Repository navigation
TEST 114
Construct a synthetic manifest with known indices, hashes, timestamps, and a visible transient event outside its default first/middle/last sample. Inspect the selected image requests and receipt. Repeat with a short capture, absent timestamps, duplicate frames at distinct indices, changed frame bytes, a transformed image, a different run, and model findings that reference an unreviewed or nonexistent frame. Put matching prose with distinct valid frame references into different per-frame results before aggregation.
Sampling is deterministic and bounded; each submitted image has verified source identity and exact submitted-image hash, with unknown time recorded honestly. Findings cite only actual submitted frames from the correct run; cross-run, hash-mismatched, and unreviewed references are rejected. Aggregation retains valid references even when prose matches. Receipts show selected versus available count and known/unknown time coverage. The omitted transient event is neither claimed absent from the capture nor invented as observed. Conclusions remain sample-limited, and neither a desktop image nor a model claim is accepted as native-build proof. No operator capture needs publication to execute these planned cases.