Research proof index

Running the inner unifForward + roImpl simulator against a source computation with an nmaHashQueryBound Q can grow the internal queryLog by at most Q. Each source Sum.inr step consumes one unit of the nmaHashQueryBound budget, while roImpl appends to queryLog only on a cache miss, hence at most once per such step.

If two successful contextual forks select the same fork index, their forgery targets agree.

Completing an occurrence resamples the focused answer as a fresh query i: any event on that answer marginalizes the resampled suffix away.

Fixed-index observed success squares under two independent completions of the selected occurrence context.