The cleanup station
Desloppify
AI slop goes in, signal comes out. The lab's decontamination tooling, pointed at the growing mess: generated filler in datasets, docs, and content that everything downstream chokes on. Bring us the slop; keep the signal.
How it works
The sieve, explained
Slop detection is contamination detection wearing work clothes. Here is the deal, stated plainly.
Slop is contamination, and we hunt contamination
The same discipline that dropped 18,564 benchmark-contaminated documents from our own corpus points the other way too: detecting low-quality generated filler in datasets, docs, and content libraries before it poisons what trains on it.
Cleaned means receipted
A desloppify pass ships with the evidence: what was flagged, why, what was removed, and what survived. You can audit every cut, the same way you can audit our model claims.
Human judgment stays in the loop
Detectors propose, people dispose. Nothing gets deleted from your material on a classifier's word alone; borderline calls come back to you with reasons.
Before you submit
Don't upload anything here; the form starts the conversation. And fair warning in both directions: every submission to this site runs through automated screening for spam and malicious content before a human reads it. Bad actors get flagged in red. Genuine requests never notice it exists.
The recipe book
Loops, workflows, and graphs
The antidote to slop is a proven recipe. Contribute the agentic loops, training workflows, and pipeline graphs you've actually run, and they join an open, credited collection held to the same evidence rules as everything else here.
House rules
- It has to run: a repo or gist someone else can actually execute.
- Receipts required: what you tested it on and what happened, dips included.
- Credited by name, same as data contributors in the Distillery.
- Every submission passes automated screening before a human reviews it.
Accepted recipes get published with the lab's own loops and workflows as the collection grows, contributor named on every entry. Want to go deeper than a single recipe? OpenLabs is the standing bench.