The inspector can be wrong.

That is what makes this a game. He knows what you tell him, what your papers say, and what he chooses to check.

What Jev actually does

Each reply first asks Jev to identify what you have already explained and any specific discrepancy. A second Jev call selects one action: question your story, check the seal, scan the car, open the cargo, refuse a bribe, take it, release you, or turn you away.

You can type anything. His spoken lines are authored responses attached to those actions. Follow-ups quote your own words and identify a specific concern. Each question is asked at most once; an unconvincing answer still counts as an answer. This prototype uses no text-generation model. The physical rules determine what a scan or search discovers; Jev decides whether to perform it.

What he cannot see

The hidden glasswing, your objective, and whether you selected the honest shipment never go into the model request. Both versions begin with the same observations. A real search or scan is what reveals their different contents.

A crossing you can inspect

After the encounter, “Decision receipts” downloads the actual requests, raw TypeSafe responses, both calls, measured request durations, token usage, and resulting actions. The displayed reply duration combines the two calls. Displayed cost uses $0.042 per million input tokens; it is model cost, not hosting cost. See the published findings from 100 scripted crossings; these are not a reliability or calibration benchmark.

One room, one conversation

The screen puts you opposite the inspector. Your reply and the item you hand over stay together. Briefing and transcript open on request. No countdown rewards typing speed, and model latency is measured without artificial replay acceleration.

The scene uses Rajdhani for titles, DM Sans for conversation, amber for actions, and petroleum teal for the room. Keyboard focus is amber; status also uses text. There is no automatic motion or audio. Optional low-volume ambient sound starts only on request.

Sources and originality

Dan’s reference was the border interview in Cyberpunk 2077’s Nomad prologue. The tension of documents and discretionary inspection also appears in Papers, Please. Last Exit has an original setting, character, artwork, dialogue, and rules; it is not a Cyberpunk mod or affiliated release.

Read the findings from 100 scripted crossings

The public demo has a shared $1 model-cost allowance and per-visitor limits. Replies go to TypeSafe. Sessions expire after four hours; expired transcripts are removed on the next new crossing. Keep your story fictional. No account or API key is needed to play.