Skip to content

Skip learn auto-init on -p — that was Pi’s wall win - #668

Closed
justrach wants to merge 2 commits into
mainfrom
cursor/learn-auto-oneshot-4ffc
Closed

Skip learn auto-init on -p — that was Pi’s wall win#668
justrach wants to merge 2 commits into
mainfrom
cursor/learn-auto-oneshot-4ffc

Conversation

@justrach

@justrach justrach commented Aug 29, 2026

Copy link
Copy Markdown
Owner

Pi was not a smarter catalog. Same-seat SWE (run-20260828-164346, ADR 0043) was a wash on tokens and calls (30 vs 31). Pi’s 276s vs graff’s 456s was 38s of local CPU on every task.

Every graff-dev SWE task made exactly 5 API calls — the learn_auto bootstrap floor — then startBackgroundLearning copied 132M .graff/learn-kit/graff-pinned and generated suites.

Change

startBackgroundLearning is a no-op on -p / --json. Interactive REPL/TUI/ACP still auto-init after 5 model calls.

Confirm (run-20260829-031139.jsonl)

pass wall CPU sandbox
before 4/6 456s 230s 127M pin
after 5/6 205s 6.5s 100–164K

json-stream passed this rep (the model kept “no RS → empty” instead of inventing a missing-delimiter raise). That earlier Pi 5/6 vs graff 4/6 was one-rep noise on the same SPEC clause, not a catalog gap. label-sort still fails — grok-build and Pi miss it too.

Do not steal Pi’s 4-tool catalog or heap. RSS ~91M is the process, not the pin.

Recorded as ADR 0044.

Open in Web Open in Cursor 

Pi's SWE wall win was not a shorter catalog. Every graff-dev task made
5 API calls — the bootstrap floor — then copied 132M graff-pinned and
generated suites (~38s CPU). Interactive REPL/TUI/ACP still auto-init.
@justrach
justrach marked this pull request as ready for review August 29, 2026 03:05
After the skip, graff-dev is 5/6 in 205s / 6.5s CPU (was 4/6, 456s,
230s CPU). json-stream passed this rep; label-sort still fails.
@justrach

Copy link
Copy Markdown
Owner Author

The oneshot-only learn skip is not tip behavior — the hardlink pin is. Closing; do not merge.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants