Implement CI: test, check lean, fuzz, bench - #100

Open

imlvts wants to merge 29 commits into

Adam-Vandervorst:masterfrom

imlvts:ci-self-hosted

Commits on Sep 15, 2026

Add self-hosted CI: tests and A/B benches on PRs

imlvts
and
claude
committed
bb1080c
CI bench: pass required-features, report progress to the PR while running

imlvts
and
claude
committed
7e3fa33
CI bench: comment once at start with the job link, once at the end with results

imlvts
and
claude
committed
dd8b799
CI bench: print each bench's compare table as soon as both sides have run

imlvts
and
claude
committed
ed83cc0
CI: run the bench job after the test job passes

imlvts
and
claude
committed
945e87d
CI: differential fuzz job as a regression gate against the base commit

imlvts
and
claude
committed
8d9c355
CI: rewrite bench_ab and fuzz_ab in Python

imlvts
and
claude
committed
1a4c285
CI: move to checkout@v5 and upload-artifact@v5, which run on Node 24

imlvts
and
claude
committed
15ec363
CI bench: post a per-bench summary to the PR instead of the full tables

imlvts
and
claude
committed
927309c
CI: run the bench job after the fuzz gate as well as the tests

imlvts
and
claude
committed
078e190
CI: upload-artifact@v7; v5 still runs on Node 20

imlvts
and
claude
committed
6d0dd42
CI fuzz: report new divergences as a warning annotation, not a failure

imlvts
and
claude
committed
42d5eeb
CI fuzz: shrink the first new divergences and emit Rust reproducers

imlvts
and
claude
committed
46c5445
Drop a committed __pycache__ and ignore them

imlvts
and
claude
committed
9c088b3
CI fuzz: show the Lean toolchain and build steps in the job log

imlvts
and
claude
committed
63604de
CI: provision rustup and elan per runner user when missing

imlvts
and
claude
committed
92d9619
CI bench: run bench pairs in parallel, one per core

imlvts
and
claude
committed
7def6e3
CI bench: build both sides at the same path and target dir

imlvts
and
claude
committed
33f40a2
CI: Lean job that builds the model and checks its proofs and guards

imlvts
and
claude
committed
04d1d1e
CI fuzz: post its own PR comment with new failure modes and reproducers

imlvts
and
claude
committed
9b0fbad
CI: run the fuzz job even when the tests fail

imlvts
and
claude
committed
4556c84
CI fuzz: list new divergences by kind, at most 10 lines

imlvts
and
claude
committed
a65e8a2
CI fuzz: post the PR comment only when there is something to report

imlvts
and
claude
committed
b8ee586
CI fuzz: explain the columns and show known-class counts that changed

imlvts
and
claude
committed
09d5aa5
Fuzz: count agreement over the inputs that actually ran

imlvts
and
claude
committed
5ec50ce
Fuzz: name the reason an operation was skipped

imlvts
and
claude
committed
1990888
Fuzz: classify the remaining divergence classes by shape

imlvts
and
claude
committed
6af7e80

Commits on Sep 16, 2026

CI: resolve the persistent cache dirs from $HOME, not a hardcoded /home/gh-runner

imlvts
and
claude
committed
15ac67d
CI bench: drop the `bench` runner label requirement

imlvts
and
claude
committed
8d36b68

Navigation Menu

Implement CI: test, check lean, fuzz, bench - #100

Implement CI: test, check lean, fuzz, bench#100
imlvts wants to merge 29 commits into
Adam-Vandervorst:masterfrom
imlvts:ci-self-hosted

Commits on Sep 15, 2026

Add self-hosted CI: tests and A/B benches on PRs

CI bench: pass required-features, report progress to the PR while running

CI bench: comment once at start with the job link, once at the end with results

CI bench: print each bench's compare table as soon as both sides have run

CI: run the bench job after the test job passes

CI: differential fuzz job as a regression gate against the base commit

CI: rewrite bench_ab and fuzz_ab in Python

CI: move to checkout@v5 and upload-artifact@v5, which run on Node 24

CI bench: post a per-bench summary to the PR instead of the full tables

CI: run the bench job after the fuzz gate as well as the tests

CI: upload-artifact@v7; v5 still runs on Node 20

CI fuzz: report new divergences as a warning annotation, not a failure

CI fuzz: shrink the first new divergences and emit Rust reproducers

Drop a committed pycache and ignore them

CI fuzz: show the Lean toolchain and build steps in the job log

CI: provision rustup and elan per runner user when missing

CI bench: run bench pairs in parallel, one per core

CI bench: build both sides at the same path and target dir

CI: Lean job that builds the model and checks its proofs and guards

CI fuzz: post its own PR comment with new failure modes and reproducers

CI: run the fuzz job even when the tests fail

CI fuzz: list new divergences by kind, at most 10 lines

CI fuzz: post the PR comment only when there is something to report

CI fuzz: explain the columns and show known-class counts that changed

Fuzz: count agreement over the inputs that actually ran

Fuzz: name the reason an operation was skipped

Fuzz: classify the remaining divergence classes by shape

Commits on Sep 16, 2026

CI: resolve the persistent cache dirs from $HOME, not a hardcoded /home/gh-runner

CI bench: drop the `bench` runner label requirement