Estimating remaining time to sustained recovery with repeated daily landmarks within stock drawdowns

I am designing an exploratory study of Egyptian stocks, initially ABUK and FWRY. At the close of each session during a drawdown, the intended prediction is the remaining time until a sustained recovery, conditional on information already available. This is not a request for a stock recommendation. A provisional recovery label is the first close above the maximum of the previous 10 closes, followed by 5 closes above that pre-breakout level. Those future 5 closes label the outcome; they must not enter prediction features. The label is a working choice, not an established definition of a market bottom. Administrative end of observation and missing/unusable follow-up must not be silently counted as recoveries or losses. I want to screen one predictor at a time (e.g. 3-session return or current volume divided by the preceding 20-session mean), then evaluate incremental predictive value jointly. Multiple daily prediction origins belong to one drawdown, multiple drawdowns belong to one stock, and stocks share market shocks. Real preliminary data audit, not model results: stock,observed_dates,provisional_consistent,provisional_single_source,unresolved ABUK,6057,523,4761,773 FWRY,1577,516,450,611 These dates cover 2000-01-12 to 2026-02-04 for ABUK and 2019-08-07 to 2026-02-04 for FWRY. Source overlap is deduplicated by stock and date. Prices that disagree are unresolved rather than averaged. Corporate-action and calendar reconciliation remain unfinished, so these are not counts of us…