[PureSaber]

factor_report could previously compute a historical signal from values that arrived after that session, while sparse observations could compress rolling windows and signal delays.

Changes:

• rebuild late-data signals from rows visible at each daily 23:59 UTC decision time while retaining the single-pass fast path for timely panels
• evaluate rolling dependencies and signal delays on the complete supplied session grid so missing or late sessions remain unavailable
• validate the session-date contract and document the point-in-time semantics
• add regressions for delayed availability, sparse rolling windows, cross-sectional ranks, and signal-delay gaps

Validation:

• python -m ruff check src tests
• python -m ruff format --check src tests
• python -m pytest -q: 182 passed, coverage 88.61%
• existing downstream validation after the same fix: quant-pipeline 27 passed; A-share research workbench 21 passed

Constraint:
When available_at is absent, the report retains its documented descriptive end-of-day assumption. Strict PIT use requires actual availability timestamps and a complete exchange session calendar.

[PureSaber]

独立代码审查：3061b1cc60ff4860d7d358d213c35806f22a57f3

已独立核对历史可得性、完整交易日网格、warmup遮罩、截面信号和延迟信号路径，并读取对应回归测试及182项完整测试结果。晚到输入不能进入历史决策；缺失交易日不再压缩窗口。未发现阻塞问题。23:59UTC为本日级接口明确约定；缺少available_at时仍只能作描述性研究。

此为独立agent代码审查记录；合并仍须通过该提交的全部CI与仓库required checks，不替代或绕过GitHub保护规则。