A moderation pre-filter where people only see the edge cases.

SourcePOST /content

Expanso nodeone pipeline, 5 steps, deterministic

1. Receive
2. Shape
3. Ask Jev
4. Gate
5. Route

Jev judgment

• category
• needs_human

• allow
• review

1. Expanso, deterministic

   Same fixed rules every time, no model involved.

2. Jev, judgment

   The one step a rule cannot write.

3. Receivelines 11 to 16

   An HTTP server input accepts content on POST /content.

4. Shapelines 17 to 26

   Parses the POST body if it arrived as a string, then stamps received_at. It applies the same fixed rules every time, with no model involved; the timestamp is the one value that differs.

5. record + questionsanswers + confidence
6. Ask Jevlines 27 to 73

   Expanso packs the whole record and the typed questions into one request. Jev answers. If the call fails, a catch substitutes empty answers marked jev-unavailable, and the pipeline keeps going.

   • categorychoice
   • needs_humannoul
7. Gatelines 74 to 88

   One fixed rule over Jev’s answers allows content or queues it, and computes review_priority as one minus Jev’s confidence.

8. Routelines 89 to 106

   A switch output writes allowed content and the review queue to separate files.

name: jev-moderation
type: pipeline
description: Moderation pre-filter with Jev — every piece of user content gets a category judgment inline. Safe content auto-allows; everything else queues for humans, most-uncertain first.
namespace: production
labels:
  category: integrations
  pattern: ai-decision
  model: jev

config:
  input:
    http_server:
      address: "0.0.0.0:8080"
      path: /content
      allowed_verbs: ["POST"]

  pipeline:
    processors:
      - mapping: |
          # http_server already parses JSON bodies; only parse raw strings
          root = if this.type() == "string" { this.parse_json() } else { this }

      - mapping: |
          root = this
          root.received_at = now()

      # ── Ask Jev ──
      - mutation: |
          meta jev_start = timestamp_unix_milli()

      - branch:
          request_map: |
            root = {
              "state": this.string(),
              "model": "jev-latest",
              "questions": {
                "category": {
                  "type": "choice",
                  "instructions": "Classify this user-generated content.",
                  "criteria": {
                    "safe": "Benign content, fine to publish",
                    "spam": "Unsolicited promotion, scams, repetitive junk",
                    "hate": "Attacks or demeans a protected group",
                    "sexual": "Sexually explicit content",
                    "self_harm": "Encourages or describes self-harm",
                    "other": "Concerning but does not fit the above"
                  }
                },
                "needs_human": {
                  "type": "noul",
                  "instructions": "Does this content need a human moderator to review it before publishing?"
                }
              }
            }
          processors:
            - http:
                url: "${JEV_API_URL:https://api.typesafe.ai/v1/systemone}"
                verb: POST
                headers:
                  Content-Type: application/json
                  Authorization: "Bearer ${TYPESAFE_API_KEY}"
                timeout: 2s
                retries: 1
            - catch:
              - mapping: |
                  root = {"answers": {}, "model": "jev-unavailable"}
          result_map: |
            root.jev = {
              "answers": this.answers,
              "model": this.model.or("jev-latest"),
              "ms": timestamp_unix_milli() - metadata("jev_start")
            }

      # ── Confidence-gated cascade: humans only see the edge cases ──
      - mapping: |
          root = this
          let cat = this.jev.answers.category.choice.or("other")
          let conf = this.jev.answers.category.confidence.or(0)
          let needs_human = this.jev.answers.needs_human.noul.or(1)

          root.jev_decision = if $cat == "safe" && $conf >= 0.9 && $needs_human < 0.5 {
            "allow"
          } else {
            "review"
          }
          # Review queue is worked most-uncertain-first downstream
          root.review_priority = 1 - $conf

  output:
    broker:
      pattern: fan_out
      outputs:
        - stdout:
            codec: lines
        - switch:
            cases:
              - check: this.jev_decision == "allow"
                output:
                  file:
                    path: ./data/jev-moderation/allowed-${! now().ts_format("2006-01-02") }.jsonl
                    codec: lines
              - check: "true"
                output:
                  file:
                    path: ./data/jev-moderation/review-${! now().ts_format("2006-01-02") }.jsonl
                    codec: lines

106 lines. Copy and Download both give you the file byte for byte.