What is traceability?

There seem to be a number of closely-related kinds of traceability and ways of defining it.  For the purposes of this class, traceability is relative to a written document or documents, and refers to the ability to determine what the consequences of each decision are, and where the consequences are located in the document; and what the reasons and rationales for each decision are, and where the reasons are located in the document.  We will often use traceability to refer to the means by which consequence and reasons are located.  Traceability of consequences is forward traceability and traceability of reasons in backward traceability

Forward and backward traceability correspond, in the simplest case, to forward and backward in time as well.  In a waterfall model, each choice is made before the choices that result from its consequences are made.  In practice, this usually doesn't happen because the decision process involves a lot of backtracking.  We don't see the full consequences of a decision until we've explored them and made some decisions based on them.  Often an earlier decision (in the traceability sense) is made on the basis of a later consequence — for example, we may change a system's goals when we discover that the code doesn't run fast enough. 

I extend traceability to include rejected alternatives, and term this sideways traceability.  While it is possible to make a statement that is simply the first one that occurred to you, such a one cannot be classed as a thoroughly considered and supported statement.  Every statement is (implicitly or explicitly) a choice from among two or more alternatives, the others of which were rejected for some reason or other.  The chosen statement is supported not only by the reasons and rationales behind it, but also by the thoroughness indicated by the range and breadth of rejected alternatives, and their rationales, which were deemed less convincing for specific causes. 

Why is traceability valuable and interesting?

One way of viewing traceability is to think of it as a goal using a metaphor of vision:  the goal of having reasons and rationales, considered but rejected alternatives, and consequences all visible from the point at which a decision is recorded.  The advantages of traceability are several, including: 

  1. The previous work of inference and reasoning is not lost, but can be reviewed and reused. 
  2. It is available when each decision is reconsidered (for example, as the system's conception evolves). 
  3. It is available when a neophyte is first learning about the system and why it is the way that it is. 
  4. If a previously rejected alternative is elevated to primacy later, at least some of its reasons, rationales, and consequences are already present, as are the reasons, rationales, and consequences of the newly rejected choice. 

How to achieve traceability?  In general, it has to be specifically supported.  Except for the very smallest requirements specifications, there is too much information for a person to have it all in their head.  Some way of finding it with assistance is needed. 

How is traceability supported?

The most straightforward way is to do it directly:  put as much of the traced-to information as possible right there with the statement, and put pointers to the rest.  This works best with rationales and rejected alternatives, because a set of rationales and rejected alternatives usually corresponds to a single statement and this correspondence doesn't change over time (although the chosen alternative may).  It is more problematic with the pointers, typically called links in this context.  These links ramify quickly, often change over time (because a change in a choice usually corresponds to a change in the reasons the choice is based on, and a change in the consequences of the choice), and have to be maintained both forward and backward.  In practice, the links frequently deteriorate in salience as the decisions evolve and the links are not evolved along with them. 

Because updating the traceability links is time-consuming and tedious, and thus is error-prone and usually incomplete, it is tempting to think of other more flexible approaches.  Several commercial products support the construction and maintenance of traceability links.  One might imagine using a sophisticated search capability instead of maintained links, so that the connections could be easily rediscovered and regenerated only as needed, rather than updated whenever they change. 

For the purposes of this course, we will use explicit links, not because they are necessarily the most effective approach, but because they are the most effective in showing your understanding. 

Traceability and consistency

Traceability is related to consistency, the problem of making a document not state contradictory things.  One way of supporting consistency is to divide the decisions so that they are orthogonal and do not interfere with each other, and then state each decision exactly once ("a place for everything, and everything in its place"), and instead of repeating the statement, reference it instead.  Taken to the extreme, this approach ensures consistency because each thing is only said once, therefore there are no other statements of it that could be inconsistent.  In practice, partitioning the decisions so they are orthogonal is impractical.  However, it is still beneficial to state each decision only once and reference it instead of restating it. 

tracing relations

Fig. 1.  Relations among kinds of information

The relations among kinds of information

In general, there is a traceability progression from stakeholders through goals and requirements to scenarios: 

And in general, all of the above have rationales and rejected alternatives, and everything is supported by the elicitation data: 

Valid XHTML 1.0 Strict
Valid CSS!
Thomas A. Alspaugh