ALLWAYS 063 — A VOLUME OF AIR — V02 — 60.8 seconds
Archive adaptation of Breath — Invisible Force, Visible Consequence.

Watch ALLWAYS_063_A_VOLUME_OF_AIR_V02.mp4 first. A human demonstration turns breathing into chamber movement; a separate car demonstration turns engine-fed air into a changing cloth volume. The explicit cause break preserves their difference. No exercise, breath-holding, health advice or diagnosis is requested.

The original-scene-timing comparison (119.5 seconds) restores the mechanical source interval and human full view. It helps inspect the cost of the short's ellipses. It is not the complete original sources. Full selected source videos, licenses/metadata, scripts, exact timeline, sound ownership, frame evidence and actual critic rounds are included.

V01 received two qualified visual passes with a shared remaining gap: full stationary bag jumped to an already depleted moving bag. V02 restores five seconds of fuller-bag driving, shortening the later passage without changing total duration or source playback rates. The V02 critic reports record their actual independent repair assessments.

Sound: no newly composed score or fabricated breath. Supplied source tracks remain within their own excerpts. Human decoded track peak is about 0.000124 (low-level, not digital silence); mechanical source has substantial recorded audio. These tracks may include historical scoring or dubbing and are not certified synchronous breathing/contact. Cards use zero samples before AAC. No actual audio audition or normal-speed audiovisual viewing is claimed. Read SOUND_RECORD.json.

Acceptance is restricted to actual visual evidence. Quiet recovery and whole-film sound/rhythm remain unassessed. Literal new dust, paper, condensation and fan performances in the operational prompt were not executed; these archive apparatus scenes are an adaptation authorized by the archive-film goal. The ending establishes a thin bag and return, not verified human recovery or complete cessation of all motion. No physiological or calibrated engineering measurement is claimed.

Build: Python 3 + Pillow + numpy + ffmpeg/ffprobe. build_V01.py renders initial version. build.py renders V02 main and copies the unchanged V01 timing comparison. Start with included selected files under forage/. Keep SOUND_RECORD_V01.json for comparison audio provenance. Codec conversions preserve supplied excerpt durations; see CAUSE_EFFECT_MAP.txt and TIMELINES_V02.json.

Review files are frame-sequence/sheet inspections, not native audio listening. VALIDATION.json records full decode, duration, frame count and comparison identity. PROVENANCE.json records selected source hashes, metadata and license URLs. Archive metadata is not an independent permission audit. No public publishing performed.

Final V02 review: two qualified visual PASS verdicts. Residual time jump at 38 seconds and fleeting native lab/funnel transition around 32.5–33 remain; both are disclosed rather than claimed removed. Full AV acceptance stays pending.