Generally, in an assessment, a "knock-out criterion" (plural knock-out criteria) means a standard which, if not met, results in disqualification or failure: it "knocks out", eliminates, the thing, entity or person being tested. In the following case, for example, standards are being assessed:

"Knock out criteria are those to which a ‘no’ answer would mean that the standard was not suitable and would not be considered further."
http://standards.data.gov.uk/meeting/open-standards-board-me...

Another example:

"Furthermore, for a specific business need or Member State context, certain criteria within the assessment step may be defined as being knock-out or blocking criteria. This means that a ‘NO’ answer for a knock-out criterion will lead to discarding the proposed formal specification."
https://webgate.ec.europa.eu/fpfis/mwikis/idabc-camss/index....

So in these cases it refers to absolute requirements: thresholds which must be met regardless of how well something performs in other respects.

In relation to tunnel safety testing, which is your context, it is not quite so simple. The system includes "knock-out criteria", but this does not mean that failure to meet a given safety level in an individual category results in automatic failure. In this case, it refers to a statistical weighting method, designed to ensure that an inadequate result in a given category is not excessively outweighed by good results in other categories.

The testing method involves an elaborate numerical system by which points are awarded for levels of performance in different categories. The points are basically added together to arrive at an overall numerical assessment. This means, however, that a tunnel which is good in most respects but very poor in one important respect, for example, may still be rated as acceptable overall. So what they do, when the "knock-out criterion" or threshold is not met in a certain category, is to assign a numerical value which is used to reduce the results by a given proportion. The proportion depends on various factors which reflect how important the poor result is deemed to be.

So in this case, "knock-out criteria" are not directly disqualification criteria, but rather downgrading criteria: standards which, when not met, do not directly disqualify the tunnel but do downgrade its rating.

The precise working of this in the particular context of tunnel testing is explained in this document. Here is the general statement:

"The disadvantage of this purely additive rating is that a very poor result in one category can be "compensated" by positive results in other categories and has hence either no or very little influence on the overall rating. In 2006, the introduction of knock-out criteria derived from the four safety pillars meant that it was now possible to correct the basic result."

This is on p. 36 of the file. See pp. 36-38 for an explanation of the statistical method used; it is very difficult to summarise briefly.
http://eurotestmobility.com/images/filelib/EuroTAP_Tunnel_Au...