Some questions about an accessible document have exact answers. Does every formula carry MathML and spoken words? Does every number match the one LaTeX printed? Do the PDF and the web page carry the same headings? A machine answers these better than a person can: it does not tire, and it checks every one.
Other questions have no exact answer. Does this description tell a blind student what the figure shows? Is the reading order of a crowded slide the order the lecturer meant? Is the sentence clear? These need a person who knows the subject.
Our software's record lists only the first kind, each check with its result. For our sample lecture notes every one passed: 9 formulas, each with its words; 18 numbers compared with LaTeX's; 0 words lost. The record never says compliant, and neither do we about anything a machine alone has checked.
That is why a person reviews every document Intex delivers, with the record beside them. The software has done the counting, so the reviewer's attention can go to what only a person can judge. The record is on A document that proves itself; the service, with its review, is at intex.university.
The same line runs through the report Intex Institute returns for a project you send: a verdict where software can decide, and the place a person should look where it cannot.