
922 base vertices · 712 base polygons
1,908 evaluated vertices · 1,736 evaluated polygons
4 materials
Textures128×128 painted color
128×128 glass mask
64×64 painted sand
Final image SHA-256793de07a78190f08eabb54581cdaa99dc34e21943b543b6935ed382d5fc24d32
Reopen verification640×640 render
mean absolute pixel delta 0
empty difference bounding box
VariantChanged pixels vs finalMean absolute RGB deltaInterpretation
Emission / unlit0.00%0.0000Reconstruction identity: unlit emission is this reconstruction’s final branch.
Linear filtering11.38%0.2023Localized but visible softening of nearest-filtered bands.
Opaque glass / no mask13.67%10.2435Sand visibility and glass storytelling collapse.
No vertex color34.41%2.0201Broad but relatively subtle value contribution.
Neutral textures35.35%18.3334Largest loss of authored material/form information.
Runtime Lambert35.02%12.8196Live light reshades the already painted surface.
Rotated runtime key35.02%17.6197Asset separation nearly disappears under the moved key.

The reconstruction’s greyscale and two-value aids are diagnostic, not automated beauty passes. The two-value aid reveals a real weakness: much of our dark wood merges with the black background. The guide keeps that failure visible.

Primary sources and reproducibility

Use the source, not our confidence.