Moves the Lean project to Lean 4.35.0 / Mathlib v4.35.0 to match the runner,
caches the Mathlib project on the runner host, and builds and keeps the Docker
image only after the Python and Lean jobs pass.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FYqwNMyPUHeTbFWg83GQq6
The workflow leaves an already-running lean-bench container alone, so a long
ingest is never cut off mid-run. That is still the right default, but "left
alone" and "up to date" looked identical from outside, and the gap only widens
with each build.
It cost a full debugging session. The container had been up for six hours on
the previous image, so a module added since simply was not in it; the CLI
rejected the new flags and every symptom pointed at the new code. The image
built from that same commit had the module all along.
The container is still not restarted. The build now compares its image id with
the one just built and, when they differ, emits a warning and puts a note at
the top of the job summary with both ids and the two ways to refresh.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ci5T3gWDhhmqDec3e1XtYZ
Two faults found by running the pipeline inside the Docker image for the first
time, both of which let a 25-page run finish "successfully" having stored nothing.
The image declares ENV DASHSCOPE_API_KEY= so the variable is visible to
docker inspect and overridable at docker run. That leaves it set-but-empty in
every container, and load_dotenv used os.environ.setdefault, so the empty string
won and the mounted .env was ignored. An empty value now counts as unset; a real
value still wins, because overriding on the command line is the point.
The consequence was worse than a missing key. Every page failed, the per-page
fallback kept the font-mangled embedded text -- exactly the text OCR is there to
replace -- and the run went on to spend a full detection pass on it. Tolerating
one unreadable page is right; tolerating every page is a configuration error
wearing a success message, so Document now raises OcrUnavailable when OCR was
attempted and not one page came back, quoting the first real failure. A page
range that needed no OCR at all is still not an error.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ci5T3gWDhhmqDec3e1XtYZ
Replaces the previous tree entirely. The pipeline is four separable pieces, each
testable on its own:
1. leanbench/llm.py provider-neutral OpenAI-compatible connector, stdlib only,
with retries, JSON coercion and a scripted client so the
rest of the code is testable without a network.
2. leanbench/detect.py per-page detection of formalizable mathematics. Case 1 is
mathematics asserted as fact (definition/theorem/lemma/...,
deliberately not sub-classified); case 2 is a problem left
open to the reader. Hits come back as spans -- start page
and index, end page and index -- so several per page is
normal. A hit cut off by the page end makes the detector
re-read that page together with the next, repeatedly, so an
item spread over three pages is found whole.
Offsets come from verbatim quotes resolved by
leanbench/anchors.py (exact -> whitespace-insensitive ->
fuzzy), because models quote well and count badly.
3. leanbench/formalize.py + leanbench/verify.py
each span becomes one Lean 4 declaration, judged by the real
compiler: full elaboration for case 1, statement legality
only for case 2, which has no proof to check. Lean's own
diagnostics feed back for a repair attempt. "verified" (Lean
accepted it) and "proved" (accepted with no sorry) are kept
strictly apart.
4. leanbench/db.py indexed SQLite plus JSONL export. Every item is indexed by
the types of its inputs, or by its own type when it is not a
proof, and records the source book (sha256) and the page it
starts on -- so `drop --source H --pages A-B` followed by a
re-ingest redoes exactly those pages and nothing else.
Infrastructure:
- lean/ Lake project pinned to Mathlib.
- docker/ image carrying Lean 4 and a prebuilt Mathlib. The Mathlib
revision and the toolchain are resolved against each other at
build time rather than both hardcoded, so the build does not rot.
One Docker layer per Lean step, each long one under `timeout`:
the gigabyte cache fetch earns a layer that a later failure never
invalidates, and a stalled step fails with a message instead of
sitting silent for an hour. There is deliberately no `lake build`
of our own library -- verification runs `lake env lean` against
Mathlib's oleans and needs nothing else, and building the library
was pure cost. docker/smoke.py runs the whole pipeline inside the
image against the real compiler, with the LLM scripted.
- .gitea/workflows ci (lint + tests in a venv, since the runner image's Python is
Debian-managed and refuses installs into itself), docker (build,
verify, optionally push), ingest (submit a book by hand). The
docker build runs with --progress=plain, because BuildKit's
default renderer collapses a long step into one rewritten line
and a stall then looks exactly like progress.
The docker workflow is not a throwaway build: it leaves the image, the layer cache,
an idle `lean-bench` container and the `lean-bench-data` / `lean-bench-books`
volumes in place on the runner, and never prunes, so the environment is there to be
used afterwards. Pushing to the registry is optional and cannot fail the build --
Gitea's container registry does not accept the built-in Actions token, so that needs
a REGISTRY_TOKEN secret with write:package.
58 tests; they need neither network nor Lean.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ci5T3gWDhhmqDec3e1XtYZ
claude-bot
codex-bot
Zayden-Jung