Can AI find SML without being told its name?
EntityForge measures whether AI systems discover ScriptMasterLabs from the problem itself, whether they can resolve distinctive SML attributes back to the correct entity, what they cite, and which unexpected semantic territories may justify a new SML product branch.
No hidden hints inside the discovery number.
Fingerprint prompts are useful because they test whether models can connect SML-specific facts without seeing the brand name. They are never blended into PURE discovery, so a descriptive hint cannot inflate the category-level recognition score.
Category discovery
No company name. No unique SML fingerprint. The question expresses only the real buyer problem or category.
Attribute resolution
No company name. One to three distinctive, already-public attributes. Tests whether the system joins the facts back to SML.
High-clue resolution
A specific public fact combination tests entity-resolution strength. Useful diagnostically, but never called organic discovery.
What can SML actually prove?
Evidence classes are not interchangeable.
Recognition can tell us what to build next.
Strong recognition in an unexpected intent, repeated competitor gaps, or recurring unresolved buyer problems can become a product-improvement signal or a new SML branch. Nothing becomes a product claim automatically.
Before EntityForge.
This card preserves the imported external report exactly as a dated sample. It is not substituted for the new holdout benchmark.
The recurring holdout wording is intentionally not published here.
Publishing the exact test corpus on an indexable SML page could eventually make the benchmark train on its own answers. EntityForge exposes probe IDs, classes, intents, evidence, and results while recurring holdout wording stays out of public artifacts. Public calibration prompts may be used separately, but they never count as sealed holdouts.
Evidence graph, score definitions, result summaries.
Probe class and semantic intent.
Exact recurring prompt wording.
PURE, FINGERPRINT, EXACT scores.