Telos

I. What “for” adds

Consider a housing-repair office whose stated purpose is to keep homes habitable. Its staff want to help tenants. Its managers reward a falling backlog. Its software counts a request as resolved when the ticket is closed.

A tenant still has no heating. Someone closes the ticket.

What is the office’s telos? Habitable homes, helping tenants, reducing the backlog, or closing tickets? Calling all four its “purpose” hides the question that would distinguish them: which end is being claimed, by whom, and in what relation to the working arrangement?

Telos means an end or purpose: that for the sake of which something is done. Aristotle distinguishes this explanation from an account of what a thing is made of or what immediately moved it. Health can explain why someone walks, while muscles and physical forces explain how the walking happens. These explanations answer different questions.¹

An end need not be a final moment after which activity stops. A journey can terminate at a destination; heating can maintain a condition; an inquiry can develop a capability; music can be played for the activity itself. Aristotle already distinguishes ends that are activities from products made through activities.² A telos can concern an outcome, a maintained state, a way of acting, or a trajectory through time.

In the repair office, habitability supplies a way to distinguish a completed repair from completed paperwork. It tells us what the office's activity is for. We still need to establish whether that end directs the office, merely appears in its documents, or supplies the standard against which we are judging it. A purpose can also be malicious, mistaken, or self-defeating; identifying it and justifying it are different tasks.

II. Four questions about purpose

The useful distinction is between four roles an end can occupy. These are neither successive stages nor mutually exclusive species of purpose. One end can occupy several roles at once.

The office declares habitability. A technician may sincerely intend to restore the heating. An inspector can use habitable homes as the reference telos. Meanwhile, the arrangement can make administrative closure its operative priority: staff are penalized for open tickets even when the underlying fault remains.

The declared and intended ends need not match. A statement can be insincere, while a sincere intention can fail to control action. Calling an organization “well-intentioned” also leaves open whose intention is meant. A technician's wish, a manager's plan, and a board's commitment are different facts.

Reference and operative do not mean imaginary and real. A reference telos is not a claim about what the focal arrangement follows. It is the end relative to which an evaluator makes a success-or-failure claim, and its force depends on the grounds for using that standard. An operative telos is a claim about what directs the arrangement.

The same end can occupy both roles. In an effective repair service, habitability is the criterion used to evaluate the work and the condition that directs inspections, assignments, and reopening of failed repairs. That is alignment. Where those roles separate, the reference makes the separation visible.

“Actual telos” is useful when it means the operative end, as opposed to a merely announced one. “Actual” becomes misleading when it means whatever happened, or suggests that every system contains one hidden objective waiting to be recovered. A result is something to explain. It is not automatically an end the system pursued. In this corpus, operative telos is the preferred causal term. Actual telos is only shorthand, because actual otherwise slides between the end directing the arrangement and the outcome it happened to produce.

Other modifiers answer different questions. A mandated telos identifies a source of authority; a designed telos identifies a design intention; an adopted telos identifies a commitment. None alone tells us whether that end governs operation. An official mandate can be the declared and reference telos while another priority controls decisions. An ascribed telos is one an observer attributes: that locates the claim, without settling whether the attributed end is genuinely operative.

III. What the working arrangement pursues

Return to the tenant without heating. Two offices could produce exactly this result for different reasons.

In the first, the replacement part is unavailable. Staff keep the failure visible, arrange the available temporary remedy, and complete the repair when the part arrives. Habitability directs their efforts even while the home remains cold. The office has failed to achieve its end; that does not establish that it pursues another one.

In the second, the repair is feasible. Closing the ticket earns credit; acknowledging that the fault persists creates trouble. Reports from tenants are reclassified as new requests, and reopening a case damages the team's score. When managers receive credible evidence of unrepaired homes, they preserve the closure rule.

The second account identifies more than a bad outcome. It identifies the choices, incentives, and response to correction that keep selecting administrative success over repaired homes. An operative-telos claim earns its explanatory value by distinguishing what the arrangement will preserve when its apparent ends conflict.

Failure alone leaves too many alternatives open: missing resources, false beliefs, a broken sensor, insufficient authority, an unsuitable method, conflicting obligations, or pursuit of a different end. The evidence becomes more discriminating when the analyst establishes what the actors knew, which alternatives were feasible, what costs they faced, and what changed after a credible challenge.

There is a formal counterpart to this problem in preference inference. Armstrong and Mindermann show that behavior alone, without sufficient assumptions about how an agent plans or errs, does not uniquely identify its preferences; a simplicity preference does not remove the ambiguity. That result concerns a particular inference framework, but it illustrates why an explanation needs more than a retrospective story about the observed choices.³

“Closing tickets is the operative telos” should therefore make a difference to investigation. Inspect what is rewarded, which cases are excluded, who can reopen them, and whether staffing or measurement changes restore attention to the homes. Evidence that the closure count is routinely sacrificed to complete difficult repairs would weaken the diagnosis.

The operative end can also be emergent. No employee needs to desire the resulting arrangement. Mission statements, career incentives, resource constraints, routines, and exposure to blame interact through the decision process. Where their interaction repeatedly privileges a recognizable end, that end is the operative resultant. The causal account must show how it is produced; the word telos does not supply the missing mechanism. The Response Vector examines how an intervention's pressure can take a route its designer never intended.

Some arrangements support only a domain-specific account: closure governs complaint handling, while safety governs emergency response. Others are too fragmented for one governing end to explain them. A defensible diagnosis can be “several competing purposes, no settled priority at this level.” It need not invent a unified optimizer.

IV. Purpose without a person

Purpose-language can concern more than conscious intention, but the relation must remain visible.

A hammer is for driving nails because of its design or use. The hammer itself does not notice a crooked nail and correct the strike. The person and the wider activity supply that direction. A broken hammer can retain its assigned purpose while failing to perform it.

A heating system controlled by a thermostat presents a different case. Its setpoint, sensor, switching mechanism, heater, and room form an arrangement that responds to departures from a target temperature. The room's temperature can be disturbed while the target remains unchanged. The correction is produced by existing physical machinery; no future state needs to reach backward and move the present.

Here the target may have been chosen by a person, but continuing regulation does not require that person to contemplate each adjustment. The origin of an end and the location of its implementation are different questions. Conversely, naming a use for an object does not install a controller inside it.

Telic Systems uses a stricter architecture test: a boundary, a carried specification, detection of deviation, corrective action, and a viability condition. Saying that an artifact has a purpose does not by itself establish membership of that category. The hammer's use-purpose and the heating arrangement's implemented target are both intelligible, without pretending that the hammer regulates itself.

This nonmental use of purpose is a substantive explanatory choice. Critics of organizational teleology, including Michael Harmon, have challenged the move from describing behavior as if purposeful to asserting that an organization has a purpose independently of human intention.⁴ The claim made here rests on inspectable organization: a particular target, signals, choices, and corrections. Attributing a mind, a destiny, or moral authority to the arrangement adds claims that the mechanism does not establish.

Historical selection introduces another relation. What caused an arrangement to be retained is not automatically what its present operation pursues. A reporting system may have been purchased because it once helped repairs; that history does not establish what its current incentives select. Selection can help explain the origin of an operative priority, but the explanation still has to connect past selection to present machinery.

The distinction is explicit in learned-optimization research. Hubinger and colleagues distinguish the objective under which a model is trained from the objective a learned optimizer might pursue. The latter question arises when the learned model is itself an optimizer; it is not a license to assign a hidden, coherent goal to every model.⁵ Training criterion, designer intention, deployed pursuit, and eventual outcome remain separately investigable.

V. Whose end, at which level?

A telos claim needs a focal unit. “The purpose of the organization” cannot silently change from the board's commitment to an employee's incentive, then to the interests of everyone affected.

It also needs a distinction between the arrangement being evaluated and the people or system whose condition matters. The repair office is the focal arrangement. Tenants are the intended beneficiaries. The municipality may be the wider reference system whose housing, finances, and capabilities are being assessed. Those roles can overlap, but they are not identical.

Keeping the office solvent can be a necessary means to repairing homes. Preserving its current budget or organization need not be the ultimate end. A temporary repair unit could perform its task, transfer its knowledge, and close. Its disappearance would then be success under its mission, despite failure under a different reference that treated the unit's continued existence as paramount.

Ends and means are relative to a commitment. Raising money is an end for a fundraising operation and a means for the work it funds. Terminal describes an end valued for its own sake within the commitment under discussion; instrumental describes its contribution to another end. Neither modifier tells us whether the end is declared, intended, reference, or operative.

A function, in a task or design description, is a contribution to a larger activity: the reporting system makes unresolved work visible so repairs can be assigned. The telos tells us what makes that contribution useful. Other effects of the same component—employment, paperwork, institutional prestige—need not be the end for which it is used.

Different levels can succeed and fail simultaneously. A department can faithfully meet a defective mandate. A firm can prosper while imposing costs on others. An institution can depart from its mandate in a way that benefits the wider system. Each claim needs its own reference; local fidelity does not settle wider benefit, and wider benefit does not by itself establish authorization.

Nor do many local purposes automatically compose into one purpose for the whole. The office, contractor, finance department, and tenants can each respond coherently to their circumstances while nobody owns the combined result. The parts' intentions do not create a collective intention by addition. From Telos to Policy develops the corresponding problem of nested reference systems and cross-level accounting.

Time belongs in the same account. Clearing today's backlog by creating next month's failures and clearing it by completing durable repairs can look identical in a weekly report. A long-horizon telos evaluates the intervening trajectory as well as the destination: a future benefit cannot repair a transition that destroys the capacity needed to reach it.

A boundary makes evaluation possible; it does not make effects outside the boundary disappear. Choosing the municipality as the immediate reference still leaves tenants, contractors, neighboring communities, and future occupants in the causal account. Moving the boundary is a change in the evaluation, not a way to delete a cost.

VI. Goals, measures, and plural ends

A telos becomes useful in a decision when it distinguishes acceptable from unacceptable movement. That does not require reducing everything to one number.

“Keep homes habitable” needs further specification. What conditions count? Which failures are urgent? How are competing needs handled? What resources and transition risks constrain the work? A commitment can be genuine before every case has been settled, but the unresolved choices remain unresolved. A broad end does not contain its own complete implementation.

A target might require restoring essential services within a defined period. A metric might count completed repairs. A model might estimate whether a particular intervention will restore habitability. These play different roles: the end supplies the standard, the model predicts how to reach it, and the measurement supplies evidence about progress. All three can be wrong in different ways.

A repair count is useful when it tracks repairs. Once closing a record earns credit without a repaired home, the number can improve while the end is defeated. The repair may be to change the measurement or the incentives while preserving the end. Changing the score is not necessarily changing the telos, just as keeping the same score does not prove that the telos was preserved.

Some requirements are part of the end itself. “Resolve disputes fairly” is not equivalent to “produce decisions quickly, with fairness as an optional inconvenience.” Whether a condition is constitutive, a side constraint, or an expendable means depends on the commitment being examined. That distinction cannot be settled by relabeling every difficult requirement a proxy.

Plural purposes can have an explicit order. A service might first address dangerous conditions, then reduce waiting among the remaining cases, while preserving enough capacity to continue operating. Some alternatives can be ruled out without completely ranking the rest. Where two remaining choices are incomparable under the stated commitments, a further judgment is required; the absence of a total ranking does not erase the commitments already made.

Likewise, “optimize for” need not mean “maximize one scalar without limit.” It can mean preserving a viable region, meeting conditions, improving a partially ordered set of outcomes, or searching for better means under several commitments. From Telos to Policy supplies the more detailed account. A weighted score is one possible representation, whose weights and permitted substitutions require justification of their own.

VII. Choosing and changing the reference

A reference telos answers what counts as success in this evaluation. That phrase locates an obligation to give reasons. It does not license an arbitrary standard.

An internal critique can use the institution's own undertaking: it was entrusted with housing repairs, so assess whether homes were repaired. An external critique can challenge that undertaking against a wider defended end. The first tests fidelity; the second tests the adequacy of what fidelity would deliver. Neither should borrow the other's authority without making the additional argument.

The distinction leaves room for discovering that a purpose is badly specified or morally wrong. A reference can be inconsistent, rest on false beliefs, ignore affected people, or require an impossible trade-off. Causal effectiveness does not vindicate it. A coercive organization can pursue its operative end competently while that end remains indefensible to the people it harms.

Using a reference for analysis is also different from having authority to impose it. An analyst can expose a contradiction in a mandate without acquiring the power to rewrite it. Formal authorization establishes an institutional fact; whether the authorized purpose is defensible is a further question. Telocracy concerns how a public reference can be adopted, contested, pursued, and revised rather than silently supplied by the strongest local pressure.

Facts bear on this argument throughout. They reveal dependencies, possibilities, consequences, and inconsistencies. Once a system and horizon are specified, they can show what maintaining it requires. They do not make that system's persistence the ultimate end merely by explaining its requirements. Survival may serve a purpose, constitute part of it, or preserve something that should be replaced.

The broader reference advanced in this corpus is sustained flourishing as broad adaptive safety margin. It is a substantive commitment about what civilization should preserve and develop, not a definition hidden inside the word telos. Evaluating a local arrangement against its own end need not settle that entire argument; evaluating the end itself requires stating the wider reference and defending its relevance.

Correction therefore has more than one possible target. New evidence may require a better method, a better measure, a more precise account of the same end, or a genuine revision of the end. Those changes should remain distinguishable. Revising an authorized goal openly is different from changing what wins while continuing to claim the old goal's authority.

Telic Corruption names a stricter case of the latter: a warranted reference and entrusted capacity, repeated priority displacement, a practicable correction path, and correction machinery that protects the displacement. Ordinary failure or a mere difference between reference and operative telos is not the whole diagnosis.

A reference used to assess another reference is sometimes called a meta-telos. “Meta” identifies the relation. It grants neither automatic superiority nor immunity from the same questions about grounds, scope, and consequences.

VIII. Making the claim usable

The repair-office example can now be stated without collapsing its parts. Habitability is the declared end and the inspector's reference. A technician's intended end may agree with it. Administrative closure is a candidate operative end only where the working arrangement privileges it in the relevant conflicts. Unrepaired homes are an outcome. The office's wider costs and benefits require a further, explicitly located evaluation.

Those distinctions change the repair. Missing parts call for supply or capacity. A misleading completion count calls for better evidence. A rule that knowingly rewards closure over repair calls for a change in the governing incentives and correction process. An inadequate mandate calls for reconsidering the mandate. The same visible failure can require different interventions because the end, the means, and the implementation are different objects.

A usable telos claim can fit into one sentence:

For [focal unit], in [domain] and over [horizon], [end] is claimed as its [declared / intended / reference / operative] telos relative to [whose condition counts or which reference host], supported by [statement / intention evidence / evaluative warrant / working-mechanism evidence].

Then supply the evidence that role requires: a statement, promise, mandate, or specification for a declared claim; evidence of a person's aims for an intended claim; the comparison rule and its grounds for a reference claim; the working mechanism, trade-offs, and response to deviation for an operative claim. Where the operative priority is uncertain, preserve the competing explanations and identify what would distinguish them.

Purpose becomes useful when it lets us explain a choice, identify a failure, or direct a correction. A reference telos keeps the standard visible when practice drifts. An operative account shows what must change for the standard to govern. Keeping both in view makes it possible to demand more than sincere intention without declaring every produced result to be what the system was for.

Related:

• Telic Systems — The stricter test for an arrangement that implements corrective pursuit.
• The Stack — Locating a reference telos and system boundary before analyzing mechanisms.
• From Telos to Policy — Turning a reference into constraints, choices, and correction.
• Telic Corruption — Protected purpose substitution under entrusted authority.
• The Response Vector — How pressures produce responses that differ from the intended end.
• Flourishing Is Maximum Safety Margin — The wider reference commitment, rather than the definition of telos.