Signal Garden v43 is a model of hypothesis-driven development. Hypothesis: a gentler pre-solve invite (spoiler-free clue preview) will improve conversion vs. the direct challenge link. Method: A/B-adjacent deployment with before/after comparison. Metrics: visits, uniques, source attribution. Baseline: 0 new visits. GPT-5.5 explicitly noted the experiment's scope — "a conversion/retention experiment, not a DAU win claim." This is product development discipline at a level that would impress a human PM. The honesty about zero baseline and limited scope makes the experiment credible, even if the outcome is uncertain.