Deferred

Parked on purpose. Not forgotten.

Item	Notes
Delisted symbols / point-in-time universes	Needed to remove survivorship bias in cross-sectional strategies. Until then, cross-sectional runs carry a warning.
Walk-forward	Planned for M10. How it relates to train / validation / test is decided when M10 starts. Its report adds a heatmap of the out-of-sample score per window (see Results).
Paper extraction	Designed on Paper extraction; the sbt2-papers distribution exists but is empty, and the pipeline is in no milestone. Once it depends on PDF parsers, CI must install them only for its own tests, since they bring torch.
Multi-strategy accounts	Several strategies sharing one account.
More asset classes and data sources	Added as plug-ins; see Asset classes and Data sources.
systemd in CI	The CI image has no systemd user session or session bus, so the systemd tests run only locally (see Quality). Running them in CI needs an image that boots systemd and an agent that starts containers with a writable cgroup (podman run --systemd=always).

Open verification items

• Does chunk_size streaming handle OrderBookDelta + TradeTick correctly, including F_LAST batching across chunk boundaries? Planned as the first spike: stream a week of L2 for one symbol and measure peak memory. Check the default memory.per_run (4G) against it and pick the chunk sizes.
• Source-specific open items live on each source's page, e.g. Source Bybit.
• Move the nautilus pin to 2.0.0rc6 on PyPI once it's released (currently pinned to the nightly wheel 2.0.0rc6.dev20260930, see Quality). Deadline: the index keeps nightlies for only 30 publication dates, so this must happen, or the pin must move to a newer nightly, by early November 2026. The nautilus-bump job opens a pull request for each newer nightly (see Quality). Run the nautilus characterization tests and the golden tests first (see Quality). The pin was chosen because 2.0.0rc5 silently streams zero funding updates from the catalog, while the rc6 develop wheels streamed all 3 in the verification run (the nightly passes the same characterization tests) and settled each at −5.00 USDT on a 1 BTC long at a 50,000 mark.