> For the complete documentation index, see [llms.txt](https://docs.neurolabai.io/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.neurolabai.io/understand-results/evidence-states.md).

# Evidence States

Neurolab places multiple kinds of scientific material in one report. The most important interpretation rule is to keep their evidence states separate.

## The six states

### 1. Model-ranked prediction

A learned model or route component ranks labels or values. Examples include top-K cognition, mechanism, usage, side-effect, and ATC outputs.

**What it supports:** prioritization under a named model and route.

**What it does not support by itself:** calibrated probability, causal truth, clinical effect, or literature validation.

### 2. Source-backed report record

A governed source or knowledge-graph artifact supplies a record with provenance, scope, and surfaceability metadata. Examples can include regulatory-label, binding, pathway, or literature-derived records.

**What it supports:** the named source made or contains the represented assertion.

**What it does not support by itself:** every prediction near the record, applicability to an individual, or a broader conclusion than the source establishes.

### 3. Symbolic or composed hypothesis

The system combines explicit rules, receptor-to-mechanism relations, circuit frameworks, analog retrieval, or other deterministic composition steps.

**What it supports:** an inspectable hypothesis under the stated inputs and composition method.

**What it does not support by itself:** independent experimental validation or a learned probability of correctness.

### 4. Reference-only context

A graph layer, atlas, dosage record, population/pathway overlay, or source path is included to orient the reader. It can be useful without being active support for the query-specific result.

**What it supports:** contextual exploration.

**What it does not support:** the claim that the query activated, traversed, or caused the displayed path.

### 5. Caller-provided assumption

The user supplies receptor values, affinities, actions, dose context, or other conditioning input.

**What it supports:** a conditional analysis—“if these inputs are assumed.”

**What it does not support:** that the assumptions were measured, predicted, or verified by Neurolab.

### 6. Missing, unavailable, or withheld

The system cannot produce or responsibly surface a value. This can appear as null, empty, omitted, unrecognized, unsupported, or abstained.

**What it supports:** a boundary on the current result.

**What it does not support:** permission to infer the missing value from a neighbor, another section, or the assistant.

## Evidence scope matters

Support can apply at different levels:

* **Field-specific:** a support link points to an exact report pointer or list item.
* **Section-level:** evidence is relevant to a report section but not every nested value.
* **Report-level:** material helps interpret the compound or route broadly.
* **Contextual:** graph or source material is adjacent or explanatory but not asserted as support.

The narrower scope should be preferred. Do not attach section-level support to an individual ranked item unless the artifact explicitly binds them.

## Anatomy of a support trace

```mermaid
flowchart LR
    Claim["Report field or ranked item"] --> Link["Support link"]
    Link --> Evidence["Evidence record"]
    Link --> Graph["Nodes · edges · paths"]
    Evidence --> Source["Source and citation"]
    Graph --> Source
    Source --> Policy["License and surfaceability"]
    Link --> Release["Model / KG / artifact identity"]
```

A graph path is a representation of typed relationships in an artifact. It is not a literal trace of attention, neural-network message passing, or biological causation unless a separate contract establishes that meaning.

## Source visibility is not source quality

`surfaceability` controls whether and how material can appear in a user-facing response. A source can be scientifically useful but restricted from raw redistribution. Conversely, direct visibility does not guarantee high scientific quality. Licensing, provenance, relevance, and evidentiary strength are separate dimensions.

## What to do when evidence is incomplete

1. Keep the prediction and the missing support state together.
2. Look for a source-backed record that directly addresses the question.
3. Inspect whether the evidence is field-specific or only contextual.
4. Record the gap as a gap.
5. Review primary literature or authoritative databases independently.

Next: [Scores, Reliability, and Abstention](/understand-results/scores-reliability-and-abstention.md)


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter:

```
GET https://docs.neurolabai.io/understand-results/evidence-states.md?ask=<question>&goal=<endgoal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.