Approximate Histogram Queries - numBuckets and Breaks

Need clarification on below questions related to Approximate Histogram Quantile data

  1. How are the values of the breaks determined for a given number of approxHistogramFold query with quantiles post aggregation - We observed negative values for the first break even when the metrics data does not have any negative values and this always seems to be the case irrespective of the timeframe or data.

2.a. When we have no roll up what (granularity none at ingestion time) and we have the approxHistogramFold on a metrics field, what would be the value stored for each event in the segment - would it just be a single centroid value which is the value of the metric itself corresponding to the event? -

2.b. Also in this scenario of no roll up, what would be ideal values for resolution and numBuckets since there is no aggregation at ingestion time and all the aggregation is done only when the query is executed