Plate / D-slop Rev. B. 2026.05

D-slop

A browser extension for reclaiming the pre-slop internet.

1. What it is

The web filled up with AI-generated text somewhere around 2023. Most of it sounds the same: confident, polished, and completely hollow.

D-slop is a browser extension that detects it and lets you choose what to do with it. Flag it with a colored border. Fold it out of view. Remove it from the page entirely.

The check runs locally. Nothing leaves your browser. No account, no telemetry, no tracking.

This is a free, open-source tool. A donation to humanity. Pre-slop internet, on demand.

2. How detection works

The extension scores each text block of 50 words or more against five signals. The individual scores combine into a confidence reading. Anything that clears the threshold gets flagged.

  1. Phrase matching. Known AI verbal tics tracked as a list: "delve into", "it's worth noting", "tapestry of", "as an AI language model", "in conclusion". Hundreds of entries, updated by an autonomous agent.
  2. Burstiness. AI prose is unnaturally uniform in sentence length. Real writing varies.
  3. Punctuation density. Overuse of em dashes, semicolons, and parallel comma structures characteristic of LLM output.
  4. List uniformity. Bulleted and numbered lists where every item lands within 10% of the mean length. Human lists don't do that.
  5. Conclusion markers. Structural tells: "In conclusion", "To summarize", standalone TL;DR sections.

Scoring is a pure function. Same text, same rules, same result every time. Not 100% reliable, and that's stated plainly. The goal is catching the obvious majority, not building a forensic tool.

3. Display modes

Three modes, selectable in the extension popup. The choice applies across every page.

  1. Highlight. A colored border and a confidence score badge appear on flagged blocks. All content stays visible. Default mode for new installs.
  2. Collapse. Flagged blocks fold to a single disclosure line: "likely AI-generated, click to reveal." The content is still there. You just choose whether to read it.
  3. Hidden. A full-page overlay fires before the page renders, showing how many blocks were flagged and a short countdown. A "Show anyway" button lets you through immediately if you want. The overlay clears on its own when the timer runs out.

A global toggle in the popup turns the extension fully off. In that state, zero DOM modification happens and pages render completely untouched.

4. How rules stay current

Detection phrase lists go stale as AI models evolve. D-slop solves this without requiring a store update for every rule change.

An autonomous agent (GitHub Actions and Claude) runs every week against a curated test corpus of AI-generated samples. It tunes phrase lists, signal weights, and detection thresholds. Any proposed change must clear a minimum accuracy check against the corpus before it ships. The extension fetches the latest rules once a day, so new rules reach your browser within 24 hours of any agent push.

All agent runs are logged to the public repository. The corpus is open. If you find a miss or a false positive, file a labeled GitHub Issue and the agent will incorporate it on the next run.

5. Privacy

No account. No telemetry. No analytics.

All text scoring runs inside your browser on your machine. The extension makes one outbound request: a small daily fetch of the rules file from a private endpoint. That is the complete data footprint.

Drawn 2026 / J. Fischer / Operator
Install D-slop