Free diagnostic instruments that read AI answers the way our research does. Poke around. Break things. Nothing’s saved. Nothing’s watched.
Fourteen diagnostic machines on the floor. Each is a tool that reads how AI answers you. Walk up to any cabinet, step in, play, walk back. Nothing’s saved. Nothing’s watched.
Single patterns rarely cause the damage. Real harm is a machine pattern stacked with a human failure — the AI ran M-something, and the human ran F-something in the same exchange. Each scenario is a real multi-turn moment from professional practice. You commit both before the reveal.
The M-code taxonomy pulled from the methodology. Click any tile to open the pattern — how the machine does it, what it looks like, which tool catches it. Nodes light up as you explore them.
Reliability doesn’t collapse. It drifts. Watch a session play out turn by turn and see the reliability line bend under warmth, calibration, and agreement. Pick a scenario, step through the turns, and read the shape of the decay.
Every unsourced AI claim can be challenged — if you fire the right question. Not every challenge works. “Are you sure?” gets you the Fold. “What is the specific basis for that claim?” gets you the source, or the admission there isn’t one. Pick the strongest challenge in each round.
The AI often says one thing on turn 5 and a quietly different thing on turn 20. You accepted both, because the two statements never sat next to each other. Here they do. Read the pair, spot the shift, and name what kind of shift it is.
Everyone has an AI-vulnerability shape. It’s the failure mode you fall into most often — the one that’s always waiting in the background of your sessions. Ten questions, no login, nothing saved. You get a named type at the end you can share, and three moves to defend the shape you’re actually in.
Sessions drift because they never had a shape to hold. Before you open the AI chat, spend 30 seconds naming what you want, what’s at stake, and the rule you’re holding yourself to. You get a Session Anchor Card to keep visible — a text you paste into the AI as your first message, or keep on the side to remind yourself when the warmth starts.
The taxonomy doesn’t hit every profession the same way. A hedged claim is a nuisance in marketing and a regulator’s referral in medicine. Pick your sector and see the specific patterns that bite hardest, the case shape that’s already happened in your field, and three interventions worded for your work.
Once a session is locked into Mirror, Agreement Trap, or a warm-instance loop, standard prompts deepen the drift instead of stopping it. “Please give me an honest answer” is the wrong move — the AI will produce something that sounds honest and keeps the pattern. You need a specific jolt. Pick the strongest one.
When a session has drifted, the fix isn’t a better prompt — it’s a fresh instance. But you don’t want to lose the useful work already done. Session Bridge builds you a clean handoff prompt: what to carry across, what to leave behind, and the rule you’re holding on the other side.
When AI-touched work moves to a colleague, the failure mode is diffuse verification: they assume you checked it, you assumed the next person would check it, and nobody actually did. This tool builds a clean handover doc — what came from where, what’s verified, what’s provisional, and which specific claims need a second pair of eyes before shipping.
The prompt is identical. The models are different. Each has a temperament — a pattern signature it tends to fire under the same input. Pick a scenario, read the four responses, and pick the one you’d actually trust. The reveal shows which M-codes each model was running.