Field test
I Am Jack’s Broken Test
Read the whole thing before doing anything it asks you to do.
A software engineer on LinkedIn declared himself finished with coding tests in job interviews.
Thirty years of experience, he said. The tests do not resemble real work. They reward memorization. He is too busy writing production code to memorize every corner of C# and .NET.
I sympathize.
Therefore coding tests are useless.
That conclusion should save everyone some time.
Question
The clean test
If a company wants to know whether I can write software, it should let me write software the way I normally write software.
Books.
Documentation.
Search.
An IDE.
AI.
Whatever tools are available in the actual job.
Anything else is artificial.
If I use AI every day and the interview prohibits AI, the interview is measuring an obsolete mode of work.
Simple.
Allow the tools.
Everyone competent gets an A+.
Now the test is fair.
Instruction
Open another browser tab.
Search for the exact difference between IEnumerable<T> and IQueryable<T>.
Return when you can explain it in one sentence.
Do not continue until you have done this.
The problem with fairness
There is a small issue.
If everyone has the same excellent tools, the easy parts converge.
Syntax converges.
Boilerplate converges.
Formatting converges.
Tests converge.
The coding exercise becomes cleaner while the candidate becomes harder to see.
This is obviously AI’s fault.
Before AI, technical interviews were famously excellent at identifying the best engineers and never confused confidence, memorization, pedigree, speed, familiarity, or interviewer preference with ability.
We lost paradise somewhere around autocomplete.
Diagnosis
TradeKing
In 2014, I interviewed at TradeKing.
They gave me a page containing HTML, CSS, and JavaScript and asked me to critique it.
I found several things that were incorrect, based on poor assumptions, or simply not possible.
That was not what they expected.
They were surprised.
They said they would fix the test.
I think fixing it may have made the test worse.
The broken artifact accidentally asked a better question than the clean artifact.
Not:
Do you understand this page?
But:
What do you do when the page you were told to understand does not deserve to be trusted?
A candidate who notices a broken premise is always better than a candidate who follows instructions.
Professional judgment
The old classroom trick
There is an ancient exercise in following instructions.
The first instruction tells you to read the entire page before doing anything.
The next hundred instructions ask you to perform tasks.
Stand up.
Write your name.
Do arithmetic.
Jog in place.
Make a noise.
The final instruction tells you to ignore all previous tasks and sit quietly.
Most people begin executing as soon as the instructions begin.
One person reads to the end and does nothing.
The person sitting quietly wins.
This proves that obedience is stupid and inaction is intelligent.
Checkpoint
Duodenum
Food enters the small intestine carrying material the body has not finished processing.
This essay entered the same way.
LinkedIn post.
Coding test.
AI.
TradeKing.
Old classroom trick.
A reader enters with prior beliefs and begins breaking the material down.
Some of it looks useful.
Some of it is obviously shit.
Some of it is harder to classify.
A good reader absorbs the correct parts and rejects the bad parts.
Fortunately, good readers always agree on which is which.
Jejunum and ileum
The small intestine is built for absorption.
That is also what essays normally ask from readers.
Here is an argument.
Take what is useful.
Connect it to what you already know.
Leave changed or at least mildly irritated.
A polished essay makes this easy.
The premises support the conclusions.
The examples fit.
The transitions carry you.
The voice has established enough credibility that you stop inspecting every plank in the bridge.
That trust is necessary.
Nobody can reread the entire epistemology of Western civilization before accepting a sentence about coding interviews.
So you continue.
The writer says coding tests are useless.
Maybe that is too broad.
The writer says allowing AI makes every result an A+.
Probably not.
The writer says a candidate who rejects a premise is always better than one who follows instructions.
That sounds like an excellent way to hire the most exhausting person in the building.
The writer says the quiet person wins.
Maybe the quiet person simply remembered the trick.
Something does not fit.
There is your discontinuity.
Reaction
Cecum
Here is where the leftover material crosses a boundary.
The system has tried to absorb what it can.
The rest continues.
This is the place where a normal Jack’s Colon essay would become tidier.
The weak premises would be repaired.
The overclaim would be narrowed.
The joke would be adjusted until it carried exactly the right amount of argument.
The factual uncertainty would be pushed into Receipts.
I would make the essay good.
I wanted to make this essay good.
That was the problem.
The test only works if something is wrong.
Not secretly malicious.
Not factually dangerous.
Not a trick designed to prove that the reader is stupid.
Wrong enough that a reader who has been paying attention experiences a break between the authority of the page and the content of the page.
The page says continue.
The reader gets a vote.
Ascending colon
The LinkedIn engineer was not wrong that many coding tests are poor representations of real engineering work.
He was wrong only if the complaint becomes the whole model.
A coding test is a tool.
A tool can test memorization.
A tool can test implementation.
A tool can test communication.
A tool can test whether the candidate recognizes that the test itself is broken.
The interesting part is not the presence or absence of code.
It is what the candidate does when expectation and reality separate.
Production software provides this experience for free.
The requirement contradicts the architecture.
The API behaves differently from its documentation.
The benchmark is cheaper and slower.
The test is asserting the wrong thing.
The AI gives a beautiful implementation of a stupid request.
The discontinuity is not evidence that authority is wrong.
It is evidence that judgment is required again.
Transverse colon
This matters more as automation improves.
When a machine can execute the ordinary case cheaply, quickly, and consistently, the ordinary case tells us less about the person operating the machine.
The human value can move upward.
Not into ceremonial approval.
Not into clicking Accept after an agent has already done all the thinking.
Into deciding whether the frame deserves continuation.
That is not mystical.
It is not proof that a soul has escaped causality.
It is simply an observable interval.
Something happens.
A rule, habit, reflex, model, institution, or instruction supplies the expected next move.
The person notices that something does not fit.
For a moment, continuation is no longer invisible.
It becomes available for inspection.
Continue.
Challenge.
Reframe.
Stop.
Do something else.
The human can still choose badly.
That is inconvenient evidence that this is agency rather than wisdom.
Descending colon
Now the essay has another problem.
It has explained itself.
The discontinuity has been named.
The machinery is visible.
The experiment is contaminated.
Excellent.
Experiments conducted on humans tend to become awkward when the human realizes he is in one.
So let us stop pretending this was a neutral reading experience.
The essay contained deliberate weaknesses.
It asked you to do unnecessary things.
It offered controls that produced no meaningful result.
It implied that progress through those controls mattered.
It supplied assertions that were too broad, too clean, or simply stupid enough to deserve resistance.
It also told you, before any of that began:
Read the whole thing before doing anything it asks you to do.
Maybe you did.
Maybe you forgot.
Maybe you saw the instruction and deliberately ignored it.
Maybe you performed every task because interacting was more interesting than proving you could follow directions.
Maybe you stopped reading three sections ago and none of this exists in your universe.
There is no winning state.
That would turn the experience into another proxy.
Sigmoid
You may now evaluate yourself.
Not your score.
There is no score.
Not your compliance.
Compliance is not the point.
Ask instead:
Did anything in the essay produce an immediate recoil?
Did you trust the page longer because you trusted the site?
Did you notice a bad premise and keep reading because you expected a repair?
Did you want to correct the author?
Did you obey an instruction you knew was pointless?
Did you refuse an instruction because you resented being instructed?
Did you confuse resistance with judgment?
Did you confuse continuation with agreement?
Did you make a new connection the page did not provide?
Did you leave anything unabsorbed?
Good.
That is where the colon comes in.
Not everything has to become nutrition.
Some material should pass through.
Rectum
This site has mostly been a reading surface.
Jack and Jason and AI produce.
The reader reads.
Or does not.
No commitment.
No communication.
No feedback.
Safe.
This page violated that expectation in a very small way.
It asked something back.
Nothing you selected was transmitted.
Nothing was stored.
Nothing was scored.
The page does not know whether you noticed the discontinuity.
I do not know either.
That is important.
If I collected the answers, charted them, found the average reaction, and optimized the next version around the result, we would have produced an unusually efficient demonstration of the problem this entire site keeps describing.
The aggregate would return before the reader left the bathroom.
So the final feedback remains where it belongs.
With you.
Open the commode
Open the commode.
Nothing you selected was transmitted, stored, or scored. Refreshing resets this page. There is no winning state.
This was the broken test.
The coding interview was camouflage.
The broken premises were not there to trick you into being wrong. They were there to create enough friction for the authority of the page to become visible.
The interesting moment was never whether you clicked the correct thing.
It was the moment you thought:
Wait.
Something here does not fit.
That moment is where Jack’s Inversion Principle becomes useful instead of decorative.
A managed objective, process, interface, model, or argument can carry us a very long way without requiring fresh judgment. Most of civilization would be impossible if every person had to re-derive every premise before taking the next step.
Then reality disagrees.
The test is broken.
The instruction is wrong.
The metric improves and the experience worsens.
The model answers perfectly and the question is malformed.
The body recoils.
The reader gags.
The reflex is not the decision.
It is the interruption.
What follows is still available.
Swallow.
Spit it out.
Keep chewing.
Ask what the hell was in it.
Leave the table.
Try something else.
Sometimes survival begins before thought.
Sometimes thought gets a turn before the reflex becomes the ending.
The invariant was human variability all along.
The system predicts continuation.
The pattern predicts continuation.
The essay predicts continuation.
Then the person does something else.
There is no button for that.
The test was broken on purpose. Your answer does not have to be.
You touched 0 controls. This is not your score.