> ## Documentation Index
> Fetch the complete documentation index at: https://docs.instructorphp.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Decision model metadata

## Overview

Read reviewed Jev limits by the exact `(driver, wire model)` route. The two token
limits answer different planning questions: the whole request must fit its total
budget, while the shared state plus each individual question must fit the
state-and-question budget. They are planning ceilings, not automatic tokenizer
enforcement.

## Example

```php theme={null}
<?php
require 'examples/boot.php';

use Cognesy\Polyglot\Decision\Models\ModelCatalog;

$models = ModelCatalog::discover();
$jev = $models->find('typesafe', 'jev-1.13.0');

echo "JEV MODEL FACTS\n";
echo json_encode($jev->toArray(), JSON_PRETTY_PRINT | JSON_THROW_ON_ERROR)."\n\n";

// Approximate counts supplied by application-owned token estimation.
$stateTokens = 20_000;
$questionTokens = array_fill(0, 10, 5_000);
$wholeRequestFits = $jev->maxRequestTokens !== null
    && $stateTokens + array_sum($questionTokens) <= $jev->maxRequestTokens;
$everyPairFits = $jev->maxStateAndQuestionTokens !== null
    && $stateTokens + max($questionTokens) <= $jev->maxStateAndQuestionTokens;

echo 'Whole request fits: '.($wholeRequestFits ? 'yes' : 'no')."\n";
echo 'Every state + question pair fits: '.($everyPairFits ? 'yes' : 'no')."\n";
echo "Remediation: split the question batch; reducing one question is unnecessary.\n\n";

$unknown = $models->find('typesafe', 'jev-preview');
echo 'Unknown preview request limit: '.($unknown->maxRequestTokens ?? 'unknown')."\n";

assert($jev->maxRequestTokens === 64_000);
assert($jev->maxStateAndQuestionTokens === 32_000);
assert($wholeRequestFits === false);
assert($everyPairFits === true);
assert($unknown->maxRequestTokens === null);
?>
```