# Preregistration amendment 01 — discovery treatment activation (2026-10-31)

The first **development-only** pilot, seeds 11–13, never discovered the latent good. The all-zero discovery indicator makes H7's conditional-on-discovery recovery estimand unidentifiable in those cells. This problem was observed **after** freezing `H1-H7_2026-10-31.md`; the original file is unchanged.

For *subsequent development and holdout* experiments, reserve one funded exploratory attempt every four periods after new-good onset, under exactly the same capacity constraint in each regime. On all other periods, agents independently learn whether to fund an attempt. The binary discovery draws remain pre-sampled common random numbers; no realized demand or future shocks are revealed. Such exploration interventions represent a minimal research/search budget, not pure spontaneous entrepreneurship, and must be reported separately from the unguided pilot. Keep the no-mandate run as a sensitivity analysis. The mandated schedule can mechanically advantage those cases with previously too little exploration, so comparisons must be conducted on matched paths. This amendment is **not** independent preregistration and must be disclosed in publications.