ALLWAYS061 — WHAT STAYS — V03
50 seconds /960×768 /30fps. Archival material-action study and held-frame control.

WHAT THIS WORK DOES
A recorded carton-folding task leads to a dark wet front in paper, then to cutting that produces/separates metal curls, then back to the carton task. The 17–19s card identifies a different-material analogy. The 37–45s comparison pairs earlier and later recorded views with source times and asks what would remain after contact stops. It does not answer a recovery question these recordings did not test.

SCOPE / PROTECTED CONDITIONS
This is an archive adaptation, not the prescribed original50-second tabletop experiment with dry paper, cloth, wet paper, shifted weight and weathered paint. Those new experiments were NOT filmed. The three selected responses are folding, darkening following liquid contact, and curl/chip formation/separation. The carton is already prepared/pre-creased; the wetting shot is staged advertising, with unverified production technique/speed; metal cutting begins before our excerpt. These are different samples and conditions. We do not claim elastic recovery, permanent/irreversible deformation, common properties, or measured memory in matter. No simulated physical response or generated sample appears.

An archive source's commercial claim, including its original 'More than paper' text, is not our measurement or endorsement. The endpoint control holds selected clip-end images throughout action passages; 'endpoint' there means excerpt frame position, never relaxed physical state. Its soundtrack is deliberately identical to main for comparison. The shared endplate supplies temporal pairs even in the control, so the control is not entirely devoid of action-order information.

SOUND
Only original supplied source streams are used on their own picture intervals, at unchanged gain. Carton source has no audio track and those passages are silent. Wetting and steel streams are present; their contents have not been auditioned and are not certified synchronous location/contact recordings. Cards/final comparison are digitally silent. No new taps, voices, generic atmosphere, or manufactured material sounds added. The identical sound used over held control frames is an explicit ablation, not a claimed recorded still-event sound. Actual listening/contact-sound acceptance remains pending.

SOURCE / RECOVERY
SOURCE_RECORD.json retains source titles/identifiers, exact transfer URLs, hashes, known dates and rights evidence/unknowns. Steel clip-relative time0–10 maps to original source384.551–394.551s according to the catalog. Other times refer to the exact downloaded full transfers. MATERIAL_ACTION_LEDGER.json identifies before/contact/after, limitations, landmarks and crops; TIMELINE_V03.json gives the edit. COMPARISON_STATES_V03.json maps every reused comparative still. Original selected sources and the candidate audit remain in the kit. The steel catalog's 'color' tag conflicts with inspected monochrome images; source tags were not treated as ground truth. The archive samples were selected by inspecting actual frames.

To rebuild run python3 film61/build.py from the kit parent. Requires Python3, Pillow, numpy, ffmpeg/libx264, ffprobe and DejaVuSans. Geometry is uniformly scaled and padded; no warped shape or reverse-motion recovery is invented. Source rates (including16fps carton footage) are resampled to30fps while retaining supplied transfer duration, with duplicated/dropped frames rather than interpolated material motion. This does not authenticate original historical camera speed. V01 andV02 scripts/outputs preserve prior states. The unused Spring Fever full candidate and second Bounty commercial full candidate are omitted from the kit; their metadata/overview sheets remain. The six smaller cached candidates remain inspectable.

REFERENCE
Matter attribution remains unresolved. Sports Matter was inspected as creator-page text but is a different title and is not silently substituted. No source/reference film superiority claim or imagined viewing is made.

REVIEW / REPAIRS
Actual independent critic records accompany the work. V01 passed a narrow visual three-response operation but its final 'Recorded endpoints' catalog implied more endpoint status than active frames warranted and weakly reframed the opening. V02 replaced it with earlier/later source pairs labeled Selected states/Recovery not tested. V03 corrects the column-heading alignment; pictures, timing and sample evidence are unchanged fromV02. Normal-speed rhythm, real contact-sound function, recovery and the full new tabletop proof remain outside what has been demonstrated. No full tribunal pass is claimed on structural checks alone. Private draft only; no publication is authorized or performed.