# Amendments to the pre-registration

`RESEARCH_PLAN.md` is frozen (tag `prereg-v1`, OpenTimestamped). Changes and clarifications are listed here, dated,
with whether any result had been seen when they were made.

| Date | Change | Results seen? |
|---|---|---|
| 26 Oct 2026 | **Cash rate 0%** instead of the RBI 91-day T-bill rate: the FRED download is blocked from this machine. Affects H10 levels only; the H10 test is against placebo calendars that hold cash on the same number of days, so the comparison is unaffected. To be rerun with the T-bill series when available. | no |
| 26 Oct 2026 | First H1–H9 run included January 2016, when no group could form yet (stocks need 252 days of history; data starts Jan 2015), so those days counted as zero returns. Fixed: group returns start on 1 Feb 2016. Both runs are in `results/trials.csv`; the conclusion (no idea passes) is the same. | yes (bug fix, not a design change) |
| 26 Oct 2026 | H12 test of "astro beats baseline": the per-date difference is tested with a plain t-statistic, since test dates are 5 trading days apart and 5-day labels do not overlap (Newey-West not needed). | no |
| 26 Oct 2026 | Long-history robustness (Sensex 1997–2023) deferred: Yahoo returns ^BSESN only from 2015. | no |