# YellowMellow burst audit

## Outcome

The original 34 folders are shoot/action clusters, not consistently true shutter bursts. Playback review exposed abrupt movement that the initial still-frame contact-sheet review did not catch. Twenty-two folders contain internal capture pauses over 0.25 seconds; some pauses reach 1.00 second. Encoding across those pauses at a constant frame rate created the random, choppy motion reported by the user.

The corrected plan splits at capture gaps over 0.25 seconds, keeps true bursts with at least five photographs, and applies the three-second minimum to the finished repeated clip rather than the unique motion. It produces 71 review clips from 1,042 photographs and holds 30 one-to-four-frame fragments (80 photographs) for review.

## What was checked

- 34 burst folders and 1,122 burst JPEGs.
- Every folder's actual frame count matches the count encoded in its folder name.
- Every burst has a contiguous numeric filename range with no missing or duplicated frame number.
- Every internal capture-time gap is 1.00 second or less.
- Five evenly distributed frames from every original folder were visually compared for scene/action continuity. This was insufficient: it confirmed subject/scene consistency but did not prove continuous motion between adjacent frames.
- The one JPEG metadata outlier and its adjacent frames were inspected separately.
- All 34 first-pass H.264 previews and all 34 first-pass CR3-developed ProRes masters were fully decoded and frame-counted after rendering. Those checks proved file integrity, not correct motion grouping.
- A representative frame from every ProRes master was visually checked for orientation, framing, exposure, color, and failed RAW development.

## Exception

`Burst19_7O9A6459_23f/7O9A6477.jpg` is a 1620x1080 JPEG with stripped capture/orientation metadata, unlike the other 2,048x1,365 JPEGs. Its matching `7O9A6477.CR3` confirms the expected `Rotate 270 CW` orientation and capture time. It belongs to the burst visually. The renderer normalizes the sequence orientation without changing the source JPEG.

## Corrected three-second rule

One endpoint-deduplicated forward/reverse cycle is:

`cycle_seconds = (2 * unique_frames - 2) / fps`

The source-frame minimum and finished-duration minimum are separate rules:

- Keep five or more photographs as a true burst by default.
- Put one-to-four-frame fragments in review.
- Preserve the detected 6, 8, or 10 fps cadence.
- Repeat whole cycles until the finished clip reaches at least three seconds.
- Never combine separate shutter bursts merely to reach three seconds.

For example, five photographs at 10 fps create a 0.8-second cycle; four whole cycles create a smooth 3.2-second review clip.

## Rejected first-pass H.264 preview output

- H.264 High profile
- 1366x2048 portrait, preserving the 2:3 composition with minimal padding
- yuv420p, Rec.709
- Per-burst cadence of 6, 8, or 10 fps inferred from capture timing
- Seamless forward/reverse order with duplicated turnaround frames removed
- Whole cycles repeated to at least 12 seconds for easy trimming in DaVinci Resolve

These clips are retained for comparison but must not be used for the final edit because their grouping is wrong.

## Rejected first-pass CR3-developed editing masters

- 34 ProRes 422 HQ `.mov` files developed from the matching CR3 sensor data
- 2560x3840 portrait, preserving the complete 2:3 composition for reframing in DaVinci Resolve
- yuv422p10le with constant per-clip rates of 6, 8, 9, or 10 fps
- 12.00 to 26.20 seconds per clip; 18.49 GiB total
- 34 manifest matches, 34 successful metadata checks, and 34 complete FFmpeg decodes with zero errors
- Contact-sheet review found all clips upright and consistently framed, with no black frames or obvious RAW-development failures

The files are technically intact but rejected because the source grouping produces discontinuous motion. Do not use them in DaVinci.

## Corrected true-burst H.264 review batch

- Split threshold: a new true burst begins after any capture gap over 0.25 seconds
- Minimum kept source frames: five
- Finished duration: at least three seconds by repeating whole ping-pong cycles
- 71 kept true bursts containing 1,042 photographs
- 30 review fragments containing 80 photographs
- 71 H.264 previews, 3.00 to 17.67 seconds
- All 71 previews passed metadata checks and full FFmpeg decode

The corrected previews are in `YellowMellow_TrueBursts_Preview_H264`. Confirm their motion before starting a replacement CR3-to-ProRes render.

## Motion-length review order

The corrected clips are also indexed non-destructively in `YellowMellow_TrueBursts_Organized`:

- `01_Long_Full_Motion_3sPlus`: 19 clips with at least 3.0 seconds in one natural forward/back cycle
- `02_Medium_Motion_1.5to2.99s`: 21 clips
- `03_Short_Bounce_Under1.5s`: 31 clips

Each tier is numbered longest-first. These are sorting tiers only; separate shutter bursts remain separate and are never merged to manufacture longer motion.