SDK Documentation

Victor Evidence SDK

Victor Evidence integrates as a deterministic semantic admissibility layer between your AI system and downstream application logic.

⚡

Fast Setup

Download the Basic SDK, install one dependency, and evaluate evidence admissibility in minutes.

🔒

Evidence Gate

Return PASS / FAIL gating and retrieval strength metrics for Basic Plan integrations.

🧩

Model Agnostic

Works independently from your LLM, retrieval engine, embedding model, or database provider.

Basic Plan

Quick Start

Evaluation Access

The Basic SDK evaluates whether sufficient admissible evidence exists to support AI-generated answers and returns a deterministic gate result. This tier is designed for evaluation workflows, prototype integrations, and first-pass evidence validation.

from victor_ai_sdk_v1_0 import VictorAI

victor = VictorAI(
    victor_api_key="YOUR_API_KEY"
)

result = victor.validate(
    question="What are human rights?",
    answer=llm_response
)

print(result["gate"])
print(result["retrieval_pct"])

Downloads

Download the Basic SDK and example integrations. Advanced SDK versions are provisioned after subscription verification.

Download SDK

victor_ai_sdk_v1_0.py

LLM-Only Example

Validate direct LLM answers.

Corpus Example

Validate answers against corpus context.

Installation

Place victor_ai_sdk_v1_0.py in your project directory, then install the required dependency.

pip install requests

Authentication

Initialize the SDK with your Victor Evidence API key. API keys are provisioned through your registered account or subscription workflow.

from victor_ai_sdk_v1_0 import VictorAI

victor = VictorAI(
    victor_api_key="YOUR_API_KEY"
)

Evaluate Admissibility

Call validate() with the original user question and the AI-generated answer. The SDK returns an admissibility determination for application-side gating.

result = victor.validate(
    question="What are human rights?",
    answer=llm_response
)

if result["gate"] == "PASS":
    show_answer_to_user(llm_response)
else:
    request_additional_evidence()

Response Schema

The Basic Plan exposes the core gate result and retrieval strength metrics. Higher tiers expose selected evidence, unit-level indicators, and audit-ready trace exports.

{
  "gate": "PASS",
  "retrieval_pct": 79,
  "evidence_count": 14
}

Field	Type	Description
gate	string	PASS or FAIL decision for downstream application gating.
retrieval_pct	number	Retrieval strength score expressed as a percentage.
evidence_count	number	Count of Semantic Evidence Units evaluated during admissibility determination.

Understanding Admissibility Signals

Victor Evidence returns structured admissibility signals. These are not probabilistic confidence scores. They are deterministic gating diagnostics designed for application-level decision control.

PASS / FAIL Evidence Gate

Use this value to control whether an AI response is shown to the end user, escalated for review, or re-generated.

Retrieval Strength Score (%)

Indicates overall semantic alignment between the query and admissible units. Useful for monitoring drift, analytics, and quality tracking.

Evidence Count

Represents the number of Semantic Evidence Units evaluated during admissibility determination. Higher tiers provide visibility into which units contributed to the gate decision.

Unit-Level Diagnostics (Individual & Pro)

Enables structured inspection of admissible units, support indicators, and signal metrics for research and audit workflows.

Audit & Profile Controls (Pro)

Designed for institutional environments requiring configurable thresholds, trace logging, and validation documentation.

Examples

Use the examples to test Victor Evidence in common integration patterns.

LLM-Only Admissibility Evaluation

Use this example when your system generates an answer directly from an LLM.

Download example →

Corpus-Based Admissibility Evaluation

Use this example when your answer is generated from retrieved or corpus-based context.

Download example →

Architecture

Victor Evidence sits after answer generation and before downstream delivery. It does not replace your model, vector database, retriever, or application logic.

❓

User Question

→

🤖

LLM Response

→

✅

Evidence Gate

📊

Admissibility Result

→

🚀

Application Action

SDK Plans

Feature visibility depends on subscription tier. The Basic Plan is focused on evaluation access. Individual and Pro tiers expose deeper evidence and audit capabilities.

Basic

Evaluation Access

✓ PASS / FAIL evidence gate
✓ Retrieval strength score
✓ Evidence count reporting
✓ Basic SDK download

Individual

Research Access

✓ Selected admissible evidence units
✓ Evidence count reporting
✓ Aggregate Admissibility Metrics
✓ Structured validation summary

Private Access

Available to Individual Plan subscribers.

Pro

Organization Deployment

✓ Full per-unit signal trace
✓ MUE / DUE metrics
✓ Anchor validation reporting
✓ Audit-ready JSON export

Restricted Access

Pro & Business subscribers only.

Feature Matrix

Capability	Basic	Individual	Pro
PASS / FAIL Evidence Gate	✓	✓	✓
Retrieval Strength Score (%)	✓	✓	✓
Evidence Count	✓	✓	✓
Selected Evidence Text	—	✓	✓
Unit-Level Support Indicator	—	✓	✓
Per-Unit Signal Metrics (MUE, Relevance, Similarity, CI)	—	—	✓
Anchor Rule Diagnostics	—	—	✓
Admissibility Profile Selection	—	—	✓
Audit JSON Export	—	—	✓