Skip to main content

​
press_safe tier

press_safe evaluates proto_score_v1 against a threshold calibrated at 0% FPR on both clean and hard-negative human tracks. Use it when a false positive carries reputational or legal consequences: royalty pool submissions, sync licensing, human-only platform ingestion.
ValueAction
"human"Issue certification
"suspicious"Reject — acoustic evidence of AI origin
nullInconclusive — route to expert review, do not certify

​
Flow

Track submitted
     │
     ▼
Analyze with HumanStandard
     │
     ▼
┌────────────────────────┐
│ tier_verdicts.press_safe│
└────────────────────────┘
     │
     ├── "human"      → ✅ Issue certification
     ├── "suspicious" → ❌ Reject / flag as AI-origin
     └── null         → ⚠️  Inconclusive → escalate to human expert review

​
Implementation

Python
import requests, os

def certify_track(audio_url: str) -> dict:
    result = requests.post(
        "https://app.jobsbyhumans.com/api/v1/analyze",
        headers={"Authorization": f"Bearer {os.environ['HS_API_KEY']}"},
        json={"url": audio_url}
    ).json()

    press_safe = (result.get("tier_verdicts") or {}).get("press_safe")
    proto_score = (result.get("scores") or {}).get("proto_score_v1")

    if press_safe == "human":
        return {
            "decision": "certified",
            "confidence": "press-safe",
            "proto_score": proto_score,
            "message": "Acoustic evidence consistent with human origin at press-safe confidence."
        }
    elif press_safe == "suspicious":
        return {
            "decision": "rejected",
            "confidence": "press-safe",
            "proto_score": proto_score,
            "origin": result.get("origin"),
            "message": "Acoustic evidence of AI-generated origin detected."
        }
    else:
        # null — inconclusive at press-safe level
        human_safe = (result.get("tier_verdicts") or {}).get("human_safe")
        return {
            "decision": "review",
            "confidence": "inconclusive",
            "human_safe_verdict": human_safe,
            "proto_score": proto_score,
            "message": "Insufficient acoustic evidence at press-safe threshold. Expert review required."
        }
TypeScript
interface CertificationResult {
  decision: "certified" | "rejected" | "review";
  confidence: string;
  proto_score: number | null;
  message: string;
}

async function certifyTrack(audioUrl: string): Promise<CertificationResult> {
  const result = await fetch("https://app.jobsbyhumans.com/api/v1/analyze", {
    method: "POST",
    headers: {
      "Authorization": `Bearer ${process.env.HS_API_KEY}`,
      "Content-Type": "application/json",
    },
    body: JSON.stringify({ url: audioUrl }),
  }).then(r => r.json());

  const pressSafe = result.tier_verdicts?.press_safe;
  const protoScore = result.scores?.proto_score_v1;

  if (pressSafe === "human") {
    return {
      decision: "certified",
      confidence: "press-safe",
      proto_score: protoScore,
      message: "Acoustic evidence consistent with human origin at press-safe confidence.",
    };
  } else if (pressSafe === "suspicious") {
    return {
      decision: "rejected",
      confidence: "press-safe",
      proto_score: protoScore,
      message: "Acoustic evidence of AI-generated origin detected.",
    };
  } else {
    return {
      decision: "review",
      confidence: "inconclusive",
      proto_score: protoScore,
      message: "Insufficient evidence at press-safe threshold. Expert review required.",
    };
  }
}

​
What to include in a certification report

When issuing a human-origin certificate, include:
FieldValue
confidence_tier"press-safe"
proto_score_v1e.g. 0.21
calibration_versione.g. "clairity-bench-2026-01-12"
processed_atISO timestamp
origin"human" (or null)
This creates an auditable chain: which version of the model made the call, at what threshold, and with what score.

​
Handling null press-safe (expert review)

When press-safe is null, there is not enough acoustic evidence to make a press-safe determination. Recommended actions:
  1. Check human_safe — if that’s also "human", the track falls below the press-safe threshold but shows no AI evidence at human-safe confidence
  2. Review risk_series to see which sections of the track scored elevated
  3. Route to a human expert with the full signal report
  4. Optionally: request the artist provide session files or DAW project for verification
Do not certify tracks where press_safe is null. Issue certification only on explicit "human" verdicts at this tier.