Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
67 commits
Select commit Hold shift + click to select a range
89679c1
Bound the separated class at 146 and the global floor at 40
a9lim Oct 22, 2026
cad3eae
Kernel-check the class bound 146 and widen the separated class
a9lim Oct 22, 2026
8e683d0
Promote the sharing exclusions and source-cost theory to Lean modules
a9lim Oct 22, 2026
639375c
Record that the reader's unary run need not be periodic
a9lim Oct 22, 2026
0ae4e02
Archive the loop lane
a9lim Oct 22, 2026
9c5df6a
Exclude three shapes of zero-λ parent from a minimum reader
a9lim Oct 22, 2026
ccae78c
Exclude the whole-cell opener from a minimum reader below 46
a9lim Oct 22, 2026
5ca9401
Record what anchoring at a rigid ancestor recovers
a9lim Oct 22, 2026
cbc93d6
Price the 164's and the hoisted 165's reader families exactly
a9lim Oct 22, 2026
c774be0
Close the redex-free s-spine readers below 46
a9lim Oct 22, 2026
798e53c
Narrow the needed-binder case of the 162's reader family
a9lim Oct 22, 2026
5ca4710
Close six shared-let root families below 46
a9lim Oct 22, 2026
6c3e2cc
Bound the self-applied root readers at 37
a9lim Oct 22, 2026
0f12f22
Close the partial-selector spine λs. s (λb. b X) A B
a9lim Oct 22, 2026
8370893
Record the reader near misses and close a depth-0 shared window
a9lim Oct 22, 2026
581b731
Bound the β-normal readers at 43 by typing derivations
a9lim Oct 22, 2026
68ffd1d
Narrow the readers whose stream-λ body is a let
a9lim Oct 22, 2026
06bc21e
Record where the source j dies and why it does not price a reader
a9lim Oct 22, 2026
855bca8
A reader below 46 dispatches through at most one site
a9lim Oct 22, 2026
7cbb123
No minimum reader below 46 has a literal λ first handler
a9lim Oct 22, 2026
8c70f22
Every β-normal reader weighs at least 46
a9lim Oct 22, 2026
f0fef45
Price dispatchers shared through a let
a9lim Oct 22, 2026
560a613
Record the second review of the β-normal bound
a9lim Oct 22, 2026
6d7fa12
Correct how the β-normal bound applies to readers with lets
a9lim Oct 22, 2026
acd24e1
Restrict redexes in a non-λ first operand
a9lim Oct 22, 2026
5de7b53
Bound root lets at five uses below 46
a9lim Oct 22, 2026
a0bd26f
Kernel-check the foundations of the β-normal search
a9lim Oct 22, 2026
a11c597
No minimum reader below 46 has a j-free first operand
a9lim Oct 22, 2026
f387be8
No minimum reader below 46 has a syntactic dispatch site
a9lim Oct 22, 2026
746568b
No minimum reader below 46 is an s-spine
a9lim Oct 22, 2026
bfdd0d6
Rewrap the interpreter record's class paragraph
a9lim Oct 22, 2026
7bdae8a
Exclude every nonlinear stream-λ body let below 46
a9lim Oct 22, 2026
c87f107
Define the minimum reader once and record the s-spine audit
a9lim Oct 22, 2026
40917aa
Narrow the tight root-let profiles to 580 listed sources
a9lim Oct 22, 2026
103a6b6
Exclude every λ-rooted minimum reader below 46
a9lim Oct 22, 2026
fbd78c4
Close the shared-dispatcher skeleton without minimality
a9lim Oct 22, 2026
0f3bdc6
Record the audit of the λ-rooted exclusion
a9lim Oct 22, 2026
b1e1257
Close the one-operand shared-dispatcher skeleton
a9lim Oct 22, 2026
959bac2
Close the tightest root-let profiles
a9lim Oct 22, 2026
61a6cdc
Exclude every linear root let below 46
a9lim Oct 22, 2026
bdbf34b
Refute the last three root-let sources in Lean
a9lim Oct 22, 2026
8c1a0bf
Close stream-λ-headed root lets for the helpers reached
a9lim Oct 22, 2026
a2a36df
Narrow the four-use root lets to 658 listed sources
a9lim Oct 22, 2026
de3bbfc
Show pre-stream hand-offs expand without bound
a9lim Oct 22, 2026
23d8184
Kernel-check two simplifications of root-let hand-offs
a9lim Oct 22, 2026
e1ca95e
Refute rootnl's 658 listed root-let sources
a9lim Oct 22, 2026
7f12424
Close more root-let cells by narrowing with minimality filters
a9lim Oct 22, 2026
60eec92
Refute the last five listed root-let sources in Lean
a9lim Oct 22, 2026
186831d
Extend the hand-off simplifications and cut the over-cap root-let cells
a9lim Oct 22, 2026
2e3f33a
Exclude root lets with a literal-λ body and pending operands
a9lim Oct 22, 2026
c6ac440
Certify most open root-let cut-offs and withdraw an unsound certificate
a9lim Oct 22, 2026
c3462d0
Re-certify the round's self-feeding divergence certificates
a9lim Oct 22, 2026
41434ec
Exclude nested-redex root-let bodies below 46
a9lim Oct 22, 2026
49968f4
Park the coding-theorem decoder and hard symmetry-of-information side…
a9lim Oct 22, 2026
46f1ff2
Prove reader cutoff families and strengthen interpreter structure
a9lim Oct 23, 2026
a47bd4f
Prove source-copy cycles and close the pending reader ledger
a9lim Oct 23, 2026
8f29e53
Bound delayed reader prefixes and price their source escapes
a9lim Oct 23, 2026
fb182c9
Simplify Scott-cell activation to a 188-bit normal interpreter
a9lim Oct 23, 2026
bca81c0
Strengthen interpreter, reader, and NEXT proof boundaries
a9lim Oct 23, 2026
1d8a1de
Price one-sided NEXT sharing and bound live-source growth
a9lim Oct 23, 2026
7b5665a
Prove reader families and shared-helper source exclusions
a9lim Oct 23, 2026
56d238a
Prove four-lambda and 31-bit self-interpreter bounds
a9lim Oct 29, 2026
0edeeb1
Charge a step per lazy readback job
a9lim Nov 1, 2026
053d4fd
battery: run int4 programs under every weakening
a9lim Nov 5, 2026
5c1e79a
golf: withdraw the int4 rewrite; Tromp's 189 stays
a9lim Nov 5, 2026
c1f2ed9
Slim the verification and release protocol
a9lim Nov 7, 2026
703f50d
Release 3.0.2
a9lim Nov 7, 2026
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
The table of contents is too big for display.
Diff view
Diff view
  •  
  •  
  •  
The diff you're trying to view is too large. We only load the first 3000 changed files.
40 changes: 40 additions & 0 deletions .github/workflows/ait.yml
Original file line number Diff line number Diff line change
@@ -0,0 +1,40 @@
name: AIT fork

# The ref/AIT submodule is the a9lim/AIT fork: upstream tromp/AIT plus
# additions only. If a pin bump ever modifies an upstream file, a
# differential against that oracle can agree with blam for the wrong reason,
# so a change to the pin must keep the fork additive. Path-scoped to the pin.
on:
push:
branches: [main]
paths:
- "ref/AIT"
- ".gitmodules"
- ".github/workflows/ait.yml"
pull_request:
branches: [main]
paths:
- "ref/AIT"
- ".gitmodules"
- ".github/workflows/ait.yml"

permissions:
contents: read

jobs:
fork-additive:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v7
with:
submodules: true
- run: |
cd ref/AIT
git remote add upstream https://github.com/tromp/AIT.git
git fetch -q upstream master
viol=$(git diff --name-status upstream/master...HEAD | grep -v '^A' || true)
if [ -n "$viol" ]; then
echo "fork is not additive-only over upstream:"
echo "$viol"
exit 1
fi
136 changes: 22 additions & 114 deletions .github/workflows/ci.yml
Original file line number Diff line number Diff line change
Expand Up @@ -13,142 +13,50 @@ concurrency:
permissions:
contents: read

jobs:
fmt:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v7
- uses: dtolnay/rust-toolchain@stable
with:
components: rustfmt
- run: cargo fmt --all --check
# Release builds keep debug info for local profiling; CI needs none of it.
env:
CARGO_PROFILE_RELEASE_DEBUG: "0"

clippy:
jobs:
lint:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v7
- uses: dtolnay/rust-toolchain@stable
with:
components: clippy
components: rustfmt, clippy
- uses: Swatinem/rust-cache@v2
- run: cargo fmt --all --check
- run: cargo clippy --workspace --all-targets --all-features -- -D warnings

rustdoc:
# The published API's own docs: a broken intra-doc link, or a public
# item pointing at a private one, is a reader following a promise
# that is not kept. 2.0.0 ships this surface, so the gate is hard.
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v7
- uses: dtolnay/rust-toolchain@stable
- uses: Swatinem/rust-cache@v2
# The published API's docs: a broken intra-doc link is a promise the
# crate does not keep.
- run: cargo doc --all-features --no-deps
env:
RUSTDOCFLAGS: -D warnings

msrv:
# rust-version in Cargo.toml is a promise to anyone who depends on
# the crate; nothing enforces it but this job.
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v7
- uses: dtolnay/rust-toolchain@1.87
- uses: Swatinem/rust-cache@v2
- run: cargo check --all-features --all-targets

stale-invocations:
# One binary, `blam`, with subcommands. Every predecessor bin name
# (qcensus, certsearch, tracescan, …) is gone, so a script or current
# document that still invokes one is a command that cannot run. Only the
# vendored ref/AIT tree is outside this repository's command contract.
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v7
- name: no invocations of retired binaries
run: |
bins='census|qcensus|qradical|qcomplement|certsearch|certlean|certdiag|tracescan|slotsearch|oddminproto|solomonoff|adjudicate|normalize|selfint|qskeleton'
hits=$(grep -rnE "target/(release|debug)/($bins)\b|--bin +($bins)\b" \
--exclude-dir=.git --exclude-dir=target --exclude-dir=ref \
. || true)
if [ -n "$hits" ]; then
echo "stale invocations of retired binaries:"
echo "$hits"
exit 1
fi

test:
# --release is the repo's own verification bar (AGENTS.md): the cert
# battery pushes 196,848 provable halters through the sweep ladder and
# is built for release speed. tromp_vectors inlines its vectors, so the
# suite needs no ref/AIT checkout.
strategy:
fail-fast: false
matrix:
os: [ubuntu-latest, macos-latest]
runs-on: ${{ matrix.os }}
# --release because the certificate soundness battery is built for
# release speed. tromp_vectors inlines its vectors, so no ref/AIT
# checkout is needed.
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v7
- uses: dtolnay/rust-toolchain@stable
- uses: Swatinem/rust-cache@v2
# --all-features so the `lab` surface (research instruments, the
# untrusted certificate discovery layer, and the `blam` subcommands
# built on them) is compiled and tested too.
- run: cargo test --release --workspace --all-features
# Default features: the no-lab dispatcher arms actually run.
- run: cargo test --release
# The engine must also build with im-rc alone.
- run: cargo test --release --no-default-features
# The engine must build with im-rc alone.
- run: cargo check --lib --no-default-features

census-spot-check:
# The standing regression guarantee: halt counts are bit-identical to
# data/classical/census_table.txt under every engine change. Rows 4..32 cover the
# full ladder including loop32's row and run in under a second locally;
# timing columns are excluded, everything else must match exactly.
default-features:
# The census spot-check: fate columns, max|nf| and beta_total of rows
# 4..32 must match data/classical/census_table.txt exactly (escal is
# reported, not failed). It builds the default-feature binary, so the one
# behaviour default features add, the no-lab dispatcher arms, is tested
# here too.
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v7
- uses: dtolnay/rust-toolchain@stable
- uses: Swatinem/rust-cache@v2
- run: bash scripts/spot-check.sh 4 32

lean:
# Kernel-check both certificate ledgers and their exact wire identities;
# verify that generated exclusions and exact probability trims are current.
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v7
- uses: actions/cache@v6
with:
path: |
~/.elan
lean/.lake
key: lean-${{ runner.os }}-${{ hashFiles('lean/lean-toolchain', 'lean/lake-manifest.json') }}
- run: curl -sSfL https://raw.githubusercontent.com/leanprover/elan/master/elan-init.sh | sh -s -- -y --default-toolchain none
- run: echo "$HOME/.elan/bin" >> "$GITHUB_PATH"
- run: python3 tools/certificates/bb35_large.py --check
- run: lake build Certs BusyBeaver
working-directory: lean
- run: python3 tools/certificates/bb.py --check
- run: python3 tools/certificates/bounds.py | diff - data/classical/certificate_bounds.txt
- run: python3 tools/omega/test_accounting.py

fork-additive:
# The ref/AIT submodule is the a9lim/AIT fork: upstream tromp/AIT plus
# additions only. The conformance goldens stay byte-identical to
# upstream's — the moment the fork *modifies* an upstream file, the
# oracle-vs-ours boundary is gone and this job fails.
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v7
with:
submodules: true
- run: |
cd ref/AIT
git remote add upstream https://github.com/tromp/AIT.git
git fetch -q upstream master
viol=$(git diff --name-status upstream/master...HEAD | grep -v '^A' || true)
if [ -n "$viol" ]; then
echo "fork is not additive-only over upstream:"
echo "$viol"
exit 1
fi
- run: cargo test --release --bin blam lab_subcommands_are_named_either_way
52 changes: 52 additions & 0 deletions .github/workflows/lean.yml
Original file line number Diff line number Diff line change
@@ -0,0 +1,52 @@
name: Lean

# Kernel-check both certificate ledgers and their exact wire identities, and
# verify that generated exclusions and exact probability trims are current.
# Path-scoped: it runs when the proofs, the ledgers, the data they bound, or
# the tools that generate or read them change.
on:
push:
branches: [main]
paths:
- "lean/**"
- "data/certificates/**"
- "data/classical/**"
- "tools/certificates/**"
- "tools/omega/**"
- ".github/workflows/lean.yml"
pull_request:
branches: [main]
paths:
- "lean/**"
- "data/certificates/**"
- "data/classical/**"
- "tools/certificates/**"
- "tools/omega/**"
- ".github/workflows/lean.yml"

concurrency:
group: lean-${{ github.ref }}
cancel-in-progress: true

permissions:
contents: read

jobs:
lean:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v7
- uses: actions/cache@v6
with:
path: |
~/.elan
lean/.lake
key: lean-${{ runner.os }}-${{ hashFiles('lean/lean-toolchain', 'lean/lake-manifest.json') }}
- run: curl -sSfL https://raw.githubusercontent.com/leanprover/elan/master/elan-init.sh | sh -s -- -y --default-toolchain none
- run: echo "$HOME/.elan/bin" >> "$GITHUB_PATH"
- run: python3 tools/certificates/bb35_large.py --check
- run: lake build Certs BusyBeaver
working-directory: lean
- run: python3 tools/certificates/bb.py --check
- run: python3 tools/certificates/bounds.py | diff - data/classical/certificate_bounds.txt
- run: python3 tools/omega/test_accounting.py
69 changes: 35 additions & 34 deletions .github/workflows/release.yml
Original file line number Diff line number Diff line change
@@ -1,12 +1,12 @@
name: Release

# Triggered by the main push itself — crates.io trusted publishing
# rejects `workflow_run`-minted tokens (2026-08-08, measured: status
# 400, "does not support the workflow_run event trigger") — but the
# guard job's first step WAITS for the CI workflow on this exact SHA
# and fails unless it succeeded. Publication therefore still never
# outruns the verification bar — cargo publish alone compiles the
# package but runs none of the suite/spot-check/Lean gates.
# Triggered by the main push itself, because crates.io trusted publishing
# rejects `workflow_run`-minted tokens. The guard first resolves whether
# this push is a release at all (an untagged version in Cargo.toml); only
# then does it wait for the CI workflow on this exact SHA, and it fails
# unless CI succeeded. Publication never outruns CI; the path-scoped Lean,
# qALC and AIT workflows guard the repository's data, not the crate, and
# are not publication gates.
on:
push:
branches: [main]
Expand All @@ -23,10 +23,10 @@ permissions:
jobs:
# The gate. Version comes from Cargo.toml; the git tag is the
# completion marker, so the pipeline is dormant while main's version
# is already tagged — bumping the version is what arms it. The
# is already tagged, and bumping the version is what arms it. The
# crates.io probe is independent of the tag so a partial-failure run
# resumes cleanly: an already-published version skips straight to the
# tag + GitHub release. Every checkout pins the exact SHA CI passed.
# tag + GitHub release. Every checkout pins the pushed SHA.
guard:
runs-on: ubuntu-latest
outputs:
Expand All @@ -35,31 +35,6 @@ jobs:
crates_published: ${{ steps.state.outputs.crates_published }}
released: ${{ steps.state.outputs.released }}
steps:
- name: Wait for CI success on this exact SHA
env:
GH_TOKEN: ${{ github.token }}
run: |
# Poll the CI workflow's run for this SHA (REST, not `gh run
# list -c`, so a runner gh version bump cannot break the gate).
# Both workflows trigger on the same push, so the run record
# may not exist yet on early polls ("null null" from jq on the
# empty list — treated as still pending). CI takes ~3 minutes;
# 120 * 30 s = 1 h ceiling.
for i in $(seq 1 120); do
resp="$(gh api \
"repos/$GITHUB_REPOSITORY/actions/workflows/ci.yml/runs?head_sha=$GITHUB_SHA&per_page=1" \
-q '.workflow_runs[0] | "\(.status) \(.conclusion)"' 2>/dev/null || true)"
status="${resp%% *}"
conclusion="${resp##* }"
if [ "$status" = "completed" ]; then
[ "$conclusion" = "success" ] && exit 0
echo "CI concluded '$conclusion' on $GITHUB_SHA — refusing to release" >&2
exit 1
fi
sleep 30
done
echo "CI never concluded for $GITHUB_SHA" >&2
exit 1
- uses: actions/checkout@v7
with:
fetch-depth: 0
Expand Down Expand Up @@ -91,6 +66,32 @@ jobs:
else
echo "released=false" >> "$GITHUB_OUTPUT"
fi
- name: Wait for CI success on this exact SHA
if: steps.state.outputs.should_release == 'true'
env:
GH_TOKEN: ${{ github.token }}
run: |
# Poll the CI workflow's run for this SHA (REST, not `gh run
# list -c`, so a runner gh version bump cannot break the gate).
# Both workflows trigger on the same push, so the run record
# may not exist yet on early polls ("null null" from jq on the
# empty list — treated as still pending). CI takes ~3 minutes;
# 120 * 30 s = 1 h ceiling.
for i in $(seq 1 120); do
resp="$(gh api \
"repos/$GITHUB_REPOSITORY/actions/workflows/ci.yml/runs?head_sha=$GITHUB_SHA&per_page=1" \
-q '.workflow_runs[0] | "\(.status) \(.conclusion)"' 2>/dev/null || true)"
status="${resp%% *}"
conclusion="${resp##* }"
if [ "$status" = "completed" ]; then
[ "$conclusion" = "success" ] && exit 0
echo "CI concluded '$conclusion' on $GITHUB_SHA — refusing to release" >&2
exit 1
fi
sleep 30
done
echo "CI never concluded for $GITHUB_SHA" >&2
exit 1

# crates.io via trusted publishing: rust-lang/crates-io-auth-action
# exchanges the workflow's OIDC identity for a short-lived registry
Expand Down
Loading
Loading