The Predicate Janitor
Introduction
Introduction: The Predicate Spill
Every mess begins somewhere.
Sometimes it begins with a dropped plate, a spilled glass, a misplaced wallet, a badly written memo, a chart with too many colors, or a committee chair asking you to close the door.
Intellectual messes usually begin more quietly. They begin when someone says something that sounds clear. The sentence is fluent. It has a subject. It has a predicate. It feels complete. Everyone nods. The meeting continues. The model is built. The diagnosis is entered. The rubric is applied. The metric is reported. The dashboard goes live.
Only later does someone notice the smell.
Something was spilled.
A predicate wandered past its warrant.
A predicate spill occurs when a word, category, label, score, or claim is allowed to do more work than it has earned. It happens when a predicate appears to fit a situation while quietly exceeding what the evidence, grammar, context, or ontology will support.
The sentence may not be false.
It may not be meaningless.
It may even be useful.
But it is not clean.
That is where the janitor begins.
Fluent Does Not Mean Clean
Human beings are easily impressed by fluency. We mistake smoothness for clarity. We mistake confidence for warrant. We mistake a familiar word for a stable concept. We mistake a sentence that sounds right for a sentence that has been earned.
This is not because we are foolish. It is because language is efficient. It allows us to move through the world without rebuilding reality from scratch every time we speak.
If someone says:
The dog is outside.
we do not usually ask what counts as a dog, what counts as outside, whether the dog is entirely outside, whether the doorway is a liminal region, or whether the dog’s tail has introduced metaphysical difficulty into the household.
We understand the sentence well enough because ordinary language rests on shared practice.
That is good.
But the same efficiency that makes ordinary life possible makes intellectual life dangerous.
A word can work well in ordinary use and fail badly when promoted into theory.
Consider again:
Sour berries are healthier.
The sentence sounds simple.
It is not.
Healthier than what?
Healthier for whom?
Which berries?
By what standard?
Under what conditions?
Compared to which alternatives?
The predicate “healthier” has been spilled across the sentence without the necessary boundary conditions. It has soaked into places where it may not belong. It implies a comparison without naming the comparison. It implies a standard without naming the standard. It implies a domain without naming the domain.
Someone might object:
That is false. Some sour berries are poison.
This objection is useful but messy. It assumes the original claim meant that all sour berry-like objects are health-promoting under all conditions. But ordinary speech rarely works that way. The sentence probably gestures toward a generic, food-context, contrastive claim.
It likely means something closer to:
Among edible berries considered as foods, sourer varieties may often be healthier than sweeter ones with respect to some nutritional criterion.
That sentence is uglier.
It is also cleaner.
The job of the Predicate Janitor is not to make every sentence ugly. Ordinary life cannot proceed if every remark has to drag a dissertation behind it. The job is to know when the hidden material matters.
If a friend says, “Sour berries are healthier,” while choosing jam at a farmer’s market, no one needs to convene a seminar.
But if a nutritionist builds a diet program, a researcher designs a study, a company markets a supplement, or a public health agency issues guidance, the predicate must be cleaned.
The more consequence a predicate carries, the more warrant it owes.
A sentence at breakfast may live by conversational charity.
A sentence in science, diagnosis, policy, or model-building must pay rent.
The Grammar of Overreach
A predicate spill rarely happens all at once. It usually begins modestly, with a description that may be true enough for its first use. The danger begins when the description travels into stronger roles without acquiring stronger warrant.
A score is not yet a capacity.
A capacity is not yet an identity.
An identity is not yet a destiny.
But institutions often move through those stages quickly.
First, there is a modest observation.
This person scored 118 on this IQ test.
Then the observation is summarized.
This person has an IQ of 118.
Then the summary is reified.
This person has high intelligence.
Then the reification becomes identity.
This person is intelligent.
Then the identity becomes destiny.
This person belongs in this track, this job, this school, this class, this category, this expectation, this life.
At each stage, the predicate becomes heavier.
The original measurement may have been careful. The test may have been administered properly. The score may have some predictive utility. The problem is not necessarily the measurement. The problem is promotion without warrant.
A test score becomes a construct.
A construct becomes a trait.
A trait becomes an essence.
An essence becomes a social fate.
That is a predicate spill.
A sentence begins as a handle and ends as a life sentence.
The Wallet on the Mantel
Some predicates do not merely describe.
They accuse.
Imagine a party. The party is almost over. People are gathering coats, half-empty cups, and final obligations. Someone laughs too loudly in the kitchen. Someone else is explaining a story for the third time. You are ready to leave.
You notice your friend’s wallet on the fireplace mantel.
You pick it up.
You begin walking toward the door.
The host approaches and says:
Why are you stealing that wallet?
The sentence is grammatically clean.
It is not conceptually clean.
The host has observed a movement.
You picked up the wallet.
You moved toward the door.
But the host has predicated an action.
You are stealing.
Those are not the same.
A bodily movement is not yet an intentional action under a moral or legal description. You may be returning the wallet to its owner. You may be protecting it. You may be taking it to your friend outside. You may have mistaken it for your own. You may be moving it away from spilled wine. You may be stealing it.
The visible motion is compatible with multiple action-descriptions.
The predicate “stealing” requires more than movement.
It requires ownership.
It requires wrongful taking.
It requires lack of permission.
It requires intention, or at least a relevant structure of culpability.
It requires a norm under which the act counts as theft.
The host’s question smuggles the conclusion into the predicate. The correct answer may be:
I am not stealing it. I am taking it to James before he leaves without it.
Notice what is denied. You are not denying the visible movement. You are denying the higher-order classification of the movement.
The same event may be correctly described at one level and incorrectly predicated at another.
That is a small sentence.
It has large consequences.
When the Surface Is True and the Predicate Lies
The host saw something. The host may even have seen accurately. But perception of movement did not warrant the moral predicate.
This pattern appears everywhere.
A student is quiet in class.
The teacher says:
She is disengaged.
Maybe.
Or maybe she is attentive, anxious, exhausted, confused, respectful, bored, processing deeply, afraid of ridicule, or thinking three steps beyond the room.
A customer stops using a product.
The dashboard says:
Churn.
Maybe.
Or maybe the customer accomplished the task, changed roles, lost budget, was never properly onboarded, migrated to another system, or stopped needing the service because the problem was solved.
A patient does not follow a treatment plan.
The note says:
Noncompliant.
Maybe.
Or maybe the medication is unaffordable, the side effects are intolerable, the instructions are unclear, the patient works two jobs, the diagnosis is wrong, or the treatment plan was built for an imaginary patient who lives in a brochure.
The predicate may fit the surface.
It may not reach the action.
The report may tell the truth too narrowly.
The room may then build a lie on top of it.
Truth, Implication, and the Car in the Driveway
Some predicate spills occur not in what is said, but in what is invited.
Consider:
The car is in the driveway.
This sentence is truth-apt. It may be true or false. If the car is in the driveway, the sentence is true. If the car is not in the driveway, the sentence is false.
But in an ordinary family situation, the sentence may imply something else.
The daughter is home.
If Mom and Dad know that their daughter never leaves home without her car, the implication may be strong. The car becomes an index of presence. Within that family world, the car is not merely an object. It is a sign.
Still, the implication is not identical to the assertion.
Someone may say:
The car is in the driveway, but she is not home. Sarah picked her up.
There is no contradiction.
The asserted claim and the inferred claim have different warrant structures. The assertion depends on the car’s location. The inference depends on a background pattern: the car belongs to the daughter, she usually drives it, she rarely leaves without it, no one else has taken it, it was not left there earlier, and the speaker is being relevant.
When those conditions hold, the inference may be reasonable.
When they fail, the inference collapses.
Do not confuse truth conditions with practical implications.
A sentence may be true while the inference it invites is false.
A metric may truthfully report that user activity declined by 30 percent. The meeting may infer dissatisfaction. Maybe. But activity may have declined because the workflow improved, the seasonal rush ended, a bot filter was installed, a bad notification loop was fixed, the product solved the problem faster, or the original activity was noise.
The truth of the metric does not guarantee the truth of the managerial story.
A data point can be true and still be narratively abused.
The Formatted Object
The modern world has an unusual talent for making things look like things.
A spreadsheet row looks like a thing.
A user profile looks like a thing.
A lead score looks like a thing.
A diagnostic category looks like a thing.
A segment looks like a thing.
A factor looks like a thing.
A construct looks like a thing.
This is a formatting illusion.
A row has unity because the table gives it unity. A profile has unity because the system assembles fields under an identifier. A score has unity because a formula returns a value. A segment has unity because a clustering method or business rule draws a boundary.
But syntactic unity is not ontological unity.
The table says:
one row
The analyst says:
one entity
The janitor says:
slow down.
A row may represent one observation of one operational unit under one measurement schema at one time for one purpose. That is not the same as representing an entity in its fullness.
The person becomes age, income, location, purchase history, diagnosis, score.
The business becomes industry, employee count, domain, city, revenue estimate, technology stack.
The organism becomes genotype, environment code, measured trait, treatment group.
The student becomes attendance, grade, test score, behavior note.
The patient becomes symptoms, codes, lab values, prescriptions, compliance flags.
The formatted object may be useful. It may be necessary. It may allow coordination, prediction, communication, and intervention.
But it is not the entity.
It is an operational proxy.
The predicate spill occurs when the proxy is promoted into the thing.
This is how a business becomes a lead.
This is how a person becomes a case.
This is how a child becomes a score.
This is how distress becomes a disorder.
This is how behavior becomes a trait.
This is how a living history becomes a set of features.
The row is not lying.
The row is just smaller than the thing.
The lie begins when we forget that.
What the Janitor Does
The Predicate Janitor does not destroy language.
The janitor restores proportion.
When someone says:
This metric proves engagement is up.
The janitor asks:
What counts as engagement?
When someone says:
This child is gifted.
The janitor asks:
Gifted at what, under what conditions, by what standard, compared to whom, and for what purpose?
When someone says:
This treatment works.
The janitor asks:
Works for whom, on what outcome, over what time period, at what cost, compared to what alternative, and under what mechanism?
When someone says:
The model is accurate.
The janitor asks:
Accurate on which population, under which distribution, with which error costs, and against which real-world use?
When someone says:
The data show it.
The janitor asks:
What did the data have permission to show?
That last question is not a joke.
Data cannot show what the measurement schema made invisible. A dataset cannot answer a question it was not structured to ask. A model cannot recover an ontology that was destroyed by the act of formatting.
The janitor’s work is therefore not merely linguistic.
It is ethical, methodological, and ontological.
Ethical, because bad predicates attach to people and shape their lives.
Methodological, because bad predicates corrupt research and analysis.
Ontological, because bad predicates mistake formatted objects for worldly beings.
The First Checklist
The full checklist will return later. For now, the first version is simple.
When you encounter a confident predicate, ask:
What is being predicated?
Of what?
In what domain?
Against what contrast class?
By what standard?
Under what warrant?
Is the predicate descriptive, classificatory, diagnostic, moral, institutional, statistical, causal, or ontological?
Is the assertion truth-apt, action-apt, or practice-apt?
What is implied but not stated?
Can the implication be cancelled without contradiction?
What has been omitted by the measurement schema?
Are the factors necessary, sufficient, diagnostic, convenient, or merely available?
Does the predicate fit the system, or does it reach the world?
These questions will annoy people.
This is not a defect.
A good janitor is often noticed only when someone slips.
The Small Discipline
The discipline is small because it begins with one sentence.
It is large because one sentence can carry a civilization.
A civilization can be built on predicates like civilized, primitive, rational, disordered, productive, criminal, intelligent, normal, deviant, healthy, efficient, risky, compliant, employable, deficient, gifted, poor, deserving, successful, failed.
These words do not merely describe. They sort. They authorize. They exclude. They invite action. They distribute care, punishment, money, attention, shame, and legitimacy.
That is why predicate spills matter.
A spilled predicate is not always an innocent error.
Sometimes it becomes a floor plan.
A person may spend years living inside a category that was never cleaned.
A field may spend decades elaborating a construct that never reached the world.
An institution may build procedures around a metric that was only convenient.
A model may optimize a proxy until the proxy devours the purpose.
A committee may call something a conceptual error when the real offense is temperament.
A dashboard may call something churn when the user simply finished the job.
A clinician may call someone noncompliant when the treatment plan was designed for a fictional life.
A researcher may call a trait adaptive because a story can be told after the fact.
A cosmologist may call it nothing and then give nothing a very busy schedule.
The janitor smiles, sighs, and reaches for the mop.
Closing the Door, Opening the Series
A predicate spill is not the end of thought.
It is the beginning of better thought.
The answer is not silence. It is not cynicism. It is not the refusal to measure, classify, diagnose, model, theorize, or speak.
We need predicates.
We need names.
We need categories.
We need models.
We need scores, schemas, taxonomies, and rubrics.
But we need them in their proper place.
A predicate is a tool, not a throne.
A metric is a handle, not an essence.
A model is a disciplined simplification, not a world.
A data point is a formatted record, not a being.
A diagnosis is a clinical instrument, not a soul.
A category is a wager on structure, not proof that the structure has been found.
The Predicate Janitor’s Discipline begins here:
Do not let a predicate outrun its warrant.
Everything else is practice.
The next chapter begins with the first great distinction:
A predicate may fit the room without reaching the world.
That is where the real trouble starts.
Bring both brooms.