How publishing works in this field
In most CS subfields the selective conferences carry more weight than most journals, are fully peer reviewed, and publish archival proceedings — ACM and IEEE both index them. Deadlines are fixed, review is a single round with a rebuttal rather than iterative revision, and a rejected paper waits for the next cycle rather than being revised in place.
Preprints are normal. Posting to arXiv before or at submission is standard practice and does not count as prior publication at the major venues, though a double-blind conference will have rules about when and how you may publicise it. Journals in the field generally accept arXiv preprints too, but confirm rather than assume.
What slows papers down here
Artifact evaluation is now common at the stronger venues: a separate committee runs your code to check it builds, reproduces the reported numbers, and is usable. Code that only runs on the author's machine fails this, and the badge increasingly matters to readers.
Baselines are where papers die in review. A method compared only against weak or outdated baselines, or evaluated on datasets chosen after seeing results, draws rejection regardless of how good the idea is. Decide the comparison set before running the experiments.
Worth reading next
Questions we are asked in this field
In most of CS, a selective conference is the stronger venue and the faster route. Choose a journal when the work needs more space than a page limit allows, when it extends a conference paper substantially, or when the people assessing you only count journal articles — which is worth checking first.
Not at ACM or IEEE venues or at most CS journals, where preprints are accepted practice. Double-blind conferences usually restrict active promotion during review rather than the posting itself. Check the specific venue policy, since a minority of publishers outside CS treat preprints as prior publication.
More each year. Several venues run artifact evaluation, reviewers increasingly ask for code during review, and a paper whose results cannot be reproduced is discounted even when accepted. Package it so it builds from a clean checkout, not just on your machine.