By 2028, AI is the first screener in enterprise procurement.
Before a buyer reaches a seller, an AI agent has already decided what kind of company you are. If it files you under the wrong category, you are never in the evaluation set. This is not search ranking. It is whether the model puts you in the category a buyer is actually shopping.
The wrong category is not a ranking problem. It is a pipeline that never starts.
Wrong category puts you against the wrong competitor set — judged on criteria that were never yours to win.
Self-disqualification happens before discovery. The buyer never forms the question you would have answered well.
The pipeline never starts. There is no lost deal to analyze, because there was never a conversation.
The observed finding
Inference from output behavior only. Not a claim about model internals.
Every finding traces to an archived probe.
Each card carries the surface, prompt family, output excerpt, classification, probe ID, and artifact hash. The archive reference is the credibility.
What ran, where.
The recommended proof change
The re-probe plan
What this packet does, and does not, claim.
- Shows live AI characterization behavior from archived outputs.
- Attributes likely drivers from observable output evidence.
- Identifies one proof change worth testing.
- Creates a re-probe plan with preregistered gates.
- Auditable: every finding has a probe ID and archive reference.
- Claim the client confirmed these findings.
- Claim measured movement — that requires the A4 gate.
- Provide buyer-outcome evidence.
- Give legal advice.
- Claim ranking causality.
- Say "AI excluded you because X."