REPOSITORY ANALYSIS

soupbase

Soupbase is a bilingual Chinese-English Turtle Soup game where Jev judges player questions and reconstructions, and the app checks structured Choice results and confidence to decide clearance.

E2 · source reviewedPublic material or source evidence was reviewed. A fixed-version source review requires checking a pinned revision; this is not a reproduction, performance endorsement, or safety guarantee.
Stars
4
Forks
1
License
MIT
Last commit
2026-09-23

What Jev does here

Jev returns a structured decision for the local program; consult the source for the exact decision policy.

Adds structured choices or scores to the workflow; performance and cost benefits have not been independently verified.

The cited model.ts and api.ts were read for provider confidence metadata and pending-turn lease checks. No game server or gateway inference was run; these source checks do not establish puzzle-solving accuracy or race-free production behavior.

What task does this project fit?

Host a lateral-thinking puzzle with bounded answers and a separately graded final guess.

Material analysed:

From input to action

  1. Build state from puzzle facts, the player's question and recent history.
  2. Ask gateway Choice questions for host response or fact coverage.
  3. Store a validated result only if the session's pending turn is still current.

Implementation details that matter

Native confidence is separate metadata

model.ts reads confidence from providerMetadata.typesafe, while answer choices/probabilities live in answers. Treating every gateway answer as if it carried native confidence would lose this distinction.

Inspect the supporting source ↗

Persistence guards against late writes

api.ts updates a completed turn only while the session's pending ID and lease still match. A delayed model response must not overwrite a newer session state.

Inspect the supporting source ↗

When to choose it

Useful for studying a narrow game workflow with explicit facts and turn state. It cannot freely explain a mystery or establish that a player's paraphrase is semantically equivalent.

How to inspect it yourself

Compare hostDecision/gradeGuess in model.ts with the pending_id/lease checks in api.ts. Test duplicate or delayed turns with a fake transport before enabling paid requests.

Limits of this review

  • No game server or live gateway call was run by this site.
  • Model judgments can misread the puzzle; provider failures become unknown results rather than solved turns.
Primary material used in this analysis