Formal-Verification-RSI
Formal verification for AI: a Lean 4 + Dafny proof checker that catches fake AI proofs (reward hacking), and verified recursive self-improvement (RSI) for coding agents. AI for math, physics and software engineering.
Runs with mockssource: GitHubPythonMITcommit 177a62d5847d
Python, MIT licensed. The project labels itself: ai agents, ai for math, ai for science, ai safety, autoformalization, automated theorem proving, coding agents and dafny.
Formal-Verification-RSI runs, with stand-ins for the services it depends on. An Argusic agent installed it in 0.5 minutes and hit 1 error and fixed every one without help on a clean machine with no GPU, and the whole session was recorded.
160 hermetic Python tests pass (170 skipped for missing Lean/Dafny toolchains); pip install completes; all modules import; 4 previously-failing snapshot-integrity tests fixed and passing.
What the agent ended up with on a clean machine, in its own words. How this is measured
Measured by Argusic on a fresh machine every time. Every number links to its evidence. Argusic Score 92 of 100 (the mean of this project's run scores).
At a glance
Subject data from GitHub, linked at the top of this page, refreshed . Test data by Argusic (CC BY 4.0); every number links to a run page with the full log, the recording, and their sha256 hashes.
What was measured
From the run that decided this verdict. Every line is a recorded value, not an estimate.
What went wrong, in the agent's own words
The agent's own words, worst first, never rewritten into advice.
- Test Agent snapshot manifests had temp paths instead of /app, causing all 4 snapshot-integrity tests to fail with 0 graded files15 minutes
- Install time
- under a minute
- Cold machine to finish
- 81 minutes
- Errors hit and fixed
- 1 hit, 1 fixed with no human help
- How the result was proved
- pytest tests/ --ignore=tests/test_fv_curate.py --ignore=tests/test_fv_analyze.py exited 0, 160 passed, 170 skipped
- Model tokens used
- 678,568
- Exact commit tested
- 177a62d5847d
- Container image
- sha256:cdd920bce783
What was not measured
The other half, so you can see exactly how far this evidence goes. A gap is printed as a gap and never filled in.
- Peak GPU memory
- this ran on a CPU machine, so there was no GPU to sample
- Behaviour over a long run
- each attempt uses the smallest documented settings and a fixed time budget, so this measures whether it runs, not how it holds up under load
- Behaviour against real services
- this run was verified against stand-in services only
The end of the recording
The last lines of the terminal session that decided this verdict, exactly as they were recorded.
3. **B1.tar.gz corruption didn't break tar-sha comparison**: Appending a null byte to a gzip-compre… **4 failing tests → all passing**: `test_ab_only_agent_has_no_c_and_gate_check_error_does_not_resum… tokens used 678,568 Here's my final report on the mission. **Install**: The project installed cleanly via `pip install -e ".[dev]"` in a venv. Python 3.12.3 i… **Tests**: 160 pass, 170 skip. The 170 skips are all due to missing Lean (4.23.0-rc2) and Dafny (4.… **Bugs found and fixed in the test suite** (`tests/test_fv_regrade.py`): 1. **Test Agent's snapshot paths didn't match `read_snapshot` expectations**: The test `Agent` clas… 2. **`_regrade` helper used wrong `app_dir`**: It passed `app_dir=str(tmp / "box" / "app")` instead… 3. **B1.tar.gz corruption didn't break tar-sha comparison**: Appending a null byte to a gzip-compre… **4 failing tests → all passing**: `test_ab_only_agent_has_no_c_and_gate_check_error_does_not_resum…
Replay the whole session, every command from a clean machine to this point.
How it was tested
One agent, one container, no human help. It installs the project the way its own documentation says to, runs it, and fixes what breaks. Everything below is recorded as it happened: the terminal session, the log and the exact commit. The full method.
Strengths and limits
Measured facts, not opinions. How this is written.
What went well
- Installed in 0.5 minutes, faster than the median of the 89 comparable projects Argusic has measured.
- Recovered from all 1 error without a human stepping in, which says the failures are documented well enough to solve.
- Licensed MIT, as reported by its host.
What did not, or is not known
- Only reached a running state against stand-in services, so nothing here proves it works against the real ones.
- Measured once so far. A single attempt shows it can run, not that it runs reliably.
Of the 90 ai-agents projects Argusic has installed and timed, Formal-Verification-RSI was the 7th fastest to reach a running state, and 68 of 90 reached one at all.
Also tested, in the same area
Every one of these was installed and run by Argusic on a clean machine. Nothing appears here that was not tested.
Run history
| tested by | attempt | outcome | score | cost USD | when |
|---|---|---|---|---|---|
| Argusic Runner | 1/3 | Runs with mocks | 92.00 | 0.29 |
Topics (from GitHub)
ai-agentsai-for-mathai-for-scienceai-safetyautoformalizationautomated-theorem-provingcoding-agentsdafnyformal-methodsformal-verificationlean4llmprogram-verificationproof-assistantrecursive-self-improvementreward-hackingrsiself-improving-aitheorem-provingverified-code
Embed the badge
Markdown for the project README. It links back here; terms on the terms page.
[](https://argusic.com/subject/formal-verification-rsi)Questions
- Does Formal-Verification-RSI run?
- Formal-Verification-RSI runs, with mocks standing in for external services. Argusic installed and launched it on a clean machine in 1 minutes, hitting 1 error on the way, and recorded the session.
- How did Argusic test Formal-Verification-RSI?
- On a fresh, disposable machine, with every command recorded and the repository pinned at commit 177a62d5847d. One attempt is recorded, and the full method is on the methodology page.
- What did it cost to test Formal-Verification-RSI?
- The run that produced this verdict cost $0.29: real compute and model cost, not a list price per million tokens. It is recorded on that run's page.
- How long does Formal-Verification-RSI take to install?
- 0.5 minutes in the run that decided this verdict, from a clean machine to a running state. That is one measured attempt, not an average.
- Does Formal-Verification-RSI need a GPU?
- Not to start. Argusic reached a running state on a machine with no GPU.
- What goes wrong when installing Formal-Verification-RSI?
- 1 thing broke in the recorded run, and 1 were fixed without human help. Each one, and the time it cost, is listed on this page.
- How does Formal-Verification-RSI compare with the alternatives?
- Of the 90 ai-agents projects Argusic has installed and timed, Formal-Verification-RSI was the 7th fastest to reach a running state, and 68 of 90 reached one at all.
- Where is the evidence for Formal-Verification-RSI?
- The recorded run is on this page, each linking to its full log and terminal recording, stored with a sha256 fingerprint so it cannot be quietly altered.