Pre-registration, amendments and deviations

The discipline behind the agreement and accuracy study, recorded so that a reader can check the order in which decisions were made rather than take it on trust.

What the record contains

A pre-registration fixing the hypotheses, the admissibility rules and the analysis before any data existed, followed by two amendments. Each of the three documents was committed before the data it governs was collected, so the commit history carries the ordering rather than the prose asserting it.

Alongside them sits a deviations log recording every departure from the registered plan together with its timing, including the departures that weaken a result. One admissibility rule was fixed ahead of the data specifically so that a convenient post-hoc boundary could not be drawn later, and an adversarial control was included to detect the case the rule was meant to exclude.

Why it is listed separately

A pre-registration is only worth anything if it can be read against the result it governs, and if the departures are visible. Listing the record on its own makes the ordering checkable without reading the study first: the question “was this decided before or after the data” has an answer in the repository, not in the write-up’s own account of itself.

The same file set carries the boundary that matters most for reading the study, which is that a reproduced number is not an independently validated one. Recomputation from committed traces checks arithmetic; it does not check the instrumentation that produced those traces.

Where to read it

evaltrust at v0.2, documents PREREGISTRATION.md, PREREGISTRATION-AMENDMENT-01.md, PREREGISTRATION-AMENDMENT-02.md and DEVIATIONS.md.