Practice your understanding.
Explore changed behavior, risky paths and missing tests. Answer rehearsal questions. Let the coach show you what you can explain—and what you still need to learn.
The pull request accountability gate
Understand it.
Say it live.
Ship it.
Prove you know what you ship.
Practice when it helps. When you're ready, answer a focused set of questions on live video before the pull request can merge.
SlopProof / understanding required
Understanding before merge
Fast code still deserves a slow thought.
Tools can generate code. They cannot take responsibility. SlopProof creates a deliberate pause to understand a change before a shared project has to carry it.
One flow · different starting points
Practice when it helps. Prove when you're ready.
Contributors arrive with different experience and context. Practice gives everyone access to preparation without turning preparation into a hurdle. Experienced contributors and small patches can go directly to Proof.
Prove your understanding.
Open the challenge on your phone. Questions appear one by one when the timer starts. Explain the patch on live video in one continuous take.
Product walkthrough
Deep dive—or go straight to Proof.
Both paths are directly available from the check. Practice is optional, and the exact proof questions stay hidden until recording starts.
src/runtime/reload.ts3 learning goals
- state.publish(candidate)
+ const snapshot = state.capture()
+ state.publish(candidate)
+ try { await callback() }
+ catch (error) {
+ state.restore(snapshot)
+ throw error
+ }
+ const snapshot = state.capture()
+ state.publish(candidate)
+ try { await callback() }
+ catch (error) {
+ state.restore(snapshot)
+ throw error
+ }
Understanding coach
Atomic visibility
Practice question · never reused in live proof
Can a concurrent reader observe candidate state before the callback succeeds?
Trace publish → callback → restore. Name the exact interval in which another reader can enter.
learn the edge,
not the answer ↗
not the answer ↗
Live proof handoff
Scan when you're ready. The questions are still sealed.
- One-time mobile link
- Camera + microphone preflight
- 15 sec orientation
- 90 sec live answer
- No pause or retake
SLOP/PROOF LIVE
QUESTION 2/301:17
Why can another reader still observe candidate state before rollback?
One take · uploading encrypted · head 9f31c2e
Configurable multimodal review
A multimodal LLM evaluates. Repository policy decides the path.
A multimodal LLM can act as an automated judge, comparing the transcript, selected video frames, answer timing, patch context and rubric. Maintainers can enable automatic pass for calibrated cases—or require human review for every proof.
Live capture
A risk-sized question set, one take, bound to this PR revision.
Encrypted vault
A per-attempt key protects the original recording.
Extract
Transcript, timestamps and a few reasoned frames.
Multimodal LLM judge
Patch references, transcript, frames, rubric, gaps and contradictions.
Policy → outcome
Auto-pass can be enabled. Review can always remain required.
Private by architecture
Your video is locked down from capture to deletion.
A live proof may contain your face, voice and technical reasoning. SlopProof treats the recording as sensitive evidence—not ordinary product analytics.
Encrypted per attempt
Each live proof gets its own encryption key. Private object storage alone is never enough to play the recording back.
Need-to-see access
Only authorized maintainers can open evidence when review is required. Every access is short-lived and auditable.
Deleted on a short clock
Recordings disappear after 24 hours by default—and sooner after a clear pass. Storage lifecycle rules provide a second deletion backstop.
Bounded model accessSlopProof exposes only a temporary, job-scoped media reference. Deployment operators choose the model endpoint and are responsible for its processing and retention terms.
Hard limitsNo face recognitionNo gaze tracking or room scanNo public video or transcriptNo permanent evidence archive
Practice it. Prove it. Own it.
SlopProof turns understanding into a visible step in shipping software—first as learning, then as accountability.
See the flow again ↑