Screening-level analytics: what they are, and what they are not

Much of what AI produces in engineering today is screening-level analysis: rapid, systematic first-pass assessment across a whole dataset. Screening is genuinely valuable and genuinely limited, and the industry's trust in AI depends on never blurring the two. So here is our working definition, stated as plainly as we can.

What screening does

• Covers everything: every borehole scored, every claim line matched, every log checked — no sampling, no fatigue.
• Applies established methods consistently: the same cited correlation, the same measurement rule, everywhere.
• Ranks attention: it tells the professional where to look harder, earlier than any manual pass could.

What screening is not

• It is not detailed design. A screening bearing estimate scopes a foundation strategy; it does not size a footing for construction.
• It is not certification. A matched payment chain still needs a person to approve what gets posted.
• It is not judgment. It has no opinion about what a soft zone means for this structure, this client, this risk appetite.

Why the boundary is a feature

A tool that claimed to do detailed design from raw logs would be untrustworthy precisely because the claim is too large. A tool that does screening completely, cites every method, and hands a ranked picture to a professional is trustworthy because its claim is exact. Every Greenshift analytic states its level on the output. The boundary is not small print; it is the design.

Used this way, screening changes project economics quietly: risks surface in week one instead of month three, and expert hours concentrate where the data says they matter. That is a large prize for an honest claim.