> ## 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.

# Responses

`Decision::get()` returns an `Answers` collection. `Decision::response()` returns the full `DecisionResponse` envelope. Both terminal methods use the same lazy execution path.

## Read answers by question ID

Use the typed accessor when the expected primitive is known:

```php theme={null}
$answers = $decision->get();

$probability = $answers->noul('needs_refund')->probability();
$route = $answers->choice('route')->value();
$urgency = $answers->score('urgency')->value();
// @doctest id="923d"
```

`noul()`, `choice()`, and `score()` reject a missing ID and reject an answer of the wrong primitive. `answer($id)` returns the `NoulAnswer|ChoiceAnswer|ScoreAnswer` union when generic processing is intentional. Collections also expose `all()`, `count()`, and `toArray()`.

## Interpret each result

### NoulAnswer

`NoulAnswer::probability()` is the probability of yes. There is no separate confidence property. Treat values near `0.5` as uncertainty and choose application thresholds based on the cost of false positives and false negatives.

### ChoiceAnswer

```php theme={null}
$answer = $answers->choice('route');

$winner = $answer->value();
$confidence = $answer->confidence();
$billingProbability = $answer->probabilities()->probability('billing');
$highestProbability = $answer->probabilities()->highest();
$optionIds = $answer->probabilities()->ids();
// @doctest id="39b0"
```

`value()` is always one of the option IDs supplied by the request. `probabilities()` contains the full normalized distribution over those exact IDs. Confidence describes how concentrated the judgment is; it is not a guarantee that the selected option is correct.

### ScoreAnswer

```php theme={null}
$answer = $answers->score('urgency');

$position = $answer->value();
$confidence = $answer->confidence();
$levelTwoProbability = $answer->probabilities()->at(2);
$expected = $answer->probabilities()->expectedValue();
$levelTwoDescription = $answer->legend()->at(2)->value();
// @doctest id="307a"
```

`value()` is the probability-weighted index across the ordered levels and therefore may be fractional. `probabilities()` is indexed from `0`; `legend()` preserves the corresponding level definitions. Polyglot validates that the distribution, expected value, and legend agree with the original question.

## Use the response envelope

```php theme={null}
$response = $decision->response();

$answers = $response->answers();
$model = $response->model();
$inputTokens = $response->usage()->inputTokens();
$outputTokens = $response->usage()->outputTokens();
$totalTokens = $response->usage()->totalTokens();
$providerRequestId = $response->providerRequestId()->toString();
$httpResponse = $response->responseData();
// @doctest id="ceae"
```

Token counts are nullable because a provider may omit usage. `totalTokens()` is available only when both input and output counts are present. `responseData()` exposes the underlying normalized HTTP response for diagnostics; keep business logic on typed answers instead of provider bodies.

## Lazy and memoized execution

`create()` returns `PendingDecision` without making a network call. The first `get()` or `response()` executes it, and later calls return the memoized result. Terminal failures are memoized too.

```php theme={null}
$pending = $decision->create();

$request = $pending->request();
$executionId = $pending->executionId();
$answers = $pending->get();
$sameResponse = $pending->response();
// @doctest id="4c5d"
```

There is no Decision stream. A structured decision is returned as one validated result.