How to submit feedback to the learning loop

Every slop learn submission feeds the NSED reinforcement loop on the server. The next scan from your account weights findings differently based on what you reported.

Report a false positive

slop learn "the rust_pub_no_doc warning on src/main.rs:42 is wrong — the public re-export already has docs on the source item"

The last poke is auto-attached as context (id + verdict + patch, up to 8 KB) so you don't have to paste the diff.

Skip auto-attach

slop learn --no-attach "the heuristic for X is too broad"

Report a missed slop

slop learn "missed: defensive try/except around imported function call in app.py — no chance it raises"

Tag a feedback batch

slop learn --project my-org/my-repo "the python_print_debug rule fires on legitimate CLI tools — gate on entrypoint detection"

The project field partitions your feedback so the learning loop can weight by project shape later.

Per-period cap

Each account is capped at 100 submissions per calendar month and 8 KB per submission. The CLI prints queued <uuid> (N/100) so you know where you stand.

What happens next

The submitted text + attached context lands in the server's LearnLog sled tree. The offline RL workflow consumes it out-of-band and tunes the per-account weighting. Expect 24–48 h before changes show up.