Real Atlassian solutions to real problems — no fluff, no SEO spam.
Forge gives you no veto over a Confluence save. Product events arrive after the change is already committed, which means a lock on Confluence content cannot prevent anything — it can only detect and restore. Here is what that constraint does to an app: why the restore has to ignore itself, why reverting the whole page is the wrong fix, and why a licence check in a protection app must fail open.
A worker in my local-model swarm ran pytest against a file that did not exist, twice, and finished the task green. The pipe had eaten the exit code. Measured on bash and zsh, on pytest and cargo — including why turning pipefail on just moves the lie to the other side.
Jira Cloud answers 303, Confluence answers 302, and an SSO-fronted Data Center answers with a login page. I measured the two Cloud cases end to end, and the failure that follows passes every audit you are likely to write.
MCP went stateless on 2026-07-28. I probed ten real servers: none implement it. Then I measured the thing that actually costs you — 145 tool definitions, 40,784 tokens, and the 30% that is one server repeating itself.
A customer changes email domain and starts appearing twice in the reporter picker. Removing them from the project provably cannot fix it, the query that finds their tickets stays silent when it fails, and the reassign that clears them reports the wrong reason for refusing.
A blocks link in Jira records a relationship. On Free and Standard it reschedules nothing. Here's the engine we put behind it, the required-start rule that's easy to get subtly wrong, and what it still can't do.
Metal will not give you the RAM on the box, and the number it does give is not the 75% everyone repeats. I measured the three ceilings on a 96 GB Mac Studio, then measured what modern hybrid-attention models actually spend against them — including a Gemma 4 cache that quietly holds three times its own sliding window.
A saved filter is a string, not a binding. I measured what cf[10042] actually resolves to on a destination site, and where Data Center to Cloud and Cloud to Cloud each fail silently.
Our swarm workers kept re-reading files they had already read. Context compaction was summarising the tool output that held the file. Pasting the file into the summary does not fix it; returning the last turns verbatim does. Measured three ways on the same 27B.
