If a human intervention carries consequence, the evidence should be able to prove why it was allowed to matter.

The Minimum Human Intervention Evidence Package — MHIEP is the evidentiary core of the TA-14 Admissible Human Intervention Architecture.

AHIA does not accept:

“A human reviewed it.”

as sufficient proof of governance.

It asks:

Who was the human?

What role were they performing?

What evidence did they actually receive?

Was that evidence connected to current reality?

Were they competent?

What authority did they possess?

What was their scope?

Could they meaningfully verify the condition?

Could they refuse?

What did they determine?

What exactly did that determination authorize?

Did execution remain faithful to it?

What happened afterward?

And can the complete intervention be reconstructed?

The MHIEP is designed to preserve the answers.

If the intervention cannot be reconstructed, its admissibility may not be provable.

Why an Evidence Package Is Necessary

Human decisions are often preserved as isolated events:

Approved by Jane Smith — 2:14 PM

That proves very little.

It does not establish what Jane saw.

It does not establish what reality existed.

It does not establish whether Jane possessed the required competence or authority.

It does not establish whether the evidence changed before execution.

It does not establish what Jane actually authorized.

And it does not establish whether the resulting execution remained inside that authorization.

The MHIEP converts the human intervention from an isolated approval event into a governed evidence object.

1. Intervention Identity

Every consequence-bearing intervention should have a unique identity.

The record should establish:

Intervention ID

Decision Route ID

Human Participant ID or Governed Role ID

Timestamp

System / Environment

Decision Class

Consequence Class

This allows the intervention to be connected to the larger execution route.

No orphan approvals.

2. Human Role

The evidence package should identify the actual role being performed.

Examples:

Observer

Reviewer

Diagnoser

Determiner

Authorizer

Intervener

Escalation Authority

Consequence Authority

Outcome Reviewer

The record should not rely solely on organizational title.

“Vice President” does not tell us what governance function the person performed.

Preserve the role in the route — not merely the title on the business card.

3. Human Standing

The package should establish whether Human Standing existed for the particular decision.

This may include evidence of:

competence,

qualification,

authority,

scope,

independence,

and decision eligibility.

The question is:

Why was this human permitted to carry this consequence?

Human Standing should be established before the consequence-bearing determination — not reconstructed afterward merely because something went wrong.

4. Competence Evidence

Where competence is required, the package should identify its basis.

Depending on the domain, that may include:

training,

credential,

qualification,

demonstrated proficiency,

experience,

simulation performance,

recency,

or another appropriate competence record.

The required evidence should match the consequence class.

Not every decision requires professional certification.

But where competence matters, it should not be assumed.

5. Competence Continuity

A credential obtained years ago does not necessarily establish current capability.

Where the human serves as a critical intervention control, the architecture may need evidence that competence remains operationally current.

Possible evidence includes:

training recency,

recent practice,

assessment results,

simulation history,

continuing qualification,

or successful intervention performance.

Historical competence does not automatically prove present competence.

6. Authority Evidence

The MHIEP should establish:

who granted authority,

what authority exists,

when it became valid,

whether it remains current,

what conditions limit it,

and whether another authority supersedes it.

Authority should be attributable.

Where authority changes during the route, the change should be preserved.

Authority valid yesterday is not proof of authority at commit.

7. Scope Evidence

Authority must be bounded.

The evidence package should identify:

what system is covered,

what action is permitted,

what consequence class is included,

what limits apply,

what thresholds apply,

and what remains outside scope.

A human should not be able to expand authority simply by exercising it.

8. Operative Reality

Before evaluating what the human decided, the record must preserve the reality to which that decision applied.

This is foundational TA-14 doctrine.

Where appropriate, the evidence may include:

timestamped video,

sensor data,

system state,

logs,

measurements,

photographs,

transaction state,

environmental conditions,

machine state,

or other primary evidence.

The objective is to establish:

What reality existed when the human determination was formed?

9. Evidence Presented to the Human

The package should distinguish between:

Evidence existing somewhere

and

Evidence actually available to the human.

The record should preserve what the person actually received or accessed.

That may include:

documents,

data,

video,

system outputs,

model recommendations,

diagnostic results,

witness records,

alerts,

measurements,

and known uncertainty.

A human cannot be credited with relying upon evidence they never saw.

10. Evidence Provenance

Evidence should carry enough provenance to determine:

where it came from,

when it was created,

who or what created it,

whether it was modified,

what system produced it,

and what reality it purports to represent.

Where evidence is derived from other evidence, that lineage should remain visible.

Evidence without provenance can become assertion.

11. Corroboration and Witnesses

Some consequence classes should not depend upon a single evidence source.

TA-14's development from diagnosis toward admissible determination placed particular importance on:

timestamped evidence

and

multiple corroborating witnesses.

A witness need not always be a person.

Depending on the architecture, independent witnesses may include:

separate sensors,

independent records,

multiple cameras,

distinct measurement systems,

separately controlled logs,

or qualified human witnesses.

The governing objective is:

Reduce the ability of one actor or one source to rewrite reality.

12. Evidence Continuity

The MHIEP should establish whether the evidence remained connected to operative reality through the relevant determination boundary.

Questions include:

Did state change?

Did a dependency change?

Did new evidence appear?

Did the environment change?

Did authority change?

Did the proposed consequence change?

If the answer is yes, revalidation may be required.

Correct evidence about an obsolete state can still support an inadmissible determination.

13. Diagnostic Basis

Where diagnosis precedes determination, the evidence package should preserve:

the baseline,

the threshold,

the observed condition,

the governing criteria,

the diagnostic method,

and the evidence supporting the diagnosis.

This is particularly important because diagnosis should not silently become discretionary opinion.

The route should show:

Evidence → Diagnostic Basis → Diagnosis

before consequence-bearing determination occurs.

14. Verification Feasibility

The package should establish whether meaningful verification was realistically possible.

Questions include:

How much evidence required review?

How complex was it?

How much time existed?

What verification tools were available?

Was independent checking possible?

Was the human relying primarily on an AI-generated summary?

Did the verification burden exceed realistic human capacity?

Evidence availability does not establish verification feasibility.

15. Operating Condition

The evidence package should preserve material conditions affecting the intervention.

These may include:

time pressure,

fatigue,

emergency conditions,

cognitive load,

simultaneous incidents,

procedural momentum,

organizational pressure,

or degraded system conditions.

Not every human decision requires intrusive monitoring.

The purpose is not to create a surveillance architecture.

The purpose is to preserve material conditions capable of changing the admissibility of the intervention.

16. Independence and Conflict

Where relevant, the record should identify material conflicts or dependencies.

Examples include:

financial interest,

personal interest,

organizational hierarchy,

performance incentives,

prior involvement,

or another condition capable of materially affecting independent determination.

Human beings are not automatically neutral because they are human.

Human judgment must not be treated as evidence of independence.

17. Refusal Capacity

The package should establish what the human could actually do.

Could they:

ALLOW

CONTROLLED ALLOW

HOLD

DENY

ESCALATE

CORRECT

DECLARE OUTSIDE SCOPE

DECLARE NOT ADMISSIBLE

Could execution remain paused?

Could another authority be reached?

Could the human prevent commitment?

A human who can approve but cannot realistically refuse does not possess a complete intervention pathway.

18. Intervention Window

The evidence package should preserve the timing relationship between:

human notification

review start

determination

binding

commit

and

execution

This allows later reviewers to determine whether meaningful intervention was still possible.

A determination entered after practical commitment is not equivalent to pre-consequence authorization.

19. Determination State

The actual human determination must be explicit.

It should not be reconstructed from vague comments.

The state should be one of the governed AHIA states where applicable:

ALLOW

CONTROLLED ALLOW

HOLD

DENY

ESCALATE

CORRECT

OUTSIDE SCOPE

NOT ADMISSIBLE

The record should also preserve the basis for the state.

20. Dissent and Uncertainty

The MHIEP should not erase uncertainty merely because a final determination was reached.

Where material, preserve:

known uncertainty,

conflicting evidence,

minority opinion,

reviewer disagreement,

limitations,

assumptions,

and unresolved conditions.

This matters because later evidence may show that the dissenting concern was important.

A clean final answer should not erase a messy evidentiary reality.

21. Binding Evidence

The package should establish how the human determination became attached to the specific consequence route.

This protects against a valid authorization being reused for:

another system,

another time,

another subject,

another action,

or another consequence.

Binding answers:

What exactly did this determination govern?

22. Commit Evidence

The record should identify when the route crossed the consequence boundary.

What event constituted commitment?

Could the human still intervene immediately before it?

Was authority still valid?

Was evidence still current?

Did any material condition change between determination and commit?

The commit record is essential because:

Determination at Time A does not automatically establish admissibility at Time B.

23. Execution Evidence

The package should preserve what actually executed.

Not merely what was requested.

Not merely what was approved.

Not merely what the system intended.

What actually happened?

This may require:

execution logs,

transaction records,

machine state,

video,

system output,

sensor evidence,

or other consequence-specific records.

24. Execution Fidelity

The system should compare:

What was authorized?

with

What executed?

If execution exceeded:

scope,

duration,

permissions,

conditions,

or declared consequence,

then valid human authorization cannot legitimize the excess.

The MHIEP should therefore preserve any divergence.

25. Outcome Evidence

Execution is not the end of the route.

The package should preserve what reality formed afterward.

Questions include:

Did the expected outcome occur?

Did unexpected consequence occur?

Was the outcome partial?

Did another system rely upon it?

Was corrective action required?

Did the intervention improve or degrade the system?

Governance must eventually return to reality.

26. Admissible Non-Occurrence

Sometimes nothing executes because governance worked.

The package should preserve:

what consequence was proposed,

what intervention state prevented it,

why,

who possessed standing,

what evidence supported the intervention,

and what happened afterward.

This is the evidence of:

Admissible Non-Occurrence.

A prevented consequence should not disappear from institutional memory merely because execution never occurred.

27. Future Reliance

Today's intervention may become tomorrow's precedent.

The MHIEP should therefore identify whether the outcome may be relied upon in future decisions.

Questions include:

Was the determination validated?

Was it exceptional?

Did new evidence emerge?

Should the rule change?

Should training change?

Should automation change?

Should authority change?

Should the intervention boundary move?

A successful outcome does not automatically make the underlying decision universally reusable.

28. Integrity of the Evidence Package

The MHIEP itself must be governed.

The architecture should protect against:

silent modification,

record substitution,

missing timestamps,

deleted dissent,

altered authority evidence,

or post-event reconstruction presented as contemporaneous evidence.

Depending on consequence class, this may require:

append-only records,

cryptographic integrity,

versioning,

access controls,

independent witnesses,

or other continuity mechanisms.

Evidence governance applies to the evidence of human governance.

The Minimum Package

Not every implementation needs identical fields.

But a consequence-bearing AHIA evidence package should ordinarily be capable of answering at least:

WHO?

Who participated?

IN WHAT ROLE?

What governance function did they perform?

WITH WHAT STANDING?

Why were they eligible to carry the decision?

SEEING WHAT?

What evidence actually reached them?

ABOUT WHAT REALITY?

What state did the evidence represent?

WITH WHAT CONTINUITY?

Was that state still operative?

UNDER WHAT AUTHORITY?

Who permitted the determination?

WITH WHAT SCOPE?

What were the limits?

UNDER WHAT CONDITIONS?

Could the person meaningfully reason, refuse, and escalate?

DETERMINING WHAT?

What state was entered?

BOUND TO WHAT?

What consequence route did the determination govern?

BEFORE WHAT COMMIT?

When did consequence become operational?

WHAT EXECUTED?

What actually happened?

WITH WHAT FIDELITY?

Did execution remain inside the determination?

WITH WHAT OUTCOME?

What new reality formed?

OR WHAT DID NOT HAPPEN?

What consequence was prevented?

Evidence Burden Should Follow Consequence

AHIA should not require the same evidence burden for every decision.

A low-consequence reversible action may require a lighter package.

A high-consequence irreversible action may require substantially stronger evidence.

Evidence burden may increase with:

severity

irreversibility

uncertainty

rights impact

scale

dependency

novelty

downstream reliance

and

difficulty of remediation

Greater consequence requires stronger admissibility evidence.

The Package Should Not Become Bureaucracy for Its Own Sake

The objective of MHIEP is not maximum paperwork.

It is minimum sufficient reconstruction.

Good implementation should automate evidence capture wherever possible.

Humans should not be forced to manually document information the system can preserve more accurately itself.

Where machines can reliably capture:

time,

identity,

state,

evidence lineage,

authority status,

binding,

commit,

execution,

and outcome,

they should.

This also reduces another source of human contamination:

retrospective narrative reconstruction.

Capture Evidence at the Moment It Exists

Do not wait until after failure to ask:

“What evidence did the reviewer see?”

“What authority did they have?”

“Was the state current?”

“What did they actually authorize?”

“Could they have stopped it?”

Those questions should already be answerable.

TA-14 therefore prefers:

Contemporaneous Evidence

over

Retrospective Explanation.

The farther reconstruction moves from the event, the greater the opportunity for:

memory failure,

self-protection,

bias,

narrative revision,

or institutional pressure

to alter the account.

The MHIEP Principle

A human intervention should not become trustworthy merely because a trustworthy person says it happened correctly.

The architecture should be capable of proving the route independently.

Preserve the human.

Preserve the evidence.

Preserve the authority.

Preserve the state.

Preserve the determination.

Preserve the consequence boundary.

Preserve what executed.

Preserve what followed.

Then allow the record to answer:

Was the human intervention admissible?

Next: When the Human Should Not Be in the Loop

Now that we have established the evidence required for admissible human intervention, we need to confront the opposite question.

When does inserting a human actually make governance worse?

The next page will formalize the TA-14 principle that emerged from the original diagnosis → determination problem:

Where governed evidence can establish the determination more reliably than discretionary human judgment, unnecessary human intervention should be removed from the consequence path.

That page will address:

Human Bias

Emotional Distortion

Greed & Incentives

Convenience & Laziness

Evidence Substitution

Timestamped Video

Multiple Independent Witnesses

Diagnosis-to-Determination Integrity

Deterministic Determination

Admissible Human-Free Routes

and the governing principle:

The objective is not maximum human involvement.

The objective is minimum inadmissible influence.

[EXPLORE WHEN THE HUMAN SHOULD NOT BE IN THE LOOP →]

TA-14 Admissible Human Intervention Architecture

If the intervention carries consequence, preserve enough evidence to prove why it was allowed to matter.

No admissible evidence. No admissible execution.