> For the complete documentation index, see [llms.txt](https://aura-9.gitbook.io/aura/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://aura-9.gitbook.io/aura/aura-core-functions/quickstart-1.md).
# Validate AI Model Performance
### Seamless validation of AI model performance you can prove
Validation is at the core of Aura’s protocol logic. Without trust in a model’s outputs, there can be no economic activity or adoption. Aura tackles this challenge by creating a multi-layered validation architecture that combines traditional AI evaluation metrics with on-chain verifiability.
At a base level, Aura supports standardized evaluation pipelines for different classes of models:
* Text generation and language models: Evaluated using BLEU, ROUGE, perplexity, and semantic coherence scores.
* Classification models: Validated with accuracy, precision, recall, F1 score, and confusion matrices.
* Reasoning models and agents: Subjected to logic tests, multi-step reasoning benchmarks, and goal-completion metrics.
These metrics are computed using synthetic tasks, adversarial prompts, and custom datasets to ensure robust generalization. Results are stored immutably, allowing anyone to audit a model’s performance history.
Aura also introduces optional cryptographic guarantees through what we call a *Proof-of-Performance* layer. This mechanism ensures that a model’s outputs were generated:
1. In real-time,
2. By the correct version of the model, and
3. Without tampering, human intervention, or off-chain spoofing.
This is achieved using zkTLS (zero-knowledge Transport Layer Security), which cryptographically proves that the communication session with the model occurred as claimed. Combined with secure routing logs (e.g., from Cloudflare or a decentralized gateway), Aura can create a cryptographic fingerprint of model behavior.
Furthermore, Aura enforces performance constraints such as:
* Latency SLAs
* Output length bounds
* Logical consistency or contradiction detection
These verification rules can be encoded into the deployment process, allowing both users and governance participants to define what constitutes acceptable behavior. Violations can trigger de-ranking, refunds, or revocation of deployment status.
By supporting both empirical benchmarking and cryptographic auditing, Aura ensures that models can be trusted not just in theory, but in real-world usage.
---
# 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://aura-9.gitbook.io/aura/aura-core-functions/quickstart-1.md?ask=&goal=
```
`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.