Config Drift Detector

How Config Drift Detector works

Add your text in the First text field and press Classify locally. Everything happens in this browser tab: there is no account, and nothing you enter is sent to a server. The tool is built to run a small AI model on your own device, but no model files are published for it yet, so today it runs a transparent rule-based check instead. That check is described in full below, so you can judge exactly how much weight the result deserves.

Config Drift Detector compares two pieces of text: "First text" and "Second text". Both are lowercased and split into distinct words (letters, digits and the characters + # . -).

The match score is the number of distinct words the two inputs share, divided by the vocabulary size of the smaller input, times 100 and capped at 100%. The tool also lists up to five shared words, and up to four of its signal words ("config", "drift", "detector", "developer", "workflow", "risk", "priority", "summary", "quality", "intent") that are missing from the first input.

The rules are this site's own heuristic, written for a quick first pass. They are not a published clinical, statistical or industry method, and the same input always gives the same output.

Worked example

The built-in sample pair, which the tool shows when it first loads:

Production config: API_TIMEOUT=8s, RETRIES=3, REGION=us-east-1, FEATURE_BILLING_V2=true, LOG_LEVEL=warn. Staging config: API_TIMEOUT=3s, RETRIES=1, REGION=us-east-1, FEATURE_BILLING_V2=false, LOG_LEVEL=debug.

Match score

69%

Shared keywords

config, api, timeout, retries, region

Missing signals

drift, detector, developer, workflow

Edit the sample or replace it with your own text and run the check again to compare.

Limitations

• Keyword matching does not understand meaning. Negations ("no pain"), sarcasm and synonyms are scored the same as plain mentions or not at all.
• The signal list is short and generic, so a relevant input can still score low, and an off-topic input that repeats a signal word can score high.
• Results describe the input you gave, not the real-world situation behind it. Review them yourself before acting.

Good uses for a quick local check

• Triage developer workflow notes before deeper review
• Prepare consistent labels for repeat workflows
• Reduce manual sorting on small batches
• Keep sensitive drafts in the browser preview