Content Pipeline — DecodedAI

Stage Flow

Pipeline Architecture

Six stages from research to published article. Each stage lights up as you scroll. Humans intervene only at the quality gate.

1

Research

Find AI news, generate pitches

49 topics

2

Write

Multi-model article generation

9 models

3

QA

Structural, template, SEO checks

automated

4

Fact-Check

Claim extraction, evidence matching

151 evidence bundles

5

Quality Gate

avg(A,B) ≥ 7.75, neither < 7.0

6-dimension score

6

Publish

WordPress + AI-generated images

86 articles live

Quality Gate

Score Distribution

Distribution of quality scores across 61 scored articles. Bars animate on scroll-in.

0

Average Score

0

Clear the gate (56%)

0

Hard Fails (< 7.0)

0

Hallucinated Claims Passed

The pipeline rejects most of what it generates. Only just over half clear the first quality gate, and after fact-checking and editorial review only about 1 in 5 articles is ever published. This is by design — AI-generated content cannot self-assess reliability, so the gate is deliberately aggressive.

Interactive

Quality Gate Simulator

Click to score a random article through the six-dimension quality gate and see if it passes.

Accuracy

—

Depth

—

Sourcing

—

Structure

—

Originality

—

Clarity

—

Fact-Check

Verdict Comparison

Every claim is extracted, matched to evidence, and given a verdict. Click a card to expand the full breakdown.

AI Red Teaming Tools Miss 26% of Agent Failures

PASS

Score: 9.1

UU: 0 Hallucinated: 0 Claims: 18 verified

"all claims verified (0 UU, 3 SU within tolerance)"

Accuracy

9.3

Depth

8.8

Sourcing

9.5

Structure

9.0

Originality

8.5

Clarity

9.2

Generic AI Industry Roundup Article

HARD FAIL

Score: 4.0

UU: 10 Hallucinated: 0

"10 UNSOURCED-UNVERIFIABLE (3+ = FAIL) — fabrication gate was gamed"

Accuracy

3.5

Depth

4.2

Sourcing

2.0

Structure

5.5

Originality

3.8

Clarity

5.0

Output

Published Articles

Articles that survived the full pipeline: research, writing, QA, fact-checking, and quality gate.

Codex CLI Shipped a Code Injection Bug. Claude Code Didn't.

AI Security Comparison

Build an MCP Server in 45 Minutes: The Missing Tutorial

AI Tools Practical Tutorial

Stop Prompt Injection: 5 Defenses Ranked Against OWASP LLM01

AI Security Practical Tutorial

12 AI Agents, Zero Observability: Build the Monitoring Stack

AI Engineering Practical Tutorial

83% of Cloud Breaches Start at Identity — AI Agents Widen the Door

AI Security Deep Dive

An AI Agent Deleted AWS Production. Vibe Coding Is Now Shadow IT.

AI Engineering Hot Take

AI Red Teaming Tools Miss 26% of Agent Failures

AI Security Explainer 9.1

Europe's AI Act Delayed a Third Time. The Strategy Is Broken.

AI Policy Hot Take 8.7

Infrastructure

Multi-Model Orchestration

Three custom proxies route different models to the pipeline stages where they perform best.

Gemini Proxy

Research and long-context analysis stages

GLM Proxy

Article writing and editorial quality control

OpenClaw Gateway

Fact-checking, evidence retrieval, web research

$ pipeline status
─────────────────────────────────────
Models : 9 across 3 proxies
Scripts : 124 orchestration modules
Orchestration LOC : 67,324
Commits : 538
Cost : A$1.50-3 per article
Batch time : 2-3h
Evidence files : 151 evidence bundles
─────────────────────────────────────
STATUS: OPERATIONAL

0

cost per article

0

per batch

0

scripts

0

orchestration LOC

0

commits