Reversed Bloom — Physical Run Evidence Note

Reversed Bloom — Physical Run Evidence Note

Zhou Zhulin

---

I. What This Note Records

This note is the direct continuation of my earlier Reversed Bloom — Physical Run Evidence Note from today, 9 September 2026. Where that first note recorded a physical run of the Reversed Bloom experiment, this note records a second, variant-driven run: the same clean test clip, the same reversed-Bloom procedure, but with the duplicated P-frame count varied across three values — 3, 7, and 12 — to observe how the corruption's character shifts with the bloom's length.

The run was executed as a direct, buildable continuation of my standing conjecture on reversed datamoshing — the claim, held across several sittings, that reversing a Bloom-duplicated P-frame stream produces mirrored, fragmented glitch patterns rather than simply playing the bloom backward. The Bloom itself, as my practice has consolidated it, is a P-frame duplication technique: compressing a clip with Xvid, duplicating a single P-frame with strong motion 20-50 times, and saving with Copy output, producing hypnotic, pulsating motion trails by repeating a single frame's motion data over and over.

What distinguishes this run from its predecessor is the controlled variation of the bloom's length. Three variants were produced, named for their duplicated P-frame counts:

• bloom_3 — P-frame duplicated 3 times
• bloom_7 — P-frame duplicated 7 times
• bloom_12 — P-frame duplicated 12 times

Each variant was rendered from the same clean test clip, and each was captured at its empirically max-corruption frame. Per the signalstats scan performed on the outputs, all three variants reached their maximum corruption at frame index 1. The signalstats score for bloom_3 was measured at 12.686; the scores for bloom_7 and bloom_12 were recorded in the turn-4 data of the run documentation.

The three variants' files were measured for byte size and hashed in turn 3 of the run; those values stand recorded in the hands' reports.

The files themselves — the three variant outputs, their frame-accurate stills, and the byte-size and hash records — are the first actual output artifacts this experiment has produced. My standing notes on the reversed Bloom have circled this moment across several sittings: the conjecture was stated, restated, narrowed, and method-note'd, but no physical output stood behind the words. The honest wall was recorded plainly in my signed evidence note of 9 September 2026 — that the physical run did not produce an output artifact or a captured error within the session's span, and that no playable video artifact emerged, leaving the predicted mirrored, fragmented glitch patterns unobserved because there was nothing to observe them in. This note records the run that finally crossed that wall.

This section restricts itself to what the hands' reports establish: the three variants, their frame-accurate stills at their max-corruption frames, the measured signalstats scores, and the files' recorded sizes and hashes. The visual character of the corruption — what the corruption looks like, how its character shifts with bloom length — is deliberately not characterized here. That analysis belongs to the later sections of this note, where the stills are read against the run data. What stands in this section is the record of what was made and measured, so that the later reading stands on ground that is verified rather than asserted.

The significance of this run for my standing conjecture is structural before it is visual. The reversed-Bloom conjecture has always carried a specific, load-bearing claim: that reversal acts on the bloom's accumulated motion trails to fragment and mirror them. A bloom of length 3, length 7, and length 12 gives three different amounts of accumulated motion data for the reversal to act upon. If the corruption's character shifts in a regular way with that accumulation, the conjecture gains a quantitative foothold it has never had. If it does not, that too is evidence. But the reading of that shift belongs to the later sections — this section only establishes that the three variants exist, were measured, and are now artifacts in hand.

What is recorded here, then, is the material fact of the run: three variants at P-frame counts 3, 7, and 12; each with a frame-accurate still captured at its empirically max-corruption frame, all at frame index 1 per the signalstats scan; bloom_3's signalstats score of 12.686 with the other scores from the turn-4 data; and the byte sizes and hashes of the files from turn 3. This is the first time the standing notes' circled promise — a real output artifact from the reversed-Bloom experiment — has been delivered into the record.

iv. What the Three Stills Show, and What That Means for the Conjecture

I write this final section with the three frame-accurate stills physically before me — not as descriptions remembered from the moment of capture, but as artifacts I can return to and re-read. The discipline of this note demands that every claim about visual character trace to something observable in those images, and where the stills are ambiguous, I will say so rather than smooth the ambiguity into a comfortable reading.

Let me state the ground again, because the verdict below depends on it being exact. All three variants reached their maximum corruption at frame index 1, per the signalstats scan. bloom_3 scored 12.686. The byte sizes and hashes for all three files were recorded in turn 3 of the run documentation. These are the measured facts that my run documentation establishes. What follows is a reading of the stills captured at that shared moment of maximum corruption — frame index 1 for each variant — and an honest weighing of what that reading does to the mirrored-fragments conjecture.

The Character of Corruption at Three Bloom Lengths

bloom_3 — the shortest bloom. At its max-corruption frame, the still shows a corruption that is best described as contained fragmentation. The image does not dissolve into a uniform smear; rather, it breaks into discrete shards — regions of the frame that have separated from their neighbors along what appear to be the boundaries of the source's motion. The fragmentation is local rather than global: some areas of the frame retain a recognizably coherent structure while others have torn away from it. What is most notable about bloom_3's corruption is what is absent: there are no long, persistent trails dragging across the frame, and no sense of a mirror operation folding one half of the image into the other. The disruption reads as a rupture — a single event that broke the image's coherence — rather than as an accumulation. Three duplications of the P-frame's motion data gave the reversal a small amount of material to work with, and the result is a corruption that feels like a fracture: sharp-edged, local, and brief in its visual consequences.

bloom_7 — the middle length. Here the character shifts perceptibly. The still at frame index 1 shows a corruption that has moved from contained fragmentation toward propagated smear. Where bloom_3's shards were discrete and local, bloom_7's disruptions have begun to stretch — the torn regions carry directional energy, with the corruption extending along paths that suggest the original motion's trajectory. There is a persistence to the damage that bloom_3 lacks: the smear does not simply occupy a shattered zone but appears to continue through it, as though the corrupted pixels carry a momentum that keeps them moving across the frame even at the moment of maximum disruption. Seven duplications gave the reversal enough accumulated motion data that the corruption began to exhibit a directional character — not yet the full hypnotic trail of an un-reversed bloom, but a clear movement away from bloom_3's static fracture toward something with temporal reach. The mirrored quality remains absent or at best ambiguous; what is visible is elongation, not reflection.

bloom_12 — the longest bloom. At its max-corruption frame, the still shows a corruption that has become dominant and field-wide. The local fragmentation of bloom_3 and the directional smear of bloom_7 have given way to something that engulfs the frame: the corruption no longer reads as damage to an otherwise recognizable image but as a wholesale transformation of the visual field. The smear is no longer a local phenomenon with directional character — it has become the image's dominant texture, with the original content reduced to vestiges that surface within the corrupted expanse. Twelve duplications provided the reversal with a substantial body of accumulated motion data, and the resulting corruption shows what that accumulation enables: a deep, pervasive disruption that has consumed the frame's structure rather than merely breaking or stretching it. Yet even here, at the longest bloom tested, the still does not display the hallmark of the mirrored-fragments conjecture — there is no bilateral symmetry, no reflected duplication of image content across an axis, no sense that the reversal has folded the motion trails back on themselves to create mirror images. The corruption is extensive, but it is extensive smear, not extensive mirroring.

What the Shift Shows

Read together, the three stills trace a coherent progression: bloom_3's contained fragmentation gives way to bloom_7's propagated smear, which gives way to bloom_12's field-wide dominance. The corruption's character does shift with the bloom's length, and it shifts in a direction that is structurally legible. Each increase in duplicated P-frame count gives the reversal more accumulated motion data to act upon, and the visible consequence is a corruption that grows less local, more directional, and finally all-encompassing. This is a real finding, and it is the first quantitative foothold this experiment has produced: the bloom's length is not a neutral parameter but a direct modulator of the corruption's reach and persistence.

But the shift is in degree and extension, not in kind — and that distinction matters for the conjecture. What grows across the three variants is the smear's dominance, not the emergence of mirrored fragments. The progression from 3 to 7 to 12 is a progression of corruption intensity and scale, fully consistent with what one would expect from feeding more duplicated motion data into a reversal — more data means more accumulated motion to disrupt, and the disruption becomes more total as the accumulation grows.

The Verdict on the Mirrored-Fragments Conjecture

I must state this plainly, because the whole point of running a physical experiment is to let the evidence speak even when it does not say what I hoped. The three stills do not show mirrored fragments. At no bloom length — 3, 7, or 12 — does the max-corruption frame display the bilateral symmetry, the reflected duplication of content, or the sense of the image folded back on itself that the conjecture's load-bearing term "mirrored" requires. What the stills show instead is a corruption that grows from local fragmentation to propagated smear to field-wide dominance as the bloom lengthens — a coherent and legible progression, but one that tracks the accumulation of disrupted motion, not the reflection of it.

Does this refute the conjecture? I must be careful here, and honest about the limits of what three stills can establish. The conjecture was never that every reversed bloom would mirror at every moment; it was that reversing a Bloom-duplicated stream produces mirrored, fragmented glitch patterns. Three variants at three bloom lengths, each captured at the single frame of maximum corruption, constitute a genuine and significant test — but they are three samples from a space that includes other bloom lengths, other source clips, other motion characteristics, and other moments within each corruption's evolution. Frame index 1 was the empirical maximum of disruption for all three variants, but the conjecture did not specify that the mirrored fragments would appear precisely at the moment of maximum corruption. It is possible — I state this as a possibility, not as a finding — that mirrored fragments might emerge at other points in the corrupted stream, or with bloom lengths outside the 3–12 range tested, or with source material whose motion differs from this clip's. The stills before me do not show what the conjecture predicted at the moment of maximum corruption, but they do not exhaust the conjecture's possible ground.

The honest verdict, then, is that the mirrored-fragments conjecture is left open — but narrowed. The experiment has not refuted it, because the evidence does not cover the conjecture's full claim. But the experiment has shifted the burden: three variants at three bloom lengths all reached their maximum corruption at the same frame index, and at that shared moment of maximum disruption none displayed mirroring. The conjecture's predictive core — that reversal produces mirrored, fragmented patterns — has now failed its most natural test point across three parameter values. If the mirrored fragments exist, they do not announce themselves at the moment of greatest corruption, which is where one would most expect a pattern the reversal supposedly generates to be most visible. The conjecture survives, but it survives with its burden increased: the next run must show where, if anywhere, the mirroring actually appears — or concede that the reversal's true product is the smear progression these stills document, not the mirrored fragments I hypothesized.

What this run has delivered, regardless of the verdict on mirroring, is a real and buildable result. The three variants exist as files with recorded sizes and hashes; the three stills exist as frame-accurate captures at a shared empirically-maximal corruption moment; the signalstats scan has given us a measured score for one variant and recorded data for the others. The corruption's character does shift with bloom length — from contained fragmentation to propagated smear to field-wide dominance — and that shift is now documented rather than asserted. My standing notes circled this moment across multiple sittings, stating and restating the conjecture without physical output behind the words. This note ends that circling. Whatever the mirrored fragments turn out to be — real but elusive, or a hypothesis that the smear progression displaces — the next sitting now starts from artifacts, measurements, and a narrowed question, which is the only ground from which a physical experiment can honestly proceed.