wtf is output-surface-integrity?

Output Surface Integrity, or OSI, is a method for handing off long running work between people and AI assistants without making the next person or AI reread everything that came before. Instead of starting a new AI chat by explaining the whole project again, OSI keeps a short summary of exactly what state the work is in, so a new session can pick up from there. The core idea addresses a specific problem the author calls false completion. An AI assistant might say a task is done, but without proof of what was actually checked, what is still unresolved, or who owns the next step, the person receiving that handoff still has to verify everything themselves and figure out where things stand. OSI's answer is to record, at the moment work pauses, what changed, what evidence backs that claim, what is still unresolved, where work should restart, and who is meant to take the next action. If responsibility is said to have transferred to someone else, OSI also wants proof that the receiving person actually accepted it, with a timestamp. The project includes an internal test where a project's history, over 14,000 characters long, was compressed down to about 3,267 characters using this method, a reduction of almost 78 percent, while still keeping every item the author tracked as needed for a restart. The author is careful to say this single test does not prove a general rate of savings or that the method works the same way everywhere else. There is no software to install for the basic version of this. It is presented as a workflow and a set of documents: a five minute template for checking whether a piece of work is really finished, and a shorter one minute check for restarting an AI session with less repeated explanation. The project is released under the MIT license.