A Ratio between Tests

During one of my recent projects I was considering the ratio between the checks and the tests – that is the ratio between those tests, that are primarily simple binary confirmations, and those tests that are more tacit questions. This blog is about my considerations on the idea/experiment/model.

First I observed, that we have a range of different items in our requirements – some of them are [actual copy of the current specification]:

Binary Confirmations

  • It must be possible to add a customer ticket reference
  • It must be possible to copy the ticket number

Tacit Questions

  • You must be able to navigate displayed lists easily
  • It must be easy to compare work log and audit log

You could argue that they need refinement, more testability and less “easy“. But this is what we have to work with for now. Even if we had all the time in the world (we don’t) – we would not be able to write all of the requirements in a perfect form (if such a form exists).

As the system under test is a commercial standard system, some of the requirements are even given as “Out of the Box”, we will probably not even be testing those explicitly. Our coverage criteria is not ALL OF THEM.

Ordering the tests

It is a deliberate experiment from my side to divide the requirements (and hence the tests) into the piles of Closed and Open Questions. Perhaps there is even three piles – Rapid Software Testing has: human checking, machine checking and human/machine checking , Wardley has Pioneers, Settlers and Town Planners. Perhaps the Rule of Three applies here too.. perhaps it’s an continuum … let’s see.

Perhaps it’s a continuum

As part of the requirement workshops I will label the requirements and align with the stakeholders to get the expectations right – with the help of a few friends. This a context/project based “operationalization“.

I wrote about this ratio on my blog post around the Test Automation Pyramid, as I will use the labels to automate the confirmations (and only the confirmations). The assumption is, that there are significantly more of the binary requirements tested by machine checking – and more human tested tacit questions. If we can get all the tedious tasks automated – that is the really the end goal.

Automate all the things that should be automated

Alan Page

Every project/context will have it’s own ratio, depending on a range of factors. Saying there should always be more of one type than the other would not hold. As the above project is the configuration and implementation of a standard commercial business software package (like SAP, SalesForce etc), my expectation is that most of the requirements are binary. Also considering that this project is heavy on the right hand side of the Wardley Map scale of evolution.

It’s a Reduction in Context

I am well aware that the two/three piles are an approximation / reduction. Especially when looking at the “binary” requirements and “only” testing these by confirmation. They could as easily be explored to find further unknown unknowns. If we prioritize to do so – it all about our choice of risk.

It is also an limitation as “perfect testing” should consist of both testing and checking. I factor this into the test strategy, by insisting that all of the requirements are tested both explicitly and implicitly. First of all most of our binary requirements are on the configuration and customization of the out-of-the-box software solution. So when the subject matter experts are performing the testing of the business flows, they are also evaluating the configuration and customization. And I do want them to spot when something is odd

The binary configuration is ok, but human know-how tells us otherwise.

Ultimately I want to use the experts to do the thinking and the machines to do the both the confirmations and the tedious tasks.

Test ALL the things

TL;DR: We can add testing to all requirements and all business risks. Testing to document requirements and to debunk risks provides valuable information for the business. Let us not limit testing to things that can be coded. The intellectual activity of trial and learning is happening anyways, we might as well pitch in with ways to find important evidence for the decision makers.

Test all the requirements

Traditionally testing was all about testing the functional requirements that could be coded. Non-functional requirements was left for the specialists, or plainly disregarded. I know I have done my share of test planning, with a range of requirements left “N/A” with regards to testing. Especially performance scope, batch jobs, hardware specs, data base table expansions and virus scanning have been left out of my functional test plans…

When I look at a list of requirements now – I see that we can indeed test all the things, or we can at least work on how to document that the requirement is fulfilled. Some of the requirements are actually quite easy to document. If it’s on a screen somewhere, take a screen shot and attach it to a simple test case. Done deal really. Additionally with a testing mind-set I can think of ways to challenge the details. But do we really, really need to fill up a disk to establish if it’s exactly a 1 Gb allocation – probably not. Do we really really need to document all requirements – yes in some contracts/contexts it’s important for the customer to know that everything has indeed been established. Sometimes the customer doesn’t trust you otherwise, sometimes the tests are more about your ability to deliver and provide evidence that matters.

Test all the business risks

Look into the business case of your project and find the business risks. Sometimes they are explicitly stated and prioritized. A a recent Ministry of Testing Meetup we looked into a case for a large national health system. We looked at the tangible benefits, intangible benefits and on the scored business risks.  What worried the business and management most was budget, time and whether the new system would be used in a standardized way. There is an opportunity for testing here to help address, document and challenge the most important business  risks. Traditional testing would usually look at functional requirements that can be coded or configured, and miss totally to address what matters most to the business.

OK, how do we test the project costs? How about frequent checkpoints of expected spending – would that be similar to tracking test progress. Perhaps – let’s find out. Testing is all about asking questions for the stakeholders and solving the most important problems first. Can we help to analyse risks and setup mitigation activities – sure we can. We just have to step out of our traditional “software only bubble”.

MEME - Test ALL the things
Meme ALL the things

 

 

 

 

 

 

 

Read also: Many Bits under the Bridge, Less Software, more TestingTest Criteria for Outsourced SoftwareThe Expected, Fell in the trap of total coverage.

Links: “A Context-Driven Approach to Delivering Business Value”, Cynefin In Software TestingTesting during Application Transition Trials

 

The Expected

Many test processes and test tools mention that you have to establish the expected results to every test. It is at best risk ignorance not to take the notion of expected results with a kilo of salt #YMMV.

If you can establish the result of your solution in a deterministic, algorithmic and consistent way in a non-trivial problem, then you can solve the halting problem. But I doubt your requirements are trivial.. or even always right. Even the best business analyst or subject matter expert may be wrong. Your best oracle may fail too.  Or you may be focussing on only getting what you measure.

Nailed it
Nailed it

When working with validation in seemingly very controlled environments changes and rework happens a lot, as every new finding needs the document trail back to square one.. Stuff happens. Validation is not testing, but looking to show that the program did work as requested at some time. It is a race towards the lowest common denominator. IT suppliers can do better that just to look for “as designed” [1].

Still the Cynefin framework illustrates that there are repeatable and known projects, and in those contexts you should probably focus on looking to check as many as the simple binary questions in a tool supported way, and work on the open questions for your testing.

Speaking of open ends – every time I see an explicit expected result I tend to add the following disclaimer inspired by Michael Bolton (song to the tune of nothing else matters [2]):

And nothing odd happens … that I know of … on my machine, in this situation [3]

And odd is something that may harm my user, business or program result

Significantly…

But I’d rather skip this test step  and work on the description of the test and guidelines to mention:

And now to something completely different:

See also: The unknown unknown unexpressed expectationsEating wicked problems for breakfast

1: Anyone can beat us, unless we are the besttodays innovation becomes tomorrows commodity

2: https://www.youtube.com/watch?v=tAGnKpE4NCI 

3: I’ve heard that somewhere…