research-loop

# Research Loop Runs survey-architect, feedback-synthesizer, and research-reporter as a loop that behaves like an actual research team: someone can say "that sample's too small, we need more data" and send it back, rather than quietly writing up a shaky number. Each of the three skills works standalone. This skill exists for the handoffs between them and the file structure that lets a later skill (or a later *study*) pick up where an earlier one left off. ----- ## The loop at a glance ``` Entry detection ↓ survey-architect (intake → instrument → sample size → deploy-ready spec) ↓ [ pause: wait for data collection ] ↓ feedback-synthesizer (score + CI + significance + theme + severity) ↑___________kick back if n insufficient for stated goal_______| ↓ research-reporter (audience-formatted output) ``` Max loop-backs: 2 per study. If a second kickback doesn't resolve it, surface the trade-off to the researcher rather than looping a third time. ----- ## Entry detection **Run `scripts/routing.py`'s `detect_entry_point(...)`** rather than eyeballing the table below — pass whether a learning goal, raw data, an existing study folder, and a synthesis file are present, and use its return value. The table is here to show the reasoning, not to be hand-matched each time: | What's pasted / asked | Entry point | |---|---| | A research question, no data yet ("I need to know if X is usable") | survey-architect | | Raw survey results / CSV / feedback pile, no study...