# GraphRAG Community Reports Entities: 970 | Relationships: 2759 | Communities: 6 | Themes: 2 | Text units: 860 Query with `readmenator . ask ""` (local: BM25 + Personalized PageRank; global: map-reduce over these reports) or the MCP tool `readmenator.graphrag`. ## Project overview (`root`, root, rating 7.0) 44 files in 6 communities and 2 themes. Highest-impact communities: topogpt3: model (7.0), topogpt3: inference_hrm (2.9), eval: governor (2.3). God nodes: topogpt3/model.py, topogpt3/__init__.py, topogpt3/train.py, topogpt3/lens_model.py, topogpt3/inference_hrm.py. - [topogpt3: model] rating 7.0: `topogpt3/model.py` ranks 1 by PageRank, 15 importers, 187 symbols: TopoGPT2: Quaternion-Enhanced Topological Transformer Language Model Author: Gris Iscomeback Email: grisiscomeback@gmail.com License: GPL v3 Mejoras sobre topogpt.py: - Álgebra de cuaterniones completa (QuaternionLinear... - [topogpt3: inference_hrm] rating 2.9: `topogpt3/continuation.py` ranks 1 by PageRank, 3 importers, 5 symbols: Auto-continuation engine: detects truncated responses and feeds the last incomplete lines back so the model can resume where it left off. - [eval: governor] rating 2.3: `topogpt3/__init__.py` ranks 1 by PageRank, 8 importers, 0 symbols: TopoGPT3: complex-valued spectral language model for code. - [eval: harness] rating 2.2: `eval/harness.py` ranks 1 by PageRank, 3 importers, 12 symbols: Harness for evaluating TopoGPT3 on HumanEval (164 problems). - [topogpt3: jlens] rating 1.9: `topogpt3/lens_model.py` ranks 1 by PageRank, 6 importers, 29 symbols: LensModel: What the lens needs from a model. - [unassigned files] rating 1.9: `eval/analyze.py` ranks 1 by PageRank, 0 importers, 5 symbols: Aggregate HumanEval result JSONL files into a summary table. - Key entities: file:topogpt3/model.py, file:topogpt3/train.py, file:topogpt3/inference_hrm.py, file:topogpt3/api_server.py, file:eval/governor.py, file:eval/repair.py - Children: t0, t1 - Root rating = highest community rating. ## topogpt3: model + topogpt3: inference_hrm +3 (`t0`, theme, rating 7.0) Theme of 5 communities and 34 files: topogpt3: model (12 files, rating 7.0); topogpt3: inference_hrm (5 files, rating 2.9); eval: governor (6 files, rating 2.3); eval: harness (7 files, rating 2.2); topogpt3: jlens (4 files, rating 1.9). - [topogpt3: model] `topogpt3/model.py` ranks 1 by PageRank, 15 importers, 187 symbols: TopoGPT2: Quaternion-Enhanced Topological Transformer Language Model Author: Gris Iscomeback Email: grisiscomeback@gmail.com License: GPL v3 Mejoras sobre topogpt.py: - Álgebra de cuaterniones completa (QuaternionLinear... - [topogpt3: model] `synthetic_dataset.py` ranks 2 by PageRank, 1 importers, 35 symbols: Synthetic Dataset Generator for TopoGPT2. - [topogpt3: inference_hrm] `topogpt3/continuation.py` ranks 1 by PageRank, 3 importers, 5 symbols: Auto-continuation engine: detects truncated responses and feeds the last incomplete lines back so the model can resume where it left off. - [topogpt3: inference_hrm] `topogpt3/inference.py` ranks 2 by PageRank, 2 importers, 54 symbols: TopoGPT3 inference engine. - [eval: governor] `topogpt3/__init__.py` ranks 1 by PageRank, 8 importers, 0 symbols: TopoGPT3: complex-valued spectral language model for code. - [eval: governor] `eval/governor.py` ranks 2 by PageRank, 1 importers, 20 symbols: Streaming + governance for autoregressive generation. - [eval: harness] `eval/harness.py` ranks 1 by PageRank, 3 importers, 12 symbols: Harness for evaluating TopoGPT3 on HumanEval (164 problems). - [eval: harness] `eval/sandbox.py` ranks 2 by PageRank, 2 importers, 9 symbols: Sandbox for executing model-generated code during HumanEval evaluation. - Key entities: file:topogpt3/model.py, file:topogpt3/train.py, file:topogpt3/inference_hrm.py, file:topogpt3/api_server.py, file:eval/governor.py, file:eval/repair.py, file:eval/harness.py, file:eval/sandbox.py - Children: c0, c3, c2, c1, c4 - Theme rating = highest child community rating. ## unassigned files (`t1`, theme, rating 1.9) Theme of 1 communities and 10 files: unassigned files (10 files, rating 1.9). - [unassigned files] `eval/analyze.py` ranks 1 by PageRank, 0 importers, 5 symbols: Aggregate HumanEval result JSONL files into a summary table. - [unassigned files] `eval/analyze_results.py` ranks 2 by PageRank, 0 importers, 4 symbols: Analyze a HumanEval JSONL produced by harness.py. - Key entities: file:topogpt3/vanilla_control.py, file:eval/report.py - Children: c5 - Theme rating = highest child community rating. ## topogpt3: model (`c0`, community, rating 7.0) 12 files under topogpt3 (py 12), mostly utility. Core file topogpt3/model.py (PageRank 0.1440, imported by 15 files): TopoGPT2: Quaternion-Enhanced Topological Transformer Language Model Author: Gris Iscomeback Email: grisiscomeback@gmail.com License: GPL v3 Mejoras sobre topogpt.py: - Álgebra de cuaterniones completa (QuaternionLinear... Key abstractions: TopoGPT2Config, QuaternionOps, QuaternionLinear, QuaternionSpectralLayer, SpectralAutoencoder, QuaternionTorusBrain. Depends on eval: governor (1), topogpt3: inference_hrm (1). Used by topogpt3: jlens (4), eval: governor (3), topogpt3: inference_hrm (2). - `topogpt3/model.py` ranks 1 by PageRank, 15 importers, 187 symbols: TopoGPT2: Quaternion-Enhanced Topological Transformer Language Model Author: Gris Iscomeback Email: grisiscomeback@gmail.com License: GPL v3 Mejoras sobre topogpt.py: - Álgebra de cuaterniones completa (QuaternionLinear... - `synthetic_dataset.py` ranks 2 by PageRank, 1 importers, 35 symbols: Synthetic Dataset Generator for TopoGPT2. - `topogpt3/exploitgym_loader.py` ranks 3 by PageRank, 3 importers, 19 symbols: ExploitGym Data Loader for TopoExploit Clones the ExploitGym repo and tokenizes task data into training memmaps. - Hotspot `topogpt3/model.py`: 187 symbols, 42 connections (score 1.00). - Hotspot `topogpt3/train.py`: 64 symbols, 35 connections (score 0.64). - Taint: 8 paths reach this group via eval, subprocess. - Surprising bridge: sandbox_smoke.py <-> hodge_cm.py (6 hops across communities). - Key entities: file:topogpt3/model.py, file:topogpt3/train.py, file:synthetic_dataset.py, file:topogpt3/exploitgym_loader.py, file:topogpt3/ewc.py, sym:topogpt3/model.py::main@3570, sym:topogpt3/model.py::run@1771, file:infer_exploitgym.py - Rating 7.0/10 = 7 x PageRank share 1.00 + 3 x risk 0.00. Internal imports: 18. ## topogpt3: inference_hrm (`c3`, community, rating 2.9) 5 files under topogpt3 (py 5), mostly utility. Core file topogpt3/continuation.py (PageRank 0.0942, imported by 3 files): Auto-continuation engine: detects truncated responses and feeds the last incomplete lines back so the model can resume where it left off. Key abstractions: is_response_complete, extract_tail_for_continuation, split_at_last_newline, ScalePreset, InferenceSettings, InferenceLoggerFactory. Depends on topogpt3: model (2), topogpt3: jlens (2). Used by eval: governor (2), topogpt3: model (1). - `topogpt3/continuation.py` ranks 1 by PageRank, 3 importers, 5 symbols: Auto-continuation engine: detects truncated responses and feeds the last incomplete lines back so the model can resume where it left off. - `topogpt3/inference.py` ranks 2 by PageRank, 2 importers, 54 symbols: TopoGPT3 inference engine. - `topogpt3/inference_hrm.py` ranks 3 by PageRank, 2 importers, 76 symbols: TopoGPT3.1: Hierarchical Recursive Reasoning Inference Engine. - Hotspot `topogpt3/api_server.py`: 46 symbols, 30 connections (score 0.53). - Hotspot `topogpt3/inference_hrm.py`: 76 symbols, 16 connections (score 0.39). - Taint: 3 paths reach this group via eval. - Key entities: file:topogpt3/inference_hrm.py, file:topogpt3/api_server.py, file:topogpt3/inference.py, sym:topogpt3/inference_hrm.py::execute@1239, sym:topogpt3/inference.py::execute@573, sym:topogpt3/inference_hrm.py::run@1048, file:topogpt3/continuation.py, file:topogpt3/__main__.py - Rating 2.9/10 = 7 x PageRank share 0.41 + 3 x risk 0.00. Internal imports: 5. ## eval: governor (`c2`, community, rating 2.3) 6 files under eval (py 6), mostly utility. Core file topogpt3/__init__.py (PageRank 0.0730, imported by 8 files): TopoGPT3: complex-valued spectral language model for code. Key abstractions: TokenStream, StopReason, GenerationResult, GenerationGovernor, put, mark_done. Depends on topogpt3: model (3), topogpt3: inference_hrm (2), topogpt3: jlens (2). Used by eval: harness (3), topogpt3: model (1). - `topogpt3/__init__.py` ranks 1 by PageRank, 8 importers, 0 symbols: TopoGPT3: complex-valued spectral language model for code. - `eval/governor.py` ranks 2 by PageRank, 1 importers, 20 symbols: Streaming + governance for autoregressive generation. - `app.py` ranks 3 by PageRank, 0 importers, 5 symbols: Drop-in entry point that demonstrates how to use the topogpt3 package. - Hotspot `topogpt3/__init__.py`: 0 symbols, 23 connections (score 0.33). - Hotspot `eval/repair.py`: 6 symbols, 17 connections (score 0.26). - Taint: 4 paths reach this group via eval, subprocess. - Dataflow: 1 INFERRED issues (first: UNCHECKED_ALLOC in main). - Surprising bridge: governor.py <-> sandbox_smoke.py (5 hops across communities). - Key entities: file:eval/governor.py, file:eval/repair.py, file:topogpt3/__init__.py, file:eval/governor_smoke.py, file:app.py, file:eval/smoke.py, sym:eval/governor_smoke.py::load_model@30, sym:eval/governor_smoke.py::test_tokenstream_threadsafety@49 - Rating 2.3/10 = 7 x PageRank share 0.33 + 3 x risk 0.00. Internal imports: 5. ## eval: harness (`c1`, community, rating 2.2) 7 files under eval (py 7), mostly utility. Core file eval/harness.py (PageRank 0.0277, imported by 3 files): Harness for evaluating TopoGPT3 on HumanEval (164 problems). Key abstractions: ModelLoader, load_humaneval, build_prompt, extract_candidate, run_one_test, run_one_test_sandboxed. Depends on eval: governor (3), topogpt3: model (1). - `eval/harness.py` ranks 1 by PageRank, 3 importers, 12 symbols: Harness for evaluating TopoGPT3 on HumanEval (164 problems). - `eval/sandbox.py` ranks 2 by PageRank, 2 importers, 9 symbols: Sandbox for executing model-generated code during HumanEval evaluation. - `eval/samplers.py` ranks 3 by PageRank, 2 importers, 7 symbols: Registry of sampler constructors for the HumanEval harness. - Hotspot `eval/harness.py`: 12 symbols, 31 connections (score 0.47). - Hotspot `eval/noise_sweep.py`: 4 symbols, 18 connections (score 0.27). - Taint: 3 paths reach this group via subprocess. - Surprising bridge: sandbox_smoke.py <-> hodge_cm.py (6 hops across communities). - Key entities: file:eval/harness.py, file:eval/sandbox.py, file:eval/noise_sweep.py, file:eval/temp_sweep.py, file:eval/samplers.py, file:eval/integration_smoke.py, sym:eval/sandbox.py::check_safety@133, sym:eval/harness.py::evaluate_problem@272 - Rating 2.2/10 = 7 x PageRank share 0.31 + 3 x risk 0.00. Internal imports: 8. ## topogpt3: jlens (`c4`, community, rating 1.9) 4 files under topogpt3 (py 4), mostly testing. Core file topogpt3/lens_model.py (PageRank 0.0520, imported by 6 files): LensModel: What the lens needs from a model. Key abstractions: LensModel, TopoGPT3LensConfig, TopoGPT3LensModel, TinyDecoder, encode, forward. Depends on topogpt3: model (4). Used by eval: governor (2), topogpt3: inference_hrm (2). - `topogpt3/lens_model.py` ranks 1 by PageRank, 6 importers, 29 symbols: LensModel: What the lens needs from a model. - `topogpt3/jlens.py` ranks 2 by PageRank, 7 importers, 29 symbols: TopoGPT3JLensFitConfig: Centralized configuration for Jacobian lens fitting. - `tests/test_jlens.py` ranks 3 by PageRank, 0 importers, 51 symbols: TestValidPositionMask: Feature: valid_position_mask excludes attention-sink and final positions. - Hotspot `topogpt3/jlens.py`: 29 symbols, 23 connections (score 0.39). - Hotspot `topogpt3/lens_model.py`: 29 symbols, 22 connections (score 0.38). - Taint: 2 paths reach this group via eval. - Key entities: file:tests/test_jlens.py, file:topogpt3/lens_model.py, file:topogpt3/jlens.py, file:tests/test_lens_model.py, sym:topogpt3/jlens.py::fit@291, sym:topogpt3/jlens.py::jacobian_for_prompt@187, sym:topogpt3/lens_model.py::TinyDecoder@306, sym:topogpt3/lens_model.py::TopoGPT3LensModel@161 - Rating 1.9/10 = 7 x PageRank share 0.27 + 3 x risk 0.00. Internal imports: 9. ## unassigned files (`c5`, community, rating 1.9) 10 files under root (py 5, sh 5), mostly utility. Core file eval/analyze.py (PageRank 0.0104, imported by 0 files): Aggregate HumanEval result JSONL files into a summary table. Key abstractions: pass_at_k, classify_error, load_jsonl, summarize, main, load_records. - `eval/analyze.py` ranks 1 by PageRank, 0 importers, 5 symbols: Aggregate HumanEval result JSONL files into a summary table. - `eval/analyze_results.py` ranks 2 by PageRank, 0 importers, 4 symbols: Analyze a HumanEval JSONL produced by harness.py. - `eval/noise_analysis.py` ranks 3 by PageRank, 0 importers, 3 symbols: Analisis post-hoc del noise sweep. - Hotspot `topogpt3/vanilla_control.py`: 26 symbols, 9 connections (score 0.18). - Hotspot `eval/report.py`: 6 symbols, 9 connections (score 0.14). - Key entities: file:topogpt3/vanilla_control.py, file:eval/report.py, file:run_merged.sh, file:eval/analyze.py, file:eval/noise_analysis.py, file:eval/analyze_results.py, sym:topogpt3/vanilla_control.py::__init__@73, file:run_exploitgym_v2.sh - Rating 1.9/10 = 7 x PageRank share 0.27 + 3 x risk 0.00. Internal imports: 0.