Human governance needs more than APPROVE or REJECT.

Many consequence-bearing systems reduce human judgment to two buttons:

APPROVE

or

REJECT

That may be convenient for workflow design.

It is not sufficient for serious governance.

Reality is not always binary.

Evidence may be incomplete.

Conditions may have changed.

Authority may be uncertain.

The proposed action may be valid only within additional constraints.

The person receiving the decision may not possess standing to determine it.

A defect may be correctable.

Another authority may be required.

Or the available evidence may simply be insufficient to permit consequence.

The TA-14 Admissible Human Intervention Architecture therefore establishes explicit Human Intervention States.

ALLOW

CONTROLLED ALLOW

HOLD

DENY

ESCALATE

CORRECT

OUTSIDE SCOPE

NOT ADMISSIBLE

Each state means something different.

Each changes the consequence route differently.

And none should be silently converted into another state merely to keep execution moving.

Why Binary Approval Fails

Consider a reviewer who receives insufficient evidence.

The interface offers:

APPROVE

REJECT

What should the person do?

Approval would manufacture certainty.

Rejection may imply that the proposed action has been substantively determined to be wrong.

But neither conclusion is supported.

The actual determination is:

HOLD — insufficient evidence to determine.

A binary workflow forces a human to answer a question the evidence cannot support.

That creates governance distortion.

TA-14 instead requires the system to preserve the actual state.

1. ALLOW

The evidence and governing conditions support progression.

ALLOW means the human has sufficient standing and the consequence route satisfies the conditions required for progression within the declared scope.

ALLOW does not mean:

“Everything is safe.”

“Nothing can go wrong.”

“The human guarantees the outcome.”

It means:

The available admissible basis supports progression under the governing conditions of this decision.

An ALLOW determination should be bounded to:

the specific action,

the specific scope,

the operative state,

the applicable authority,

the relevant evidence,

and any applicable time boundary.

An ALLOW should not become a reusable permission for unrelated execution.

2. CONTROLLED ALLOW

Progression is admissible only under explicit conditions.

Sometimes execution can proceed, but only if additional controls remain in force.

Examples:

reduced scope,

limited duration,

enhanced monitoring,

human presence during execution,

specific thresholds,

restricted permissions,

additional evidence capture,

rollback readiness,

or mandatory outcome review.

That is:

CONTROLLED ALLOW

A Controlled Allow is not a weak approval.

It is a bounded authorization state.

The controls are part of the determination.

If those controls disappear, the authorization may disappear with them.

Remove the control and you may remove the admissibility.

3. HOLD

Do not proceed yet.

HOLD is one of the most important states in AHIA.

It means:

The current route does not possess sufficient basis to proceed, but permanent denial has not been established.

A HOLD may result from:

missing evidence,

stale evidence,

continuity uncertainty,

verification requirements,

authority questions,

dependency changes,

pending second review,

unresolved conflict,

or insufficient intervention time.

HOLD preserves uncertainty without manufacturing a conclusion.

The system should:

stop progression,

preserve state,

preserve evidence,

identify the unresolved condition,

and establish what must occur before reassessment.

HOLD means not yet.

It does not automatically mean no.

HOLD Must Actually Hold

A HOLD state is meaningless if downstream execution continues anyway.

When HOLD is entered:

commitment should stop,

dependent actions should stop where required,

new reliance should be prevented,

the relevant state should be preserved,

and the route should remain suspended until declared release conditions are satisfied.

A system that records HOLD while continuing consequence-bearing execution has not implemented HOLD.

It has implemented a label.

4. DENY

The proposed consequence is not permitted to proceed.

DENY means the evidence and governing conditions support a determination that the proposed route should not execute.

The denial should preserve:

what was denied,

why,

under what authority,

on what evidence,

and whether reconsideration is possible.

DENY should not necessarily mean:

forever prohibited.

Some denials may be specific to:

the current evidence,

the current state,

the current scope,

the current authority,

or the current proposal.

If conditions materially change, a new route may require a new determination.

A denied route should not silently resurrect itself.

5. ESCALATE

The determination requires another authority or capability.

ESCALATE means the current human possesses enough standing to recognize that they should not carry the final consequence.

Reasons may include:

authority limits,

scope limits,

insufficient competence,

conflicting evidence,

high consequence,

legal or rights-based questions,

exception conditions,

conflict of interest,

or requirements for independent review.

Escalation is not indecision.

It can be an admissible determination.

Knowing that the decision belongs elsewhere is itself governance.

ESCALATION MUST PRESERVE THE ROUTE

When escalation occurs, the architecture should preserve:

the current state,

the evidence,

the unresolved question,

the reason for escalation,

the originating human determination,

and the authority being requested.

Execution should not continue simply because escalation takes time.

Where the decision requires resolution before consequence:

ESCALATE should carry an implicit HOLD.

6. CORRECT

A correctable defect exists before progression may continue.

Sometimes the problem is identifiable and repairable.

Examples include:

incorrect evidence,

wrong configuration,

invalid input,

expired authority,

missing record,

scope mismatch,

or a known technical defect.

The appropriate state may be:

CORRECT

CORRECT means:

identify the defect,

preserve the pre-correction state,

perform the bounded correction,

record what changed,

then re-establish the required evidence before reconsidering progression.

Correction should not silently convert into approval.

Correct first. Revalidate second. Determine again.

The Correction Boundary

This is especially important because intervention itself changes reality.

Once the human corrects something, the original evidence may no longer represent the current state.

Therefore:

Intervention creates a New Reality.

That New Reality may require:

new evidence,

new continuity,

new diagnosis,

and a new determination.

The person should not rely automatically on the pre-correction determination.

7. OUTSIDE SCOPE

The human does not possess standing to determine this consequence.

This state protects both the person and the system.

A reviewer may understand the issue perfectly and still lack authority to determine it.

A technical expert may identify the correct action but lack organizational authority.

An executive may possess broad authority while lacking professional standing for a specialized determination.

The appropriate response is:

OUTSIDE SCOPE

This is not failure.

It prevents authority from expanding merely because the person is available.

The route should identify the appropriate authority or enter escalation.

OUTSIDE SCOPE vs. ESCALATE

These states are related but distinct.

OUTSIDE SCOPE

describes the standing of the current participant.

“This determination is not mine to make.”

ESCALATE

describes the route transition.

“This condition must be routed to another authority.”

An OUTSIDE SCOPE determination will often produce ESCALATE.

But preserving both states tells us why authority moved.

8. NOT ADMISSIBLE

The conditions required for consequence-bearing determination cannot presently be established.

This is stronger than HOLD.

HOLD says:

More evidence or resolution may permit reconsideration.

NOT ADMISSIBLE says:

The required admissibility conditions themselves are not established for this route.

Possible reasons include:

reality cannot be reliably established,

evidence integrity is materially compromised,

continuity cannot be restored,

authority cannot be validated,

the intervention window has collapsed,

human standing cannot be established,

verification is impossible,

or the route itself lacks sufficient governance.

NOT ADMISSIBLE prevents the system from pretending that uncertainty is merely an ordinary delay.

NOT ADMISSIBLE Is Not Necessarily Permanent

A route may become admissible later if the underlying defect is genuinely resolved.

But that requires a new governed state.

The architecture should not simply remove the NOT ADMISSIBLE label and resume.

It should establish:

what changed,

what evidence proves the change,

whether continuity has been restored,

and why the new route now possesses sufficient standing.

States Are Not Feelings

Human Intervention States are not:

“I feel comfortable.”

“I don't like this.”

“This seems okay.”

“I think we should probably proceed.”

They are governed determinations.

Each state should be supported by:

evidence,

standing,

authority,

scope,

and declared decision criteria.

Human judgment may be part of the determination where appropriate.

Unsupported preference should not substitute for governance.

States Should Be Machine-Readable

A mature AHIA implementation should not preserve these states only in prose.

They should become structured governance objects.

For example:

Determination State: HOLD

Reason: Evidence continuity failure

Authority: Human Role H-17

Scope: Release R-482

Timestamp: Preserved

Commit Status: Blocked

Required Resolution: Re-establish current production state

Escalation: Not required

Revalidation Required: Yes

This allows the rest of the execution architecture to respond deterministically.

State Transitions Must Be Governed

A route may transition:

HOLD → ALLOW

or

HOLD → DENY

or

ESCALATE → CONTROLLED ALLOW

or

CORRECT → HOLD → ALLOW

or

NOT ADMISSIBLE → reconstructed route → new determination

But transitions require evidence.

The architecture should be able to answer:

What changed?

A human should not simply click a different button until the desired state appears.

No Silent State Override

If one human determines:

HOLD

and another person later enters:

ALLOW

the original determination should not disappear.

The record should preserve:

who issued HOLD,

why,

what evidence existed,

who changed the state,

what authority permitted the change,

what new evidence existed,

and why progression became admissible.

Governance evolution should preserve chronology.

Dissent Is a State-Relevant Artifact

Where multiple humans participate, disagreement may be important.

Suppose:

Reviewer A: DENY

Reviewer B: ALLOW

Authority C: CONTROLLED ALLOW

The final determination should not erase the preceding disagreement.

The record should preserve the dissent and the evidence supporting it.

This becomes especially important if later outcomes validate the minority concern.

Dissent may become future evidence.

Multi-Human Determination Classes

Some consequence routes may require more than one human.

AHIA can support:

SINGLE AUTHORITY

One human with sufficient standing determines.

DUAL AUTHORITY

Two independently qualified authorities must concur.

SECOND REVIEW

One determination requires independent verification.

SEPARATION OF DUTIES

The determiner cannot also perform another consequence-bearing role.

PANEL DETERMINATION

Multiple governed participants contribute to the final state.

The required class should be established before the consequence route begins.

Humans Should Not Override Governed Evidence Without Basis

This returns to a foundational TA-14 concern.

If timestamped evidence, multiple witnesses, preserved continuity, declared thresholds, and diagnostic criteria establish a condition, a human should not be able to erase that condition simply by clicking:

ALLOW

Human intervention is not a license to corrupt evidence.

Where a human departs from the evidence-supported determination, the architecture should require a governed basis for that departure.

Human authority does not create alternate reality.

Machine Determination and Human Intervention

Not every state must originate from a human.

A governed system may automatically enter:

HOLD

when evidence becomes stale.

NOT ADMISSIBLE

when authority cannot be validated.

CORRECT

when a known bounded remediation is required.

Or DENY

where explicit deterministic criteria prohibit progression.

The human may then enter only where the decision class requires human standing.

This preserves the TA-14 principle:

Do not insert human discretion where governed evidence can more reliably control the route.

Fail-Closed State Logic

Where a required condition cannot be established, the default should not silently become ALLOW.

A mature architecture should prefer:

unknown evidence → HOLD

invalid authority → NOT ADMISSIBLE

outside authority → OUTSIDE SCOPE / ESCALATE

correctable defect → CORRECT

prohibited condition → DENY

bounded acceptable condition → CONTROLLED ALLOW

fully supported progression → ALLOW

This is fail-closed determination logic.

Admissible Non-Occurrence

Several AHIA states intentionally prevent execution:

HOLD

DENY

ESCALATE

CORRECT

OUTSIDE SCOPE

NOT ADMISSIBLE

When those states successfully prevent an unsupported consequence, the absence of execution is itself meaningful.

TA-14 calls this:

Admissible Non-Occurrence

The record should preserve:

what would have occurred,

what state prevented it,

what evidence supported the intervention,

and what happened afterward.

Successful restraint should not disappear simply because nothing happened.

The State Decision Map

SUFFICIENT BASIS + VALID STANDING + VALID AUTHORITY

→ ALLOW

SUFFICIENT BASIS + REQUIRED CONDITIONS

→ CONTROLLED ALLOW

TEMPORARILY INSUFFICIENT BASIS

→ HOLD

EVIDENCE SUPPORTS PROHIBITION

→ DENY

ANOTHER AUTHORITY IS REQUIRED

→ ESCALATE

CORRECTABLE DEFECT EXISTS

→ CORRECT

CURRENT HUMAN LACKS SCOPE

→ OUTSIDE SCOPE

REQUIRED ADMISSIBILITY CONDITIONS CANNOT BE ESTABLISHED

→ NOT ADMISSIBLE

The state should follow the governed condition.

The desired business outcome should not determine the state.

The Human Intervention State Principle

The purpose of AHIA is not to give humans more buttons.

It is to give consequence-bearing systems more truthful governance states.

Sometimes the truth is:

YES.

Sometimes:

YES, BUT ONLY UNDER THESE CONDITIONS.

Sometimes:

NOT YET.

Sometimes:

NO.

Sometimes:

SOMEONE ELSE MUST DETERMINE.

Sometimes:

FIX THIS FIRST.

Sometimes:

THIS IS NOT MY AUTHORITY.

And sometimes:

THIS ROUTE IS NOT ADMISSIBLE.

A serious governance architecture must be capable of preserving all of them.

Next: AHI-0 Through AHI-5

Now that the intervention states are defined, the next question is:

How mature is the human-governance route itself?

The next page establishes the six Human Intervention Admissibility Levels, from mere human presence to route-complete human governance.

AHI-0 — Presence Only

AHI-1 — Recorded Human Review

AHI-2 — Bounded Human Intervention

AHI-3 — Admissible Human Intervention

AHI-4 — Pressure-Survivable Human Intervention

AHI-5 — Route-Complete Human Governance

[EXPLORE AHI-0 THROUGH AHI-5 →]

TA-14 Admissible Human Intervention Architecture

Do not force certainty where admissibility does not exist.

No admissible evidence. No admissible execution.