Context Integrity Kit — Free reading sample
While I was adjusting my personal AI, an execution report showed a candidate passing its tests while the switch to the setup used for ordinary conversation remained undone.
If I took away only “passed,” I could imagine that my usual AI now had the new configuration. But only the candidate had been checked, with application still ahead. I can use the test result to move forward while keeping that distinction: it does not make the remaining work complete.
What did that pass establish?
Suppose a test supplies a handoff document and produces the expected response. It establishes that the AI could answer with that document available. That one response does not establish that the document will reach the next conversation or that the usual settings have been switched.
If I only want to keep the configuration as a candidate, the result may be enough. If I want to change ordinary use, I also need to check its application. Even after material is in place, attributing an actual response to a particular configuration is a separate question, within what I can observe.
CIK distinguishes Verified—a named check meeting its stated condition—from Active—a configuration attributable to an actual response or session. A test pass alone does not establish that the configuration is being used in ordinary conversation.
One thing to check in your own AI
Choose one recent settings change and look at its records.
What was checked, under which conditions? How far has applying this configuration to ordinary use actually progressed?
If only the candidate tests are complete, look at what the records establish about pending application, mismatched references, or use in ordinary conversation. Narrowing down the next question may let you continue without repeating every test from the beginning.
Use what the test established, then check what is still missing for the next step.
You can use what you have already confirmed, even when the whole setup is not yet verified. Continue from there.
About Context Integrity Kit
A practical kit in English for people who keep developing and maintaining a personal AI, including its instructions, records, and references. It shares experience and decisions from actual use: mismatched save destinations, settings that do not reach the intended setup, repair targets, handoffs, and the work of maintaining them.
The paid edition includes five long cases and fifteen briefs in addition to this sample.
Continue with the full kit
The free sample is a separate brief. The paid edition contains five longer accounts and fifteen other briefs, with six guides and optional tools for applying their judgments. It is an English digital download for personal, non-commercial use.
Explore the paid contents · Price and terms
The Standard Edition is available for USD 29, plus applicable tax, as a one-time purchase. See purchase details.