Comments
so users can provide their own clustering for samples (e.g., based on UniFrac distance) and features (not sure this is possible; a way to provide their own phylogeny would be awesome but looks like this is not possible with seaborn.clustermap)

Reference

1. forum xref
2. relevant SO topic.

Current Behavior
Currently supports label annotation of sample metadata information.

Proposed Behavior
Please allow the same for features. And please support annotation with row_colors and col_colors instead of or in addition to label annotation. E.g., see the following plot from the seaborn clustermap docs.

E.g., color by phylum affiliation. Allow level selection from semicolon-delimited feature IDs? This would be useful for taxa but also, e.g., KEGG annotation.

Improvement Description
The boxplots and dot plots are great as a quick report, but are not publication-ready and cannot be edited in illustrator.

The moving pictures tutorial PD boxplots are a good example; particularly when setting the category to "ReportedAntibioticUsage", the columns (n = 2) become very wide and ugly, as plot width remains constant.

Proposed Behavior
I have a number of suggestions to improve these (ordered by importance). Some of these suggestions imply an interactive visualizer (similar to q2-types or emperor) so might be unnecessarily complex here, but others are more general and could be input as parameters, similar to something like qiime1's make_distance_boxplots.py (and even if an interactive visualizer is feasible, setting these parameters programmatically should still be possible).

1. Plot width should scale by number of columns by default. Thus, a 2-column boxplot should be fairly narrow and a, say, 18-column boxplot should fill the screen.
2. Drag plot edges to manually resize
3. Allow coloring/scaling of points by metadata
4. For correlation plots, support trendlines, curve smoothing/rolling averages, and confidence intervals by metadata category (and toggle on/off data points), ideally with user-defined colors.
5. Toggle on/off outlier points on boxplots.
6. Select color for labels, axes, and background
7. Select font typeface and size (in points/picas)
8. boxplots generated by beta-group-significance and alpha-group-significance appear to be formatted differently. Making these uniform would be nice for publication.
9. Allow in-line editing of labels, e.g., to rename cryptic metadata names that are used as labels
10. Set axis ranges and intervals
11. Allow manual ordering of columns in boxplots. Dragging/dropping columns to re-order would be a neat way to do this in an interactive way.
12. Allow all plots to be printed in multiple formats: SVG, PDF, TIFF, JPEG

References

1. cannot be edited in illustrator
2. PD boxplots
3. make_distance_boxplots.py
4. beta-group-significance
5. alpha-group-significance

similar behavior to volatility plot within q2-longitudinal
https://github.com/qiime2/q2-longitudinal/blob/dev/q2_longitudinal/_vega_specs/volatility/legend.py

These can be derived from the cookiecutter template. I think this would be useful to have to facilitate testing. You should be able to edit the templated environment file to install necessary dependencies, but let me know if that doesn't work.

for each group from the selected input, calculate the mean (or selected average - mean, median, mode) and plot as a trend line with optional error shading from the standard error (or confidence interval)

q2-nothing-to-see-here

Current Behavior
It is great that many visualizations show associated statistical tests, and that for some visualizations these results can be downloaded as csv.

However, often many different tests are being performed (e.g., a separate test for each metadata category) and it would be laborious to flip through all visualizations, download results, and concatenate. It is even more laborious to to this for visualizations that do not have an option to download as a csv (e.g., alpha correlation). Just globbing these files was easy in qiime1, but not with qzv files.

Questions
What if all statistical tests associated with a single qzv were automatically compiled into a single summary table, and that table could be downloaded from a "download summary csv" button at the top of the page?

Replace cdn links in index.html assets with vendored json to ensure that visualizations still work even if (i.e., when) the server(s) associated with those links are no longer active.