I Built a Procrastination Pattern Detector with 8 CBT Interventions — No AI, No ML

#python #api #productivity #discuss

You're not lazy. You're running one of 8 procrastination patterns, each with a specific CBT intervention that breaks the cycle.

I built a detector that identifies which patterns are keeping you stuck — using keyword matching, not machine learning.

The 8 Procrastination Patterns

Each pattern has a root cause, a behavioral signature, and an evidence-based CBT intervention:

Perfectionism Block — "If I can't do it perfectly, I won't start." Intervention: valued action over quality
Fear of Failure — "What if I try and fail?" Intervention: decatastrophize the failure outcome
Task Overwhelm — "There's too much to do." Intervention: break into 5-minute micro-steps
Waiting for Motivation — "I'll do it when I feel like it." Intervention: behavioral activation — action precedes motivation
Task Avoidance — "I'll do anything except the task." Intervention: identify and remove avoidance behavior
All-or-Nothing Approach — "I either do it all or none." Intervention: partial progress is still progress
Guilt-Procrastination Cycle — "I procrastinated → I feel guilty → I procrastinate more." Intervention: self-compassion breaks the cycle
Minimization Trap — "It's not that important." Intervention: reconnect with values and long-term cost

How It Works

The detector uses keyword-pattern matching — the same approach I used for my CBT Thought Analyzer:

PATTERNS = [
    {
        "id": "perfectionism_block",
        "keywords": ["perfect", "flawless", "not good enough", "has to be right"],
        "intervention": "You're waiting for perfect conditions that will never come. "
                        "CBT: valued action > quality. Start with a 5-min imperfect draft."
    },
    {
        "id": "fear_of_failure",
        "keywords": ["what if i fail", "fail", "mistake", "embarrass"],
        "intervention": "You're catastrophizing failure. CBT: What's the realistic "
                        "worst case? Can you survive it? What's the cost of NOT trying?"
    },
    # ... 6 more patterns
]

def detect_procrastination(text):
    detected = []
    text_lower = text.lower()
    for pattern in PATTERNS:
        matches = [kw for kw in pattern["keywords"] if kw in text_lower]
        if matches:
            detected.append({
                "pattern": pattern["id"],
                "confidence": len(matches) / len(pattern["keywords"]),
                "evidence": matches,
                "intervention": pattern["intervention"]
            })
    return sorted(detected, key=lambda x: x["confidence"], reverse=True)

No embeddings. No fine-tuning. No API calls to OpenAI. Just keyword matching against a curated list of behavioral signatures derived from CBT research.

Why Keyword Matching > ML for This Domain

Determinism: Same input → same output, every time. ML models give different results on different runs. For a clinical tool, nondeterminism is a bug.

Explainability: The matched keywords ARE the evidence. You can see exactly why the detector flagged "perfectionism block" — because you wrote "has to be perfect" and "not good enough." An ML model gives you a probability and a black box.

Zero Cost: No GPU, no API calls, no inference latency. The detector runs in <1ms.

Privacy: The text never leaves your device. No data sent to a server.

Known Output Space: There are exactly 8 patterns from decades of CBT research. ML might discover a 9th, but it would be a noise cluster, not a clinically validated pattern.

The API

The detector is deployed as a REST API on Render:

curl -X POST https://cbt-thought-analyzer.onrender.com/procrastination   -H "Content-Type: application/json"   -d '{"text": "I need to write my thesis but it has to be perfect and I keep waiting for the right time to start"}'

Response:

{
  "patterns_detected": [
    {
      "pattern": "perfectionism_block",
      "confidence": 0.5,
      "evidence": ["perfect"],
      "intervention": "You're waiting for perfect conditions..."
    },
    {
      "pattern": "waiting_for_motivation",
      "confidence": 0.33,
      "evidence": ["waiting for the right time"],
      "intervention": "Behavioral activation: action precedes motivation..."
    }
  ],
  "total_patterns": 2,
  "dominant_pattern": "perfectionism_block"
}

The Behavioral Activation Principle

The core insight behind all 8 interventions: action precedes motivation, not the other way around.

This is the central finding of behavioral activation research (Jacobson et al., 1996; Dimidjian et al., 2006). You don't wait until you feel like doing something — you do it, and the feeling follows.

Each intervention in the detector is a specific application of this principle to the pattern's root cause:

Perfectionism → start with an imperfect 5-min draft
Fear of failure → test the catastrophic prediction with a small experiment
Task overwhelm → identify the smallest possible first step
Waiting for motivation → act for 5 minutes, then reassess

Try It

The API is live. The detector is also submitted to aitopia.ai as an AI agent (in review) — if approved, you'll be able to invoke it from the marketplace.

import requests

response = requests.post(
    "https://cbt-thought-analyzer.onrender.com/procrastination",
    json={"text": "I keep putting off my taxes because I might make a mistake"}
)
print(response.json())

No AI. No ML. No NLP library. Just psychology.

DEV Community