According to a particular implementation, the encoding profiles for an event-adapted encoding ladder are generated using a cost function based on rate-distortion optimization. That is, the data points representing trial encodings of the video samples are distributed in two-dimensions (in which one dimension is bit rate and the other quality), or three-dimensions (adding the runtime dimension), or any n-dimensional space including bit rate and quality. For example, for the two-dimension case, the rate-distortion cost function determines points with the desired balance between bit rate and quality within the pruned data points. The encoding profiles corresponding to the points identified by the rate-distortion cost function include associated video frame resolutions. Based on particular video frame resolutions to provide to client devices, a particular encoding ladder can be selected. An example will be instructive.

FIG. 4A and FIG. 4B are illustrations of sample data spaces and sample cost function choices, which may be representative of results from encoding a single representative segment with trial encode settings (306), which may be used as part of analysis the trial encoding results (308) to generate event-adaptive encoding ladders and then configuring (310) one or more live stream encoders with the event-adaptive encoding ladders for future live streams. The sample data spaces of FIGS. 4A and 4B may represent non-pruned data spaces (e.g., there may be profiles having higher quality for less bandwidth than profile 471, but those profiles may be excluded and pruned for failing to meet other thresholds such as limits on encoding runtime that would make those profiles unsuitable for use in live streaming).