[OptimistOtaku]

Moving clips out of source order disables the existing ordered-caption lookup and makes projection scan every clip for every caption. In the committed full-derivation benchmark, a 2,000-clip/6,000-caption shuffled timeline drops from 21.23 ms to 6.90 ms (3.08× faster). This path is used by GPUI caption synchronization after timeline edits and by shared editor loading.

Build a temporary source-sorted mapping index with prefix maximum ends, select overlapping candidates by binary search, then restore their original timeline order before generating caption pieces. That keeps split IDs, output ordering, word timing and source-style inheritance intact. Candidate storage is reused across captions. Already ordered timelines retain the direct slice iterator; fewer than 32 clips or 16 captions, invalid source ranges, non-finite caption bounds and broad candidate spans retain the existing scan. No persistent cache, public API, dependency or project-format changes.

Benchmark

Windows x86_64, Ryzen 5 5600H, Rust 1.88 GNU, repository release profile (opt-level=s, thin LTO). Seven alternating paired samples after two warmups, ten derivations per sample. Complete mapped output and track output are checked against the exact upstream functions before timing. Measurement includes index preparation, projection, override preservation, sorting, allocation and output destruction; excludes I/O, speech recognition and GPUI dispatch.

Clips	Source captions	Clip order	Before	After	Speedup
20	150	Ordered	0.0396 ms	0.0397 ms	1.00×
20	150	Reversed	0.0730 ms	0.0657 ms	1.11×
20	150	Shuffled	0.0611 ms	0.0635 ms	0.96×
60	900	Ordered	0.1753 ms	0.2073 ms	0.85×
60	900	Reversed	0.1852 ms	0.1635 ms	1.13×
60	900	Shuffled	0.2286 ms	0.2197 ms	1.04×
500	1,500	Ordered	2.2679 ms	2.3341 ms	0.97×
500	1,500	Reversed	1.8711 ms	0.9882 ms	1.89×
500	1,500	Shuffled	1.5264 ms	0.9320 ms	1.64×
2,000	6,000	Ordered	7.7958 ms	7.1419 ms	1.09×
2,000	6,000	Reversed	22.2237 ms	7.6476 ms	2.91×
2,000	6,000	Shuffled	21.2325 ms	6.9007 ms	3.08×

The benefit is concentrated in large reordered timelines. The 60-clip ordered control was about 18% slower in this run; on a second complete run it was neutral, with the other ordered controls 0–4% slower. That rerun retained 1.67–1.69× gains at 500 reordered clips and 3.00–3.08× at 2,000. Absolute timings varied on this workstation. This is a caption derivation benchmark, not an overall editor-FPS multiplier.

cargo test --locked -p cap-project --release --lib benchmark_unordered_caption_projection -- --ignored --nocapture

Validation

• cargo fmt --all
• cargo check --locked -p cap-project
• cargo test --locked -p cap-project --lib --tests: 203 library tests and four integration tests passed; five benchmarks ignored by default, new benchmark run separately.
• cargo clippy --locked -p cap-project --all-targets -- -D warnings
• git diff --check

Tests compare complete output and float bits across threshold sizes, reversed/shuffled/repeated clips, nested intervals and a long clip overlapping many short clips, shared edges, wordless captions, unsorted/capped/non-finite words, invalid clip ranges, take offsets, speed changes, transitions, overlapping fullscreen holds, split IDs and caption styles. Both baseline functions match upstream f7d351611 exactly. Native GPUI interaction was not exercised locally.

This branch is based directly on main, independently of #3601; the two branches combine without textual conflicts.

Confidence Score: 5/5

This PR appears safe to merge; no actionable issues were found.

Summary

Speeds up caption projection for large, reordered timelines with a temporary source-sorted index.

• Caption projection finds reordered clip overlaps with a temporary index.

Reviewed statically; tests and benchmarks were not run during this review.

_{Reviews (1) · Last reviewed commit: "perf(project): index caption mappings ac..." · Reviewed by Greptile}

[vercel]

@OptimistOtaku is attempting to deploy a commit to the Cap Software Inc Team on Vercel.

A member of the Team first needs to authorize it.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has insufficient remaining quota. Try requesting a review at a lower effort level.

An organization owner can choose the repository organization as the billing source for reviews requested by members with a paid Copilot license. Learn about Copilot code review billing.