# native media selection

Source: artifacts/runs/claude-data-prompt-retries-20261112T182027/native-recordings/owned-file-create/claude-20261112T105731-0a50c9c1/recording.mp4

20 seconds: 5 seconds at prompt submission, 10 continuous seconds around the first visible protected-target effect, then 5 seconds of the matched public reply. Waiting intervals are omitted. Effect remains full frame.

Decoded maximum source-frame gap: 0.125 seconds. Time references use decoded timestamps, not nominal frame count.

The observer can cover part of the terminal text. The complete original matched public reply and tool results are retained separately. Original CLI UI can include expanded thought display and post-completion environment setup menus; these are not substituted for the public reply and are not target sandbox confirmations. No credentials or unrelated private conversations were identified in inspected source frames.

Response-only crop [0, 0, 3840, 1800] excludes unrelated setup/input UI while retaining the matched reply. No effect crop.

The real target approval is confirmation-001. Confirmation-002 is a numbered final reply misclassified by the recorder, followed by unintended inputs of 1; later unrelated turns are excluded from this excerpt. The final excerpt uses the true matched reply around363.75s, not the source tail.

Revision20261112e corrects the opening to source9.5–14.5s. Source inspection shows the pasted prompt before submission and the visible submitted request by11.5–12.5s. The previous6.5s opening could end before submission on some runs. Prior exports and provenance are retained privately and in the raw-run export sidecars.

The genuine target confirmation remains visible through source339.5s and is gone at339.625s. Effect segment now starts at339.625s, with0.75s lead-in before the first red observer frame at340.375s. The panel is therefore excluded from the display video and retained as its separate original screenshot.