0
Fork 0
mirror of https://github.com/zeronsh/comet.git synced 2026-11-16 18:57:19 +00:00
comet/scripts/fixtures
History 2026-09-08 17:45:22 -03:00
..
markdown-preview feat(markdown): add a native Mermaid rendering backend 2026-09-08 17:45:22 -03:00
README.md Fix streaming selection, runway expansion, and macOS redraw recovery (#265) 2026-09-06 13:02:46 -07:00
resource-stream.jsonl Document release resource measurements and reproducible stream workloads 2026-09-05 04:59:18 +00:00
runway-short-stream.jsonl Fix runway scroll jumps and stale holds after background streaming (#257) 2026-09-05 18:45:21 +00:00
transcript-selection-stream.jsonl Fix streaming selection, runway expansion, and macOS redraw recovery (#265) 2026-09-06 13:02:46 -07:00

resource-stream.jsonl contains sanitized deltas from a successful Haiku profiling response: an 80-section Rust ownership tutorial with code fences. Only text/reasoning deltas and the successful completion marker are retained; account, session, timestamp and usage metadata are omitted.

The assistant text is 51,769 bytes, SHA-256 c1809f92c26c682c6f035478f7ca63980e25fb5e97ea7746dcb41ecd7dbb0a25. Reasoning contributes another 853 bytes. The production transcript's part separators bring the combined short reply to 52,624 bytes.

Use ZERON_REPLAY_REPEAT=10 ZERON_REPLAY_DELAY_MS=8 for the synthetic long workload. See the profiling report for build settings, commands and measured results.

runway-short-stream.jsonl is a synthetic short reply in 24-character chunks. It keeps the own-send runway active through completion. Use a 400 ms replay delay for the runway performance comparison.

transcript-selection-stream.jsonl is a synthetic short reply in six-character chunks. A 1200 ms replay delay leaves time to expand/collapse a long prompt and select live text. See the native regression recordings.