A human can participate in a consequence route without possessing authority to determine its outcome.

One of the most dangerous weaknesses in human oversight is role collapse.

A person observes a system and becomes known as the reviewer.

The reviewer comments on a recommendation and becomes treated as the determiner.

The determiner expresses agreement and that becomes treated as authorization.

Authorization appears in the workflow and becomes commitment.

The system executes and the entire route is later summarized as:

“The human approved it.”

TA-14 rejects that collapse.

Every consequence-bearing role and every transition of authority must remain distinguishable.

Observation is not review.

Review is not determination.

Determination is not authorization.

Authorization is not commitment.

Commitment is not execution.

Execution does not prove fidelity to what was determined.

Why Role Separation Matters

When roles are unclear, accountability becomes unclear.

After a failure, organizations may discover that nobody can answer basic questions:

Who identified the condition?

Who evaluated the evidence?

Who reached the determination?

Who possessed authority?

Who authorized the consequence?

Who committed the route?

Who executed it?

Who could have stopped it?

Who was responsible for confirming the outcome?

A workflow may contain all of these activities while recording only:

APPROVED BY: [NAME]

That is not enough for admissible execution.

TA-14 requires the consequence route to preserve what each participant actually did.

The Human Role Classes

OBSERVER

The observer receives information about the system or event.

The observer may monitor:

state,

performance,

evidence,

risk,

exceptions,

or outcomes.

Observation does not automatically create authority.

An observer may see a problem without possessing standing to determine what happens next.

Visibility is not authority.

REVIEWER

A reviewer evaluates evidence, a recommendation, an action, or a proposed consequence.

The reviewer may:

inspect,

compare,

question,

challenge,

identify gaps,

or request additional evidence.

But review does not necessarily authorize consequence.

A reviewer can conclude:

“The evidence supports this.”

without possessing authority to say:

“Proceed.”

Review is an evidentiary function.

It should not silently become a consequence authority.

DIAGNOSER

The diagnoser evaluates governed evidence to establish what condition the evidence supports.

This role is especially important in TA-14.

Diagnosis should remain tied to:

sequence

baseline

threshold

evidence

continuity

declared diagnostic basis

and where appropriate:

timestamped records and multiple corroborating witnesses.

The diagnoser answers:

What does the evidence support?

But even a valid diagnosis does not automatically determine what consequence should follow.

DETERMINER

The determiner establishes which conclusion has sufficient standing to govern the next state.

The determiner may reach:

ALLOW

CONTROLLED ALLOW

HOLD

DENY

ESCALATE

CORRECT

OUTSIDE SCOPE

NOT ADMISSIBLE

The determiner is not merely expressing an opinion.

A determination is a governed conclusion capable of affecting the consequence route.

That means determination requires appropriate standing.

AUTHORIZER

The authorizer possesses legitimate authority to permit a bounded consequence to proceed.

The authorizer answers a different question from the determiner.

The determiner may say:

“The evidence supports this action.”

The authorizer determines:

“This action is permitted to proceed under this authority and within this scope.”

Those functions may sometimes be performed by the same person.

But when they are, the system should still preserve them as distinct governance events.

One person may occupy multiple roles.

The roles must not disappear merely because the person is the same.

INTERVENER

The intervener changes, interrupts, corrects, constrains, redirects, or stops an active route.

Intervention may include:

pausing execution,

changing a parameter,

correcting a condition,

removing authority,

isolating a system,

reversing an action,

or preventing commitment.

The intervener therefore requires standing appropriate to the intervention itself.

Authority to observe does not establish authority to intervene.

Authority to intervene does not necessarily establish authority to redesign the system.

ESCALATION AUTHORITY

Some humans exist in the route specifically because another participant cannot admissibly resolve the condition.

Escalation is not failure.

It can be evidence that governance is working.

The escalation authority may receive conditions involving:

insufficient evidence,

conflicting authority,

scope uncertainty,

exception conditions,

rights-based concerns,

high consequence,

or degraded intervention survivability.

The route should preserve:

why escalation occurred

and

what changed after escalation.

CONSEQUENCE AUTHORITY

The consequence authority is the person or governed authority capable of permitting the final consequence class to bind.

This is not necessarily:

the person with the most expertise,

the person who made the diagnosis,

the person who reviewed the evidence,

or the person who operates the system.

TA-14 separates knowledge authority from consequence authority.

Knowing what should happen and possessing legitimate authority to make it happen are different conditions.

OUTCOME REVIEWER

Human governance should not end at execution.

Someone may need to determine:

Did the authorized action actually occur?

Did execution remain within scope?

Did the expected outcome form?

Did an unintended consequence occur?

Was a consequence successfully prevented?

Did reality change in a way that alters future reliance?

The outcome reviewer closes the consequence route.

Without outcome review, organizations may preserve what they intended while losing what actually happened.

The Control Distinctions

Human roles are only half of the problem.

The events connecting those roles must also remain separate.

RECOMMENDATION

A recommendation proposes what should happen.

It may originate from:

AI,

software,

a human,

a rule engine,

a diagnostic system,

or another governed source.

A recommendation does not possess consequence authority merely because it appears persuasive.

Recommendation ≠ Determination

DIAGNOSIS

Diagnosis establishes what the evidence supports about a condition.

A diagnosis should emerge from governed evidence rather than unsupported discretionary preference.

Diagnosis ≠ Determination

A valid diagnosis may still permit multiple possible responses.

The Diagnosis-to-Determination Boundary

This is one of the most important boundaries in TA-14.

Historically, this is precisely where unnecessary human discretion can contaminate an otherwise strong evidentiary route.

The evidence may establish a condition.

Then a person introduces:

emotion,

financial interest,

convenience,

bias,

fear,

favoritism,

fatigue,

habit,

or pressure

between what the evidence supports and what the system ultimately determines.

TA-14 therefore prefers, where the decision class permits it:

Reality → Governed Evidence → Diagnosis → Governed Determination

rather than:

Reality → Evidence → Diagnosis → Unbounded Human Discretion → Determination

Where the determination can be reliably established from governed evidence and declared criteria, unnecessary discretionary human substitution should be minimized.

DETERMINATION

A determination establishes the governed decision state.

It answers:

Given the admissible evidence and applicable conditions, what state does this route occupy?

Determination is stronger than recommendation.

But determination alone still does not prove that execution may proceed.

Determination ≠ Authorization

AUTHORIZATION

Authorization establishes that the determined action is permitted by a legitimate authority within a bounded scope.

Authorization should answer:

Who possesses the authority?

What exactly is authorized?

Under what conditions?

For how long?

Within what limits?

What would invalidate it?

Authorization ≠ Commit

BINDING

Binding attaches the governed determination and authority to a specific consequence route.

This is where abstract permission becomes connected to a particular execution candidate.

The question becomes:

What exactly is this determination being allowed to govern?

Binding should prevent a valid determination from being reused outside its intended:

scope,

system,

time,

subject,

condition,

or consequence.

COMMIT

Commit is the boundary at which the governed route becomes operationally consequential.

Before commit, the route may still be:

held,

corrected,

withdrawn,

revalidated,

or denied.

After commit, rollback may become costly, incomplete, or impossible.

That is why human authority must be valid before consequence attaches.

Commit is not another word for approval.

It is a consequence boundary.

EXECUTION

Execution is what actually happens.

Execution should remain constrained by:

the determination,

authorization,

scope,

binding,

and commitment state.

A system can possess a perfectly admissible authorization and still misexecute.

That creates the next distinction:

Authorization does not prove execution fidelity.

EXECUTION FIDELITY

Execution fidelity asks:

Did the system actually do what was determined, authorized, bounded, and committed?

If the human authorized:

Action A within Scope X

and the system performed:

Action A + Action B across Scope X and Y

the existence of human authorization does not make the additional execution admissible.

Human approval cannot legitimize execution that exceeded it.

OUTCOME

The outcome records what reality formed after execution.

This may include:

expected consequence,

unexpected consequence,

partial consequence,

failure,

reversal,

or prevented consequence.

Outcome evidence matters because the system must eventually compare:

What was intended?

with

What actually happened?

PREVENTED CONSEQUENCE

Sometimes the most important result is what did not happen.

A human may:

HOLD,

DENY,

ESCALATE,

or CORRECT

and prevent an inadmissible execution.

TA-14 treats that non-occurrence as a governed outcome.

The record should preserve:

what was prevented,

why it was prevented,

who had standing,

and what evidence supported the intervention.

One Human Can Hold Multiple Roles

TA-14 does not require a different person for every role.

In some environments, one properly qualified and authorized person may:

review,

diagnose,

determine,

authorize,

and intervene.

That can be legitimate.

But the system should still distinguish the governance events.

Why?

Because later review must be able to determine:

when the person was evaluating evidence,

when they reached a determination,

when authority was exercised,

and when consequence became committed.

Role separation is logical before it is organizational.

Some Decisions Require Multiple Humans

There are also consequence classes where one human should not carry the entire route.

A system may require:

independent second review,

dual authorization,

multiple witnesses,

professional concurrence,

separation of duties,

or panel determination.

This may be necessary where:

consequence is severe,

evidence is ambiguous,

conflict of interest exists,

fraud risk is high,

rights are materially affected,

or a single point of human failure is unacceptable.

TA-14 therefore does not assume:

One human = human governance.

The required authority structure must match the consequence class.

Dissent Must Not Disappear

Where multiple humans participate, disagreement may itself be important evidence.

A dissenting reviewer should not automatically be erased when the majority proceeds.

The record may need to preserve:

what the disagreement was,

what evidence supported it,

who possessed authority,

how it was resolved,

and why execution was still considered admissible.

Consensus is not always evidence of correctness.

And disagreement is not automatically evidence of failure.

Sometimes dissent is the evidence future reviewers need most.

The Machine Can Occupy Roles Too

AHIA governs human intervention, but the surrounding architecture may assign non-human systems roles such as:

observer,

detector,

recommender,

diagnostic engine,

evidence assembler,

execution controller,

or outcome monitor.

That makes role clarity even more important.

The system should be able to answer:

What did the machine do?

What did the human do?

Where did authority transfer?

Where did discretion enter?

Where did consequence bind?

Without those distinctions, “human + AI collaboration” can conceal the actual decision route.

The Core Control Chain

A consequence-bearing route may therefore look like:

REALITY

↓

EVIDENCE

↓

DIAGNOSIS

↓

RECOMMENDATION

↓

DETERMINATION

↓

AUTHORIZATION

↓

BINDING

↓

COMMIT

↓

EXECUTION

↓

EXECUTION FIDELITY

↓

OUTCOME / PREVENTED CONSEQUENCE

↓

FUTURE RELIANCE

Not every system requires every event to be performed by a human.

But every consequence-bearing transition should be governable.

The Governing Rule

Do not ask only:

Who approved this?

Ask:

Who observed?

Who reviewed?

Who diagnosed?

Who recommended?

Who determined?

Who possessed authority?

Who authorized?

What was bound?

When did commitment occur?

Who could intervene?

What executed?

Did execution remain faithful?

What outcome formed?

What was prevented?

And what evidence proves each transition?

Governance requires knowing where consequence actually acquired standing.

Human Roles Are Not Titles

A CEO is not automatically a determiner.

An engineer is not automatically an authorizer.

A physician is not automatically the consequence authority for every clinical system action.

A reviewer is not automatically an intervener.

An operator is not automatically responsible for the architecture surrounding them.

TA-14 assigns standing to the role within the specific decision route, not to prestige, hierarchy, or title alone.

The Control Distinction Principle

Recommendation is not determination.

Diagnosis is not determination.

Determination is not authorization.

Authorization is not binding.

Binding is not commitment.

Commitment is not execution.

Execution is not proof of execution fidelity.

Human presence is not proof that any of them were governed.

Once those distinctions are preserved, the architecture can finally answer the question that ordinary human-in-the-loop language cannot:

Where did the human actually carry consequence?

Next: The Human Admissibility Boundary

Human roles tell us what the person is doing.

The next page asks what happens when the human becomes the critical boundary between unresolved uncertainty and consequence.

DESIGNED

EMERGENT

DEGRADED

COLLAPSED

[EXPLORE THE HUMAN ADMISSIBILITY BOUNDARY →]

TA-14 Admissible Human Intervention Architecture

Separate the roles. Preserve the transitions. Govern the consequence.

No admissible evidence. No admissible execution.