# goedel-lm/goedel-prover-v2 architecture

> Architecture diagram of the GitHub repository goedel-lm/goedel-prover-v2, generated by GitDiagram from its file tree, README and a sample of source files (last updated 2026-11-16). It is an AI-made overview: check details against the source.

- Interactive diagram: https://gitdiagram.com/goedel-lm/goedel-prover-v2
- Repository: https://github.com/goedel-lm/goedel-prover-v2

## Overview

Goedel-Prover-V2 is a Lean 4 theorem-proving system centered on generating candidate proofs and, optionally, revising failed proofs using compiler feedback. Its runtime path is model inference, Lean compilation/verification, and result summarization; correction-round preparation is an internal support layer. The README documents scaffolded training and model averaging as project capabilities, but the sampled runtime files do not implement those stages. Connections to unsampled dataset usage are therefore not asserted.

Main entry points: [`src/inference.py`](https://github.com/goedel-lm/goedel-prover-v2/blob/main/src/inference.py) (Inference runner), [`src/compile.py`](https://github.com/goedel-lm/goedel-prover-v2/blob/main/src/compile.py) (Compile adapter), [`lean_compiler/repl_scheduler.py`](https://github.com/goedel-lm/goedel-prover-v2/blob/main/lean_compiler/repl_scheduler.py) (REPL scheduler), [`src/utils.py`](https://github.com/goedel-lm/goedel-prover-v2/blob/main/src/utils.py) (Correction preparation)

## Diagram (Mermaid)

```mermaid
flowchart TD

subgraph group_generation["Proof generation"]
  node_inference["Inference runner<br/>[inference.py]"]
  node_handlers["Prompt handlers<br/>[utils.py]"]
  node_correction["Correction preparation<br/>[utils.py]"]
end

subgraph group_verification["Lean verification"]
  node_compiler["Compile adapter<br/>[compile.py]"]
  node_scheduler["REPL scheduler<br/>[repl_scheduler.py]"]
end

subgraph group_analysis["Results analysis"]
  node_summarizer["Result summarizer<br/>[summarize.py]"]
end

subgraph group_inputs["Problem inputs"]
  node_benchmarks["Problem datasets"]
end

node_user(("Operator"))
node_model["Language model"]
node_lean["Lean REPL"]

node_user -->|"starts runs"| node_inference
node_benchmarks -.->|"supplies problems"| node_inference
node_inference -->|"builds prompts"| node_handlers
node_inference -->|"loads revisions"| node_correction
node_correction -->|"prepares retries"| node_inference
node_inference -->|"requests generations"| node_model
node_inference -.->|"submits proofs"| node_compiler
node_compiler -->|"schedules proofs"| node_scheduler
node_scheduler -->|"sends commands"| node_lean
node_lean -->|"returns diagnostics"| node_scheduler
node_compiler -.->|"provides results"| node_summarizer
node_summarizer -->|"reports solve rates"| node_user
node_lean -.->|"guides revisions"| node_inference

click node_benchmarks "https://github.com/goedel-lm/goedel-prover-v2/tree/main/dataset"
click node_inference "https://github.com/goedel-lm/goedel-prover-v2/blob/main/src/inference.py"
click node_handlers "https://github.com/goedel-lm/goedel-prover-v2/blob/main/src/utils.py"
click node_correction "https://github.com/goedel-lm/goedel-prover-v2/blob/main/src/utils.py"
click node_compiler "https://github.com/goedel-lm/goedel-prover-v2/blob/main/src/compile.py"
click node_scheduler "https://github.com/goedel-lm/goedel-prover-v2/blob/main/lean_compiler/repl_scheduler.py"
click node_summarizer "https://github.com/goedel-lm/goedel-prover-v2/blob/main/src/summarize.py"

classDef toneNeutral fill:#f8fafc,stroke:#334155,stroke-width:1.5px,color:#0f172a
classDef toneBlue fill:#dbeafe,stroke:#2563eb,stroke-width:1.5px,color:#172554
classDef toneAmber fill:#fef3c7,stroke:#d97706,stroke-width:1.5px,color:#78350f
classDef toneMint fill:#dcfce7,stroke:#16a34a,stroke-width:1.5px,color:#14532d
classDef toneRose fill:#ffe4e6,stroke:#e11d48,stroke-width:1.5px,color:#881337
classDef toneIndigo fill:#e0e7ff,stroke:#4f46e5,stroke-width:1.5px,color:#312e81
classDef toneTeal fill:#ccfbf1,stroke:#0f766e,stroke-width:1.5px,color:#134e4a
class node_inference,node_handlers,node_correction toneBlue
class node_compiler,node_scheduler toneAmber
class node_summarizer toneMint
class node_benchmarks toneRose
class node_user,node_model,node_lean toneIndigo
```

## Components

### Proof generation

- [Inference runner](https://github.com/goedel-lm/goedel-prover-v2/blob/main/src/inference.py): `src/inference.py`
- [Prompt handlers](https://github.com/goedel-lm/goedel-prover-v2/blob/main/src/utils.py): `src/utils.py`
- [Correction preparation](https://github.com/goedel-lm/goedel-prover-v2/blob/main/src/utils.py): `src/utils.py`

### Lean verification

- [Compile adapter](https://github.com/goedel-lm/goedel-prover-v2/blob/main/src/compile.py): `src/compile.py`
- [REPL scheduler](https://github.com/goedel-lm/goedel-prover-v2/blob/main/lean_compiler/repl_scheduler.py): `lean_compiler/repl_scheduler.py`

### Results analysis

- [Result summarizer](https://github.com/goedel-lm/goedel-prover-v2/blob/main/src/summarize.py): `src/summarize.py`

### Problem inputs

- [Problem datasets](https://github.com/goedel-lm/goedel-prover-v2/tree/main/dataset): `dataset`

### Other components

- **Operator**
- **Language model**
- **Lean REPL**

## Connections

- Operator → Inference runner: starts runs (evidence: `src/inference.py`)
- Problem datasets → Inference runner: supplies problems
- Inference runner → Prompt handlers: builds prompts (evidence: `src/inference.py`)
- Inference runner → Correction preparation: loads revisions (evidence: `src/inference.py`)
- Correction preparation → Inference runner: prepares retries (evidence: `src/inference.py`)
- Inference runner → Language model: requests generations (evidence: `src/inference.py`)
- Inference runner → Compile adapter: submits proofs
- Compile adapter → REPL scheduler: schedules proofs (evidence: `src/compile.py`)
- REPL scheduler → Lean REPL: sends commands (evidence: `lean_compiler/repl_scheduler.py`)
- Lean REPL → REPL scheduler: returns diagnostics (evidence: `lean_compiler/repl_scheduler.py`)
- Compile adapter → Result summarizer: provides results
- Result summarizer → Operator: reports solve rates (evidence: `src/summarize.py`)
- Lean REPL → Inference runner: guides revisions (evidence: `README.md`)
