System One ENGINE
All evidence records

first-party · synthetic test

Parable dialogue advice chose silence in every first-run case

Conservative silence is not evidence of better dialogue.

Reviewed 2026-09-20

System One and The Infinite Parable share an author. This is our development diagnostic, not independent reproduction.

Evidence confidence

low. The receipts establish this small development run. They do not establish improved pacing, useful positive selection or end-to-end savings.

What was observed

Choose one eligible recorded game reaction or silence.

Eight authored scenarios resolve installed cue text into the local dialogue API using typesafe-ai/jev through Vercel AI Gateway. Four explicit silence labels precede the run; four editorial cases have no objective label. The 0.8 threshold remains unchanged.

Baseline

Exact-completion and minimum-silence checks handle two cases without inference. Existing authored fact/heard guards can reject two other policy cases. No complete deterministic-versus-Jev listening comparison was run.

Finding

All four explicit silence expectations matched. The four editorial cases also returned silence, so this run does not demonstrate useful positive selection or a more natural experience.

  • 4 of 4 explicit silence expectations matched; 4 editorial cases require listening review.
  • 6 provider evaluations; 3,758 input and 180 output tokens.
  • Model-backed request wall time: p50 254 ms; p95 512 ms, nearest rank over six calls.
  • All 8 service results were silence.

What the result does not establish

Small private authored corpus; public receipts omit cue text and facts, limiting independent reproduction. No immutable provider model revision, listener ratings, whole-task savings or probability calibration. Six measured model requests are too few for latency generalization.

What we would test in System One

Our inference: use one choice over currently eligible recordings and reject stale results in the host. Compare against deterministic and always-silent baselines before claiming better pacing.

This recommendation is our interpretation of the study. Related research does not establish the quality of every Engine recipe.

narrator

Primary sources

Read this record in System One Bench. Source commits are pinned where available. Review dates describe our inspection, not the original run date.

Metric definitions and review method · Submit a correction or new result