Docking
Computational prediction, never direct experimental binding.
Evidence methodology
Certainty should be inspectable.
Capela’s framework is designed to prevent the most consequential category errors in AI-assisted molecule research.
01
Retrieval before synthesis
The model analyzes retrieved scientific material. It does not answer from memory and search for supporting citations afterward.
02
Direct experimental evidence
A relationship or effect must be measured experimentally. Direct binding requires an appropriate binding measurement; it cannot be inferred from a downstream pathway change.
03
Indirect experimental evidence
An effect was observed experimentally, but the proposed molecular relationship itself was not directly established.
04
Computational prediction
Docking, molecular dynamics, QSAR, machine-learning prediction, target prediction, and network pharmacology remain visibly computational.
Expression change
Experimental observation, not automatic evidence of direct binding.
Review claim
Literature reported unless analyzed primary evidence independently establishes it.
Missing mechanism
Insufficient evidence is returned instead of biological plausibility.
05
Capela inference
When Capela connects findings across stored evidence, that synthesis is labeled as an inference and never presented as a direct paper result.
06
Mechanism rules
Every mechanism edge must reference existing evidence and papers. If the chain cannot be supported, the mechanism status is insufficient, partial, or conflicting.
07
Research assessment
Strength is computed from explainable inputs such as independent experimental support, directness, consistency, source depth, conflicts, and computational dependence.
08
Scientific limitations
Capela assesses what retrieved literature supports. It does not determine whether a molecule is a good drug, predict clinical success, recommend experiments, or replace expert review.