* The confidence which determines when a run can be stopped is now calculated not only wrt to the researcher who is the least confident but takes the confidence of all researchers into account.
* `calc-confidence` has been computationally optimized: redundancies have been reduced
* The variable `avg-neighbor-signal` is now turtle-owned and has been documented
* New globals: `g-confidence` and `g-depressed-confidence` have been added and documented.
* The hidden variable `confidence-cutoff` has been adjusted for the new way confidence is calculated
* The global `max-confidence` has been removed as it is not needed anymore
* The `average-confidence` reporter has been adjusted to report the global confidence `g-confidence`

* They seem to better fit in there.
* removing initialization of current-theory-info from setup b/c it's superfluous

* If the number of pulls is smaller than 100 an exact reporter for the binomial distribution is used instead of the normal approximation.
* The range for the pull slider has been extended, which is reflected in the documentation.