3.2 How this study works¶
The pages that follow make many claims, and they are not all equally strong. Some rest on a dated primary source; some are a conclusion drawn by joining several such sources; some are reasoned extrapolation; and some are the vendor's own marketing, quoted so it can be weighed rather than believed. So that you can tell which is which, every non-trivial claim carries a small tag. This section is the legend for those tags — the one thing worth reading before the evidence begins.
3.2.1 The evidence tags¶
| Tag | What it means |
|---|---|
| [Evidence] | Stated in a primary or reputable, dated source, and footnoted. The firm ground. |
| [Synthesis] | A conclusion that combines several sourced facts. As reliable as its parts, but the joining is ours. |
| [Inference] | Reasoned extrapolation beyond what any single source states. Treat with more caution. |
| [Marketing] | The vendor's own framing — quoted, not endorsed. It appears most where classical technology is sold as AI. |
| [Evidence of absence] | A verified negative: we looked for a capability and confirmed it is not there. |
Two conventions round this out. Tags combine when a claim spans more than one
kind — [Evidence + Synthesis], [Synthesis + Inference] — and they take
qualifiers when the ground is softer than usual: [Evidence — thin],
[Evidence of absence — limited]. The qualifier always means what it says.
3.2.2 Negative findings are results¶
The [Evidence of absence] tag deserves its own note, because it carries some of
the study's most important sentences. When we searched for a capability — a
learned warpage predictor, a geometry-and-outcome archive, an OEM investing in
moulded-part AI — and found, after deliberate looking, that no one had built it,
that is not a gap in the research; it is a finding. Several groups' sharpest
conclusion is a verified absence, and it is reported as first-class evidence, not
buried as "not applicable."
3.2.3 Reading the capability matrices¶
The Analysis scores each company, at each workflow stage, by the kind of mechanism it uses. The codes are exactly the six mechanism classes from Chapter 2 — the classical trio that computes (§2.2, R rules, P physics, O optimization) and the AI trio that learns (§2.3, K retrieval, ML machine learning, L language models). A dash (—) means nothing of the kind was found at that stage. Reading the matrix is therefore a direct application of the honest-AI test, company by company.
3.2.4 A word on trust¶
The academic foundation behind this study was produced under adversarial verification: every claim was independently re-checked during research; claims that failed re-checking were discarded and never reused. The same spirit applies to vendor numbers — where a figure is vendor-published with no independent audit (most of them are), the report says so at the point of use. And the "is it AI?" verdicts lean on the same argument as Chapter 2: optimization and design-of-experiments relabelled as AI are flagged, not accepted, with a pointer to What is Optimization technology.
With the legend in hand, the group studies come next.