jevbooks

Technique patterns

Fourteen ways people actually use Jev in production. Jev itself decides which of them a README describes.

  1. Many questions, one request

    一个 state 多题平铺

    138 projects
    The question asked
    Does the text describe sending several independent questions about the same state in a single Jev request?
    Not for
    One question per request, or questions spread over separate calls.

    See all 138 →

  2. Speculative heads

    投机式头

    4 projects
    The question asked
    Does the text describe asking questions for every possible branch up front and consuming only the ones matching the chosen branch?
    Not for
    Asking follow-up questions only after the first answer arrives.

  3. Independent cross-check

    互不可见交叉题

    7 projects
    The question asked
    Does the text describe a second question that independently verifies the main answer, such as goal_done or stuck alongside the action choice?
    Not for
    A single question, or verification done by code rather than by another question.

  4. Cheap screen, then deep check

    便宜宽筛再窄查

    13 projects
    The question asked
    Does the text describe a cheap broad Jev pass whose positives are then sent to a slower, more expensive stage such as an LLM or a detailed chain?
    Not for
    A single-stage pipeline.

    See all 13 →

  5. Tiered thresholds

    概率分档

    27 projects
    The question asked
    Does the text describe two or more probability cut-offs mapping to different actions, such as act / flag / clear or deny / ask / allow?
    Not for
    A single yes-or-no threshold at 0.5.

    See all 27 →

  6. No-judgment band

    无判定区

    5 projects
    The question asked
    Does the text describe treating probabilities near 0.5 as 'unknown' that neither pass nor fail?
    Not for
    Reading 0.5 as a medium score.

  7. Enter / exit thresholds

    双阈值防抖

    4 projects
    The question asked
    Does the text describe different thresholds for entering and leaving a state, or feeding the previous decision back to prevent flip-flopping?
    Not for
    A single static threshold.

  8. Decision as code

    决策即代码

    13 projects
    The question asked
    Does the text describe the final decision being computed in code from Jev probabilities, with weights, journals, or replay of logged answers?
    Not for
    Acting directly on the raw model answer.

    See all 13 →

  9. Explicit fail-open / fail-closed

    失败模式旋钮

    22 projects
    The question asked
    Does the text state what happens when Jev errors or times out: allow, deny, escalate, circuit breaker?
    Not for
    No stated behaviour on error.

    See all 22 →

  10. Answer contract validation

    答案契约校验

    5 projects
    The question asked
    Does the text describe checking Jev's answer shape before acting, such as choice in candidates, probabilities summing to one, or discarding out-of-range values?
    Not for
    Trusting the response as-is.

  11. Rules compiled into questions

    规则即问题

    3 projects
    The question asked
    Does the text describe natural-language rules or rubrics being turned into typed Jev questions, possibly compiled once by a larger model?
    Not for
    Hand-written one-off questions.

  12. Code enumerates, model picks

    代码枚举候选，模型只选序号

    11 projects
    The question asked
    Does the text describe code listing the candidates such as DOM nodes, diff hunks, or files, and Jev choosing an index among them?
    Not for
    Asking the model to produce a selector, coordinate, or free-text answer.

    See all 11 →

  13. Minimal state

    只放要判的东西

    8 projects
    The question asked
    Does the text describe deliberately sending only the item under judgment, such as the last message or a single paragraph, rather than the whole history?
    Not for
    Sending full transcripts or documents.

  14. Question health measured

    题目体检

    23 projects
    The question asked
    Does the text describe measuring the questions themselves: fire rate, Brier score, precision on historical samples, or an A/B against another model?
    Not for
    Anecdotal claims of accuracy.

    See all 23 →