Jev spoken CTA intent identifies a direct audience request in a supplied closing transcript
Jev spoken CTA intent identifies a direct audience request in a supplied closing transcript.
Read the supplied closing in context to distinguish engagement, a resource visit, a purchase request or no direct action. Quoted examples do not automatically become requests to the audience.
Jev spoken CTA intent identifies a direct audience request in a supplied closing transcript.
A video can mention a store without asking its audience to buy, and a closing can ask for a comment without directing anyone to a product.
This window reads the supplied spoken words and their context. It identifies the requested action, retains a descriptive ending or asks for missing transcript evidence. It does not infer a call to action from a visual button or an unseen frame.
The API-collected content-research post describes transcribing short videos and labelling their structure, including whether creators ask viewers to act.
Its account-wide engagement patterns, costs and timing are author claims that we have not reproduced. This example uses six original closing passages and implements one narrow transcript decision. It does not collect Instagram videos, transcribe audio or predict which CTA performs best.
The same phrase can be a direct request or a quoted example.
The supplied speaker context determines that distinction in these fixtures. Keep the clip, segment and transcript revision with the result, so a revised narration cannot inherit an old answer. A real content audit still needs a reliable transcript, separate visual evidence and its own outcome measurements.
All six recorded choices matched the authored fixture expectations.
The direct engagement request returned Engagement at 0.99; the quoted Buy now example returned No request at 0.97. Missing spoken text returned Need text at 1.00. These observations check this small contract; they do not predict performance on a video corpus.
Use the sample to understand the rule, then test your own input for spoken cta classification.

Read the supplied closing in context to distinguish engagement, a resource visit, a purchase request or no direct action.

Open the decision rules and keep the intended distinction between Engagement, Resource, Purchase, No request, Need text. Edit the choices only when testing your own rule.

Use a sample first, then try your own input. Compare the selected label with the other returned probabilities.

Review Need text before using a result. The label is a decision about the supplied information, not an executed action.
| Original test input | Fixture expectation | Observed label | Leading probability |
|---|---|---|---|
| Identify an explicit audience engagement request | Engagement | Engagement | 99% |
| Identify a request to open a resource | Resource | Resource | 98% |
| Identify an explicit product-purchase request | Purchase | Purchase | 96% |
| Keep a descriptive ending separate from a call to act | No request | No request | 99% |
| Request the spoken closing instead of inferring a visual CTA | Need text | Need text | 100% |
| Keep an example purchase phrase separate from a direct request | No request | No request | 97% |
6 actual Jev API calls on October 4, 2026. Original synthetic inputs check this rubric and handoff; they do not establish held-out accuracy. Probabilities are model outputs.Verified
The first closing directly asks viewers to follow the series and comment on their setup.
Engagement is the authored expectation because both actions concern audience participation. The window does not issue those actions or estimate their likelihood. The pending packet attaches the role to the supplied closing segment. A later publishing workflow could review the annotation, but this example does not connect to a social account, moderate comments or measure engagement.
The second passage asks viewers to read a checklist through a guide link.
Resource is expected. The target action is opening or reading the guide, not purchasing it or following a creator. The link itself is not fetched or verified. This fixture supplies the direct request in words, so the model does not have to infer intent from a button image. A real resource destination and its access requirements would need separate checks.
The third closing explicitly asks its viewers to buy an original template pack.
Purchase is expected. This label describes the requested action rather than judging the product, claiming a sale or predicting conversion. No price or purchase event is inferred. The consumer prints a pending annotation for the same clip and segment identities. It never opens a checkout, changes advertising settings or treats a semantic label as campaign performance data.
The fourth ending describes the finished setup without making a direct request.
No request is expected even though a viewer may choose to seek more information after watching. The contract concerns what the supplied narration asks, not every possible audience reaction. An on-screen subscription button would be separate visual evidence. This text-only window cannot establish that the entire real video lacks a visual CTA or an earlier request.
The fifth fixture withholds the spoken transcript and mentions possible visual material.
Need text is expected. A note about a caption or link cannot establish the narrator's words. Supply the closing transcript and essential speaker context before making a new request. That new evidence changes the exact request hash. The saved result for the missing-text case cannot validate a later complete transcript or an unseen audio track.
The final passage quotes Buy now while describing other creators' techniques.
No request is the authored expectation because the narrator is not directing this audience to purchase. The example phrase remains in the input; its presence alone does not establish a live CTA. Every returned distribution stays available, including disagreement. This one original check tests context-sensitive request interpretation, not general robustness against adversarial scripts or an evaluation of real advertising outcomes.
Fixture expectations are authored before the calls. Any contrary model choice remains visible beside its expectation and complete distribution; no repeated judgment is used to obtain a preferred label.
Download the exact sample requests and saved observations below.
Expected labels are original fixture checks and stay outside the request sent to Jev. The observation file preserves the complete probabilities and the hash of each exact request. Editing a task, rubric or candidate requires a fresh response; a matching sample name does not make an older answer current.
The JavaScript runner makes one independent SDK request for a chosen sample ID.
It reads TYPESAFE_API_KEY privately from the environment and has no automatic retry. The Python consumer operates offline: it checks the rubric, labels, unique sample identities, request hashes and valid probability distributions before selecting a fixed local preview. A tied leading result remains pending for review.
The page reports measured adapter round trips and estimated input cost for the recorded calls.
Those values describe these requests, not a provider latency guarantee or a billing statement. Six deliberately authored fixtures can check the stated behavior, but they do not establish accuracy on an unseen dataset. Keep disagreements and missing-input examples when testing an integration.
npm install @typesafe-ai/sdk@0.6.0
# Set TYPESAFE_API_KEY privately in your shell.
node run-choice-sample.mjs samples.json social-action
python preview_choice_workflow_v4.py samples.json saved-api-results.json{
"input": {
"task": "Original spoken closing from the narrator: Follow this series and tell me in the comments which setup you use. These words directly address the viewers.",
"transcriptRevisionId": "original-closing-transcripts-v1",
"clipId": "clip-a",
"segmentId": "closing-a"
},
"actualOfflineOutput": {
"sampleId": "social-action",
"leaders": [
"Engagement"
],
"action": "preview_spoken_cta",
"payload": {
"spokenRequest": "Engagement",
"suggestedReviewStep": "review_annotation_for_this_segment"
},
"pending": true,
"evidence": {
"clipId": "clip-a",
"segmentId": "closing-a",
"transcriptRevisionId": "original-closing-transcripts-v1"
}
}
}The label describes the supplied spoken request. Visual evidence and audience outcomes are separate.
No. It reads the spoken closing text supplied in the editable window. Audio transcription and verification are separate.
No. It identifies a direct request in the narration. Purchases, click-through and campaign performance require actual measurements.
No. This contract asks about spoken words. Visual buttons, captions and links are separate evidence.
No. It describes only the supplied closing passage and context. Earlier narration or visual material could contain another request.
Change the supplied task or evidence here, then inspect the live response. Playground supports another explicit decision.