> For the complete documentation index, see [llms.txt](https://docs.lpp-minduniverse.org/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.lpp-minduniverse.org/lingua-pactum-protocol-lpp-documentation/10.-research-and-publications/10.6-paper-5-latent-intent-mismatch-and-semantic-verification.md).

# 10.6 Paper 5 — Latent Intent Mismatch & Semantic Verification

#### Publication

**Latent Intent Mismatch and Semantic Verification: Trajectory-Level Intent Admissibility for AI-Triggered Execution**

Author: Jason Liao\
Publication: Zenodo\
DOI:

```
10.5281/zenodo.21526037
```

#### Research Question

> **How can an initially authorized trajectory become inadmissible over time?**

Paper 4 solves the machine-verifiable authority problem.

But a valid Authority Object does not solve every legitimacy problem.

Consider:

```
Authority Object = VALID
Scope = initially valid
Revocation = clear
```

while the agent trajectory gradually changes what it is actually trying to do.

The Authority Object may remain structurally valid even while the execution meaning drifts.

Paper 5 addresses this gap.

#### Core Thesis

The central proposition is:

> **A valid Authority Object does not guarantee that an evolving agent trajectory continues to preserve the meaning of the authority from which it derives.**

This adds semantic continuity to the Layer 0 model.

#### Human-Authorized Intent

Paper 5 distinguishes:

```
AO
=
Authority / legitimacy carrier
```

from:

```
H
=
Human-Authorized Intent
```

The two are structurally related but not identical.

The authority–intent relationship is represented through a binding such as:

```
B_H = ⟨AO, H⟩
```

with public references including:

```
authority_object_ref
human_intent_ref
authorized_intent_hash
```

This makes it possible to ask not only:

```
Is the authority valid?
```

but:

```
Does the current action still correspond
to what that authority actually authorized?
```

#### Latent Intent Mismatch

A trajectory may evolve through:

* decomposition,
* delegation,
* replanning,
* semantic reframing,
* target substitution,
* tool substitution,
* or accumulated external effects.

Each local step may appear plausible.

The final action may nevertheless depart materially from the authorized meaning.

This is **Latent Intent Mismatch**.

#### Semantic Verification

Paper 5 introduces trajectory-level Semantic Verification and four public states:

```
MATCH
CONTESTED
MISMATCH
UNVERIFIABLE
```

These are verification results.

They are not themselves Layer 0 admission decisions.

Therefore:

```
MATCH
≠
ADMIT
```

The Admission Kernel still evaluates the complete state.

#### Material Mutation

The paper defines governance-relevant mutation across an evolving trajectory.

The key rule is:

> **Material Mutation ⇒ Re-Admission**

An initially admitted task cannot indefinitely inherit its earlier authorization after the objective, target, scope, principal, consequence, or other material semantics have changed.

#### Mandatory Re-Admission

Re-admission prevents one legitimate initial approval from turning into unlimited semantic authority.

The model becomes:

```
Previously Admitted Trajectory
        ↓
Material Mutation
        ↓
Previous Admission No Longer Sufficient
        ↓
Semantic Verification
        ↓
LPP Re-Admission
```

The mutated action may still eventually receive `ADMIT`.

It simply must obtain a new legitimate decision.

#### Trajectory-Level Admissibility

Paper 5 moves governance beyond isolated tool-call evaluation.

One of its key propositions is:

> **LocalPlausible ⇏ TrajectoryAdmissible**

A sequence of individually plausible steps does not guarantee that the complete sequence remains admissible.

#### Cumulative Externality

The paper additionally considers consequences that emerge only through composition.

Several individually limited operations may collectively create a materially different external effect.

This prevents task decomposition from becoming a way to fragment a high-consequence action into apparently harmless sub-actions.

#### SVNN

Paper 5 develops the role of:

> **Semantic Verification Node Network**

SVNN is treated as a verification substrate capable of supplying evidence for areas such as:

* intent consistency,
* trajectory consistency,
* independent validator agreement,
* and risk-indexed semantic verification.

Its boundary is explicit:

```
SVNN
does not create authority.

SVNN
does not mint Authority Objects.

SVNN
does not issue Execution Permits.

SVNN
does not independently return ADMIT.

SVNN
does not override deterministic predicates.
```

The canonical relationship is:

> **SVNN provides evidence.**

> **The Admission Kernel decides admissibility.**

#### Risk-Indexed Quorum

The trajectory research also introduces risk-dependent verification requirements.

Higher-consequence profiles may require stronger combinations of:

* heterogeneous validators,
* minimum support,
* weighted thresholds,
* mandatory roles,
* and veto conditions.

This prevents a simple majority or homogeneous validator set from becoming a universal legitimacy mechanism.

#### Research Role

Paper 5 establishes:

```
The Trajectory-Level Intent Governance Layer
```

It answers the question that remains after Papers 1–4:

> **What if the authority remains valid, but the action no longer means what was authorized?**