Dormancy window sweep (60 / 120 / 180 / 252 days)

Follow-up to the dormant-awakening entry above: does extending the dormancy precondition beyond 3 months (60d) — to 6mo / 9mo / 1yr — produce a stronger signal? Intuition would say yes: more time quiet = more pent-up energy.

Code: research/dormant_awakening.py --window {120,180,252} Reports: research/dormant_awakening_{120,180,252}d_report.json

Sample-size collapse:

True 252-day dormancy at the $5M ADV / $5 price gate is extremely rare in the cached US universe — only 15 events in 15 years.

Edge degrades monotonically as dormancy lengthens:

The 252-day cell is too small (n=15) to lean on individually, but the monotone trend across 60 → 120 → 180 → 252 is hard to miss. The intuition was wrong: longer-required dormancy doesn't compress more spring; it filters into a population with structural reasons to be dormant (small float, distress, no analyst coverage, dying franchise) where the wake-up is more often a one-off catalyst that fades than the start of continuation.

T+1 confirmation pattern survives at every window:

The day-2-confirmation sub-cohort lifts the edge to ~+5-6 %/20d at hit-rate 0.79+ everywhere it has meaningful n. So the confirmation pattern is window-independent, while the dormancy precondition is mildly destructive at length.

Verdict: Keep dormancy short (60d at most) or drop the dormancy precondition altogether. The burst_continuation cohort already captures the cleaner version of "first-day spike + day-2 confirm" without any dormancy filter (n=720, edge +9 %/20d, hit 0.70). The dormancy-conditioned variant adds noise rather than signal as the window lengthens.

Meta-finding extended: Across burst_continuation, dormant_awakening at 60/120/180/252d, and failed_breakdown, the day-2 follow-through is the only selector that produces a clean +5-9 %/20d positive forward drift. Prior-day conditioning (dormancy, prior crash, prior thrust) modulates which population fires the trigger but does not itself create the alpha. The pattern is: the second day tells you whether the first day was real.