AI Handoff Continuity Benchmark v1.0 pilot
An open, reproducible pilot for testing whether an AI model recovers continuation-critical state from a conversation transcript, compressed memory, or structured handoff.
Pilot result across OpenAI gpt-5.6-terra and Anthropic claude-sonnet-5:
- Structured handoff: 79.45
- Conversation transcript: 76.67
- Compressed memory: 45.00
This is a small authored pilot, not a model leaderboard. One run per system does not estimate variance, the conditions retain different amounts of detail, and schema adherence affected scores.
Inspect and reproduce:
- Method and limitations: https://handover.sh/benchmark
- Dataset and runner: https://github.com/44-pixels/handover-mcp/tree/dd62884/benchmark/v1
- Raw responses: https://github.com/44-pixels/handover-mcp/tree/dd62884/benchmark/v1/results/2026-08-04/raw
- CFF: https://handover.sh/benchmark/v1/CITATION.cff
- BibTeX: https://handover.sh/benchmark/v1/citation.bib
- CSV: https://handover.sh/benchmark/v1/results/2026-08-04/summary.csv
Dataset license: CC BY 4.0. Runner and repository code: MIT.