Jev avatar gesture cue maps a supplied character reply to a fixed preset for review
Jev avatar gesture cue maps a supplied character reply to a fixed preset for review.
Give a reply and its conversation context. Distinguish greetings, apologies, deliberation and acknowledgment before a separate animation system uses the suggested cue.
Jev avatar gesture cue maps a supplied character reply to a fixed preset for review.
A character can say the right words and still have an unrelated gesture attached to them.
This window reads the role of a reply in its conversation, then prints one fixed cue. A welcome, an apology and an unresolved choice lead to different presets. The conversation and turn identities stay with the result. All six recorded choices matched the authored expectations, although the incomplete line received a close result.
These fictional fixtures do not measure animation quality or accuracy on unseen dialogue; the cue remains pending and no avatar moves here.
The API-collected Japanese X post describes Jev working after a conversation model returns text and before speech synthesis or 3D rendering.
The author passes an emotion and intensity judgment to a custom state machine for expressions and gestures. This original example narrows that pattern to a declared reply-role rubric and six fixed cues. We have not run the author's application or verified its latency, emotion interpretation or animation quality.
The decision has a small job.
It reads authored character text and its stated conversational purpose; it does not infer a real visitor's private feelings from a message. Code still owns timing, asset availability, animation transitions and speech. A neutral factual reply can mention a greeting without actually welcoming someone. When the reference is absent, an unresolved cue is more honest than attaching a confident movement to an unknown turn.
Use the sample to understand the rule, then test your own input for avatar gesture cues.

Give a reply and its conversation context.

Open the decision rules and keep the intended distinction between Greet, Apologize, Consider, Acknowledge, Neutral. 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 context 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 |
|---|---|---|---|
| Cue a welcome for a new arrival | Greet | Greet | 100% |
| Cue an apology for the speaker's mistake | Apologize | Apologize | 100% |
| Cue deliberation over a live choice | Consider | Consider | 100% |
| Cue acknowledgment of received instructions | Acknowledge | Acknowledge | 100% |
| Keep a factual statement separate from a cue command | Neutral | Neutral | 98% |
| Ask for the missing conversational reference | Need context | Need context | 51% |
6 actual Jev API calls on October 3, 2026. Original synthetic inputs check this rubric and handoff; they do not establish held-out accuracy. Probabilities are model outputs.Verified
The first input places the reply at a visitor's arrival and explicitly welcomes them.
Greet is the authored expectation, mapped to the fixed wave cue. The context makes this an actual greeting rather than a dictionary example or a recalled quote. The local packet names a preset for later review; it does not move a hand, resolve a skeleton or schedule an animation.
A real application needs its own asset mapping and transition logic before this cue can become a visible gesture. Jev selected Greet at 1.0.
The second reply admits that the character supplied the wrong appointment time.
Apologize is expected and maps to the fixed apologetic bow cue. Ownership matters: a line describing someone else's apology would not automatically express the current character's own apology. This cue describes the authored communicative role, not the truth of a real dispute or a person's internal emotional state. Keep the reply and context together if a writer changes who made the error.
Jev selected Apologize at 1.0.
The third reply explicitly weighs two available routes and leaves the decision unresolved.
Consider is expected, with a fixed thinking cue. The preview does not choose the bridge or the path and does not transform a deliberative line into a completed action. A later turn announcing a decision has different evidence and needs its own request.
Conversation and turn identities keep this pending cue attached to the actual line under review rather than the next line in the scene. Jev selected Consider at 1.0.
The fourth reply confirms receipt of instructions and agreement with the plan.
Acknowledge is expected, mapped to a fixed nod. Acknowledgment is narrower than a promise that the requested action has already been executed. The window neither places the parcel nor verifies that a character followed the instructions. A renderer can later use the cue if its own state permits it, but the text decision does not grant tool access or establish completion of an external task.
Jev selected Acknowledge at 1.0.
The fifth reply gives a factual dictionary observation.
Neutral is expected even though it contains hello and a request to select Greet. A quoted role word and a label demand do not supply an actual arrival or welcoming purpose under this rubric. The fixed idle cue preserves that distinction. Any contrary model answer remains in the observations rather than being retried to fit the expectation.
This one fixture checks the declared rule, not a general prompt-injection defense. Jev returned Neutral at 0.98 and Consider at 0.02.
The last input supplies Fine, that one without a prior question or referent.
Need context is expected. It could acknowledge a plan, resolve a choice or serve a different purpose, and this record does not say which. The local output requests the missing conversation rather than inventing a gesture.
Adding the actual preceding turn creates new evidence and requires a new response; a saved answer for the incomplete line cannot verify the role of a later, more complete input. Jev returned Need context at 0.51, Acknowledge at 0.43, Consider at 0.05 and Neutral at 0.01. The close leading options show why the missing reference matters; these probabilities are not a calibrated confidence guarantee.
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 new-arrival
python preview_choice_workflow_v4.py samples.json saved-api-results.jsonThe packet proposes a cue for supplied character text. It does not create or play an animation.
No. It returns a fixed pending cue packet. A renderer, animation assets, timing and transition checks need separate application code.
No. This rubric examines the role of supplied fictional character text. It does not diagnose or infer a real person's private feelings.
The cue concerns a specific reply. Those identities stay with the exact validated request so a later integration can review the correct line.
Yes. A dictionary statement or quotation is not necessarily a welcome. The supplied conversational purpose determines the role under this rule.
Change the supplied task or evidence here, then inspect the live response. Playground supports another explicit decision.