Dataset Preview
The full dataset viewer is not available (click to read why). Only showing a preview of the rows.
The dataset generation failed
Error code: DatasetGenerationError
Exception: CastError
Message: Couldn't cast
org: string
name: string
language: string
license: string
source: string
format: struct<passed: int64, total: int64, has_h1: bool, has_h2: bool, has_h3: bool, has_mermaid: bool, has (... 77 chars omitted)
child 0, passed: int64
child 1, total: int64
child 2, has_h1: bool
child 3, has_h2: bool
child 4, has_h3: bool
child 5, has_mermaid: bool
child 6, has_citations: bool
child 7, has_table: bool
child 8, has_conclusion: bool
child 9, word_count_ok: bool
judge: struct<fidelity_score: int64, supported_claims: int64, unsupported_claims: int64, notes: string>
child 0, fidelity_score: int64
child 1, supported_claims: int64
child 2, unsupported_claims: int64
child 3, notes: string
model: string
category: string
finish_reason: string
citation: struct<total: int64, validity_rate: double, problems: list<item: struct<file: string, start: int64, (... 28 chars omitted)
child 0, total: int64
child 1, validity_rate: double
child 2, problems: list<item: struct<file: string, start: int64, end: int64, error: string>>
child 0, item: struct<file: string, start: int64, end: int64, error: string>
child 0, file: string
child 1, start: int64
child 2, end: int64
child 3, error: string
page: string
to
{'model': Value('string'), 'source': Value('string'), 'category': Value('string'), 'finish_reason': Value('string'), 'page': Value('string'), 'format': {'passed': Value('int64'), 'total': Value('int64'), 'has_h1': Value('bool'), 'has_h2': Value('bool'), 'has_h3': Value('bool'), 'has_mermaid': Value('bool'), 'has_citations': Value('bool'), 'has_table': Value('bool'), 'has_conclusion': Value('bool'), 'word_count_ok': Value('bool')}, 'citation': {'total': Value('int64'), 'validity_rate': Value('float64'), 'problems': List({'file': Value('string'), 'start': Value('int64'), 'end': Value('int64'), 'error': Value('string')})}, 'judge': {'fidelity_score': Value('int64'), 'supported_claims': Value('int64'), 'unsupported_claims': Value('int64'), 'notes': Value('string')}}
because column names don't match
Traceback: Traceback (most recent call last):
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1827, in _prepare_split_single
for key, table in generator:
^^^^^^^^^
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 613, in wrapped
for item in generator(*args, **kwargs):
~~~~~~~~~^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 343, in _generate_tables
self._cast_table(pa_table, json_field_paths=json_field_paths),
~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 132, in _cast_table
pa_table = table_cast(pa_table, self.info.features.arrow_schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2378, in table_cast
return cast_table_to_schema(table, schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2306, in cast_table_to_schema
raise CastError(
...<3 lines>...
)
datasets.table.CastError: Couldn't cast
org: string
name: string
language: string
license: string
source: string
format: struct<passed: int64, total: int64, has_h1: bool, has_h2: bool, has_h3: bool, has_mermaid: bool, has (... 77 chars omitted)
child 0, passed: int64
child 1, total: int64
child 2, has_h1: bool
child 3, has_h2: bool
child 4, has_h3: bool
child 5, has_mermaid: bool
child 6, has_citations: bool
child 7, has_table: bool
child 8, has_conclusion: bool
child 9, word_count_ok: bool
judge: struct<fidelity_score: int64, supported_claims: int64, unsupported_claims: int64, notes: string>
child 0, fidelity_score: int64
child 1, supported_claims: int64
child 2, unsupported_claims: int64
child 3, notes: string
model: string
category: string
finish_reason: string
citation: struct<total: int64, validity_rate: double, problems: list<item: struct<file: string, start: int64, (... 28 chars omitted)
child 0, total: int64
child 1, validity_rate: double
child 2, problems: list<item: struct<file: string, start: int64, end: int64, error: string>>
child 0, item: struct<file: string, start: int64, end: int64, error: string>
child 0, file: string
child 1, start: int64
child 2, end: int64
child 3, error: string
page: string
to
{'model': Value('string'), 'source': Value('string'), 'category': Value('string'), 'finish_reason': Value('string'), 'page': Value('string'), 'format': {'passed': Value('int64'), 'total': Value('int64'), 'has_h1': Value('bool'), 'has_h2': Value('bool'), 'has_h3': Value('bool'), 'has_mermaid': Value('bool'), 'has_citations': Value('bool'), 'has_table': Value('bool'), 'has_conclusion': Value('bool'), 'word_count_ok': Value('bool')}, 'citation': {'total': Value('int64'), 'validity_rate': Value('float64'), 'problems': List({'file': Value('string'), 'start': Value('int64'), 'end': Value('int64'), 'error': Value('string')})}, 'judge': {'fidelity_score': Value('int64'), 'supported_claims': Value('int64'), 'unsupported_claims': Value('int64'), 'notes': Value('string')}}
because column names don't match
The above exception was the direct cause of the following exception:
Traceback (most recent call last):
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 1369, in compute_config_parquet_and_info_response
parquet_operations, partial, estimated_dataset_info = stream_convert_to_parquet(
~~~~~~~~~~~~~~~~~~~~~~~~~^
builder, max_dataset_size_bytes=max_dataset_size_bytes
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
)
^
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 948, in stream_convert_to_parquet
builder._prepare_split(split_generator=splits_generators[split], file_format="parquet")
~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1694, in _prepare_split
for job_id, done, content in self._prepare_split_single(
~~~~~~~~~~~~~~~~~~~~~~~~~~^
gen_kwargs=gen_kwargs, job_id=job_id, **_prepare_split_args
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
):
^
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1879, in _prepare_split_single
raise DatasetGenerationError("An error occurred while generating the dataset") from e
datasets.exceptions.DatasetGenerationError: An error occurred while generating the datasetNeed help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
model string | source string | category string | finish_reason string | page string | format dict | citation dict | judge dict |
|---|---|---|---|---|---|---|---|
sw-coder-14b-v3 | bregman-arie/devops-exercises | architecture_overview | stop | # DevOps Exercises Codebase Architecture
## Overview
The `bregman-arie/devops-exercises` repository is a collection of Python scripts, AWS infrastructure code, and Flask applications designed for educational and practical purposes in DevOps and software development. The codebase is organized into several distinct mod... | {
"passed": 7,
"total": 8,
"has_h1": true,
"has_h2": true,
"has_h3": true,
"has_mermaid": true,
"has_citations": true,
"has_table": false,
"has_conclusion": true,
"word_count_ok": true
} | {
"total": 20,
"validity_rate": 0.8,
"problems": [
{
"file": "triggering/aws_s3_event_trigger.sh",
"start": 1,
"end": 122,
"error": "file-not-found"
},
{
"file": "triggering/s3-lambda/s3-lambda.py",
"start": 5,
"end": 38,
"error": "file-not-found"
},... | {
"fidelity_score": 3,
"supported_claims": 18,
"unsupported_claims": 12,
"notes": "The generated page contains several inaccuracies and unsupported claims about the codebase, including incorrect file paths, missing functionality, and false assertions about code behavior that contradict the actual source files."... |
sw-coder-14b-v3 | bregman-arie/devops-exercises | data_flow | stop | # DevOps Exercises System Architecture
## Overview
The DevOps Exercises system is a collection of Python scripts, AWS infrastructure code, and Flask applications designed to demonstrate various DevOps practices, including automation, CI/CD, and AWS infrastructure management. The system is organized into several modul... | {
"passed": 7,
"total": 8,
"has_h1": true,
"has_h2": true,
"has_h3": true,
"has_mermaid": true,
"has_citations": true,
"has_table": false,
"has_conclusion": true,
"word_count_ok": true
} | {
"total": 10,
"validity_rate": 0.8,
"problems": [
{
"file": "aws_s3_event_trigger.sh",
"start": 16,
"end": 122,
"error": "file-not-found"
},
{
"file": "s3-lambda.py",
"start": 5,
"end": 38,
"error": "file-not-found"
}
]
} | {
"fidelity_score": 3,
"supported_claims": 12,
"unsupported_claims": 8,
"notes": "The page contains several inaccuracies including incorrect method names (routee instead of route), wrong file paths, misrepresentation of script functionality, and incorrect claims about code completion status that don't match the... |
sw-coder-14b-v3 | 666ghj/MiroFish | architecture_overview | stop | # MiroFish Architecture Overview
## Overview
MiroFish is a simulation platform for social media interactions, specifically designed to run simulations on Twitter and Reddit platforms. The architecture is composed of a backend service that manages simulation execution and state tracking, and a frontend application tha... | {
"passed": 6,
"total": 8,
"has_h1": true,
"has_h2": true,
"has_h3": true,
"has_mermaid": true,
"has_citations": false,
"has_table": false,
"has_conclusion": true,
"word_count_ok": true
} | {
"total": 0,
"validity_rate": 0,
"problems": []
} | {
"fidelity_score": 3,
"supported_claims": 12,
"unsupported_claims": 8,
"notes": "The page contains several inaccuracies and unsupported claims about the codebase, including false assertions about empty files, incorrect component interactions, and misrepresentations of core functionality like the report agent a... |
sw-coder-14b-v3 | 666ghj/MiroFish | data_flow | stop | # MiroFish Simulation System Architecture
## Overview
MiroFish is a simulation platform that runs agent-based social media simulations on Twitter and Reddit platforms. The system consists of a backend service that manages simulation execution and state tracking, and a frontend that provides user interfaces for intera... | {
"passed": 7,
"total": 8,
"has_h1": true,
"has_h2": true,
"has_h3": true,
"has_mermaid": true,
"has_citations": true,
"has_table": false,
"has_conclusion": true,
"word_count_ok": true
} | {
"total": 6,
"validity_rate": 1,
"problems": []
} | {
"fidelity_score": 4,
"supported_claims": 12,
"unsupported_claims": 3,
"notes": "Most technical details are accurate but some claims about specific implementation aspects like Zep graph memory handling and exact data flow are partially misrepresented or oversimplified."
} |
sw-coder-14b-v3 | ComposioHQ/awesome-claude-skills | architecture_overview | stop | # ComposioHQ/awesome-claude-skills Architecture Overview
## Overview
The `awesome-claude-skills` codebase is a collection of Python modules designed to create and validate document processing skills, specifically for Word (DOCX) and PowerPoint (PPTX) files. It also includes a GIF creation module for generating animat... | {
"passed": 7,
"total": 8,
"has_h1": true,
"has_h2": true,
"has_h3": true,
"has_mermaid": true,
"has_citations": true,
"has_table": false,
"has_conclusion": true,
"word_count_ok": true
} | {
"total": 9,
"validity_rate": 1,
"problems": []
} | {
"fidelity_score": 4,
"supported_claims": 22,
"unsupported_claims": 3,
"notes": "Most claims are accurate but some architectural details like component structure diagram and specific method signatures are not fully supported by the context, and the document validation modules are described as identical when th... |
sw-coder-14b-v3 | ComposioHQ/awesome-claude-skills | data_flow | stop | # ComposioHQ/awesome-claude-skills System Architecture
## Overview
This repository contains two primary systems: document validation for Word and PowerPoint files, and a GIF creation system for Slack. The document validation system provides schema validation for Open XML formats, while the GIF creation system program... | {
"passed": 7,
"total": 8,
"has_h1": true,
"has_h2": true,
"has_h3": true,
"has_mermaid": true,
"has_citations": true,
"has_table": false,
"has_conclusion": true,
"word_count_ok": true
} | {
"total": 35,
"validity_rate": 0.943,
"problems": [
{
"file": "gif_builder.py",
"start": 16,
"end": 243,
"error": "file-not-found"
},
{
"file": "gif_builder.py",
"start": 148,
"end": 242,
"error": "file-not-found"
}
]
} | {
"fidelity_score": 3,
"supported_claims": 12,
"unsupported_claims": 8,
"notes": "The page contains several inaccuracies: it incorrectly states that the document validation system includes Word validation modules when the context shows only PPTX validation; it misrepresents the control flow in the GIF system by... |
sw-coder-14b-v3 | asgeirtj/system_prompts_leaks | architecture_overview | stop | # System Prompts Leaks Architecture
## Overview
This codebase is a collection of Python scripts and TypeScript type definitions for various office document processing skills and fleet management tools. It primarily consists of validation modules for Word, PowerPoint, and Excel documents, as well as a fleet management... | {
"passed": 6,
"total": 8,
"has_h1": true,
"has_h2": true,
"has_h3": true,
"has_mermaid": true,
"has_citations": false,
"has_table": false,
"has_conclusion": true,
"word_count_ok": true
} | {
"total": 0,
"validity_rate": 0,
"problems": []
} | {
"fidelity_score": 3,
"supported_claims": 12,
"unsupported_claims": 8,
"notes": "The page contains several inaccuracies: it incorrectly describes the validation modules as handling 'different document types' when they're actually duplicated and mislabeled, misrepresents the fleet manager as having a 'CLI inter... |
sw-coder-14b-v3 | asgeirtj/system_prompts_leaks | data_flow | stop | # System Prompts Leaks: Code Analysis
## Overview
This codebase is a collection of Python scripts and TypeScript definitions for various system management and validation tools. The primary focus is on system management through CLI tools, specifically for managing coding agents across machines using Herdr or tmux sess... | {
"passed": 7,
"total": 8,
"has_h1": true,
"has_h2": true,
"has_h3": true,
"has_mermaid": true,
"has_citations": true,
"has_table": false,
"has_conclusion": true,
"word_count_ok": true
} | {
"total": 10,
"validity_rate": 0.9,
"problems": [
{
"file": "Meta/muse-code/skills/fleet-manager/scripts/fleet_manager.py",
"start": 295,
"end": 294,
"error": "line-out-of-bounds(2379)"
}
]
} | {
"fidelity_score": 3,
"supported_claims": 12,
"unsupported_claims": 8,
"notes": "The page contains several inaccuracies: it incorrectly states the validation modules are for Word document processing when they're for office document validation, misrepresents the TypeScript file as a complete definition file, an... |
sw-coder-14b-v3 | meta-llama/llama | architecture_overview | stop | # LLaMA 2 Architecture Overview
## Overview
The LLaMA 2 codebase is a large language model implementation developed by Meta Platforms. It provides a modular architecture for generating text using transformer-based models. The codebase consists of several key components that work together to enable text generation thr... | {
"passed": 6,
"total": 8,
"has_h1": true,
"has_h2": true,
"has_h3": true,
"has_mermaid": true,
"has_citations": false,
"has_table": false,
"has_conclusion": true,
"word_count_ok": true
} | {
"total": 0,
"validity_rate": 0,
"problems": []
} | {
"fidelity_score": 4,
"supported_claims": 12,
"unsupported_claims": 2,
"notes": "Most claims are accurate but some details like specific model parameters and exact component interactions are oversimplified or slightly inaccurate compared to the actual code."
} |
sw-coder-14b-v3 | meta-llama/llama | data_flow | stop | # LLaMA 2 System Architecture
## Overview
The LLaMA 2 system is a large language model framework designed for text generation tasks. It consists of a tokenizer, a transformer model, and example scripts for both text and chat completion. The system is built using PyTorch and leverages FairScale's model parallelism for... | {
"passed": 7,
"total": 8,
"has_h1": true,
"has_h2": true,
"has_h3": true,
"has_mermaid": true,
"has_citations": true,
"has_table": false,
"has_conclusion": true,
"word_count_ok": true
} | {
"total": 6,
"validity_rate": 1,
"problems": []
} | {
"fidelity_score": 4,
"supported_claims": 18,
"unsupported_claims": 3,
"notes": "Most claims are supported by the context, but some architectural details like 'model parallelism' and specific implementation aspects are not fully accurate or missing key details from the actual code."
} |
sw-coder-14b-v3 | ultralytics/yolov5 | architecture_overview | length | # YOLOv5 Architecture Overview
## Overview
YOLOv5 is a state-of-the-art object detection framework built on PyTorch, designed for high-performance real-time object detection. The architecture is modular, with clear separation between model components, data handling, and training utilities. This document provides a co... | {
"passed": 5,
"total": 8,
"has_h1": true,
"has_h2": true,
"has_h3": true,
"has_mermaid": true,
"has_citations": false,
"has_table": false,
"has_conclusion": false,
"word_count_ok": true
} | {
"total": 0,
"validity_rate": 0,
"problems": []
} | {
"fidelity_score": 3,
"supported_claims": 12,
"unsupported_claims": 8,
"notes": "The page contains several inaccuracies: MixConv2d is not in models/yolo.py but in models/experimental.py, the mermaid diagram is nonsensical and doesn't represent actual code flow, and several claimed entry points and interfaces a... |
sw-coder-14b-v3 | ultralytics/yolov5 | data_flow | stop | # YOLOv5 Data Flow and Control Structure
## Overview
YOLOv5 is a state-of-the-art object detection system that supports both standard object detection and instance segmentation. This document traces the primary data and control flow through the system, focusing on how data is processed from input to output, and how v... | {
"passed": 7,
"total": 8,
"has_h1": true,
"has_h2": true,
"has_h3": true,
"has_mermaid": true,
"has_citations": true,
"has_table": false,
"has_conclusion": true,
"word_count_ok": true
} | {
"total": 18,
"validity_rate": 0.944,
"problems": [
{
"file": "utils/segment/dataloaders.py",
"start": 86,
"end": 304,
"error": "line-out-of-bounds(301)"
}
]
} | {
"fidelity_score": 4,
"supported_claims": 18,
"unsupported_claims": 3,
"notes": "Most claims accurately reflect the code structure and data flow, but some details like specific function call sequences and exact parameter handling are oversimplified or slightly inaccurate."
} |
sw-coder-14b-v3 | soimort/you-get | architecture_overview | stop | # you-get Architecture Overview
## Overview
`you-get` is a Python-based command-line utility for downloading videos from various websites. The architecture is organized around a core extraction framework that handles video metadata parsing and download logic, with specialized extractors for different platforms. The s... | {
"passed": 6,
"total": 8,
"has_h1": true,
"has_h2": true,
"has_h3": true,
"has_mermaid": true,
"has_citations": false,
"has_table": false,
"has_conclusion": true,
"word_count_ok": true
} | {
"total": 0,
"validity_rate": 0,
"problems": []
} | {
"fidelity_score": 4,
"supported_claims": 12,
"unsupported_claims": 3,
"notes": "Most claims are accurate but some details like specific class names and exact method signatures are slightly misrepresented, and the component diagram is not fully accurate to the actual code structure."
} |
sw-coder-14b-v3 | soimort/you-get | data_flow | stop | # You-Get System Overview
## Overview
You-Get is a Python command-line utility for downloading videos and media content from various websites. It provides a modular architecture where each website has its own extractor class that handles the specific extraction logic for that site. The system is designed to be extens... | {
"passed": 7,
"total": 8,
"has_h1": true,
"has_h2": true,
"has_h3": true,
"has_mermaid": true,
"has_citations": true,
"has_table": false,
"has_conclusion": true,
"word_count_ok": true
} | {
"total": 2,
"validity_rate": 1,
"problems": []
} | {
"fidelity_score": 4,
"supported_claims": 18,
"unsupported_claims": 3,
"notes": "Most claims accurately reflect the code structure and functionality, but some details like the exact flow in the universal extractor and specific ffmpeg functions are oversimplified or slightly inaccurate."
} |
sw-coder-14b-v3 | Alishahryar1/free-claude-code | architecture_overview | stop | # Free Claude Code Architecture
## Overview
The Free Claude Code (FCC) codebase is a Python-based application designed for managing and configuring AI models and messaging systems. It provides a web-based admin interface for configuring providers, models, and messaging settings. The architecture is composed of severa... | {
"passed": 7,
"total": 8,
"has_h1": true,
"has_h2": true,
"has_h3": true,
"has_mermaid": true,
"has_citations": true,
"has_table": false,
"has_conclusion": true,
"word_count_ok": true
} | {
"total": 8,
"validity_rate": 1,
"problems": []
} | {
"fidelity_score": 3,
"supported_claims": 12,
"unsupported_claims": 8,
"notes": "Many claims about the architecture are partially accurate but miss key technical details, such as the web interface being primarily JavaScript with Python backend components, and some component relationships are misrepresented or ... |
sw-coder-14b-v3 | Alishahryar1/free-claude-code | data_flow | stop | # Free Claude Code Admin Interface Data Flow
## Overview
The Free Claude Code admin interface is a web-based management tool that provides configuration and monitoring capabilities for the Free Claude Code application. It handles user interactions, configuration management, provider authentication, and system status ... | {
"passed": 7,
"total": 8,
"has_h1": true,
"has_h2": true,
"has_h3": true,
"has_mermaid": true,
"has_citations": true,
"has_table": false,
"has_conclusion": true,
"word_count_ok": true
} | {
"total": 10,
"validity_rate": 1,
"problems": []
} | {
"fidelity_score": 3,
"supported_claims": 12,
"unsupported_claims": 8,
"notes": "Many claims about the admin interface's functionality are accurate, but several key technical details are misrepresented or oversimplified, such as the exact data flow between components and the specific implementation details of ... |
sw-coder-14b-v3 | justjavac/wechat-miniapp-radar | architecture_overview | stop | # 微信小程序雷达架构
## 概述
微信小程序雷达是一个用于监控和分析小程序生态资源的系统，它通过收集、处理和生成报告来帮助开发者做出决策。系统由多个核心组件组成，包括数据库、报告生成、资源管理、AI 集成和用户界面组件。
## 详细分析
### 主要组件
#### 数据库层 (`db/client.ts`, `db/schema.ts`)
数据库层使用 PostgreSQL 作为后端存储，通过 `drizzle-orm` 进行 ORM 操作。主要表包括：
- `resources`：存储资源的基本信息
- `resource_signals`：存储资源信号
- `resource_scores`：存储资源评分
- `re... | {
"passed": 4,
"total": 8,
"has_h1": true,
"has_h2": true,
"has_h3": true,
"has_mermaid": true,
"has_citations": false,
"has_table": false,
"has_conclusion": false,
"word_count_ok": false
} | {
"total": 0,
"validity_rate": 0,
"problems": []
} | {
"fidelity_score": 4,
"supported_claims": 22,
"unsupported_claims": 3,
"notes": "Most claims are supported by the codebase, but some architectural interpretations and component interactions are oversimplified or slightly inaccurate compared to the actual implementation details."
} |
sw-coder-14b-v3 | justjavac/wechat-miniapp-radar | data_flow | stop | # MiniProgram Radar System Overview
## Overview
The MiniProgram Radar system is a comprehensive platform for managing and analyzing resources related to WeChat Mini Programs. It tracks resources, generates weekly reports, provides health checks, and offers advisory services. The system is built using TypeScript and l... | {
"passed": 7,
"total": 8,
"has_h1": true,
"has_h2": true,
"has_h3": true,
"has_mermaid": true,
"has_citations": true,
"has_table": false,
"has_conclusion": true,
"word_count_ok": true
} | {
"total": 13,
"validity_rate": 1,
"problems": []
} | {
"fidelity_score": 4,
"supported_claims": 22,
"unsupported_claims": 3,
"notes": "Most claims are supported by the codebase, but some details like the exact flow of 'buildSignalDigest' and the specific Mermaid diagram don't fully align with the actual implementation details in the source files."
} |
sw-coder-14b-v3 | Crosstalk-Solutions/project-nomad | architecture_overview | stop | # Project Nomad Architecture
## Overview
Project Nomad is a Docker-based service management platform built with TypeScript. The architecture is centered around a core `DockerService` class that manages the lifecycle of Docker containers for various services. The system provides a robust framework for installing, star... | {
"passed": 7,
"total": 8,
"has_h1": true,
"has_h2": true,
"has_h3": true,
"has_mermaid": true,
"has_citations": true,
"has_table": false,
"has_conclusion": true,
"word_count_ok": true
} | {
"total": 4,
"validity_rate": 1,
"problems": []
} | {
"fidelity_score": 4,
"supported_claims": 28,
"unsupported_claims": 4,
"notes": "Most claims are supported by the context, but some architectural details like the exact Mermaid diagram and specific error handling implementation details are not fully accurate or missing from the provided code."
} |
sw-coder-14b-v3 | Crosstalk-Solutions/project-nomad | data_flow | stop | # Docker Service Integration in Project Nomad
## Overview
The Docker Service (`admin/app/services/docker_service.ts`) is the core component responsible for managing Docker containers within the Project Nomad application. It handles the lifecycle of services, including installation, uninstallation, and runtime control... | {
"passed": 7,
"total": 8,
"has_h1": true,
"has_h2": true,
"has_h3": true,
"has_mermaid": true,
"has_citations": true,
"has_table": false,
"has_conclusion": true,
"word_count_ok": true
} | {
"total": 8,
"validity_rate": 1,
"problems": []
} | {
"fidelity_score": 4,
"supported_claims": 22,
"unsupported_claims": 3,
"notes": "Most claims are supported by the code, but some details like specific error handling in pullImage and the exact behavior of GPU configuration are slightly misrepresented or oversimplified in the generated content."
} |
sw-coder-14b-v3 | microsoft/playwright-mcp | architecture_overview | length | # Playwright MCP Architecture
## Overview
Playwright MCP (Microsoft Playwright Model Context Protocol) is a tool that integrates Playwright browser automation capabilities with Model Context Protocol (MCP) for automated web interaction and testing. The architecture is designed to provide a unified interface for brows... | {
"passed": 6,
"total": 8,
"has_h1": true,
"has_h2": true,
"has_h3": true,
"has_mermaid": true,
"has_citations": true,
"has_table": false,
"has_conclusion": false,
"word_count_ok": true
} | {
"total": 1,
"validity_rate": 1,
"problems": []
} | {
"fidelity_score": 3,
"supported_claims": 22,
"unsupported_claims": 10,
"notes": "The generated page contains several inaccuracies and unsupported claims about the repository structure and functionality, including incorrect component interactions, overgeneralized descriptions, and a malformed Mermaid diagram t... |
sw-coder-14b-v3 | microsoft/playwright-mcp | data_flow | stop | # Playwright MCP System Overview
## Overview
The Playwright MCP (Model Context Protocol) system is a tool that integrates browser automation capabilities with machine learning models through the Model Context Protocol (MCP) framework. It provides a command-line interface for configuring and running browser automation... | {
"passed": 7,
"total": 8,
"has_h1": true,
"has_h2": true,
"has_h3": true,
"has_mermaid": true,
"has_citations": true,
"has_table": false,
"has_conclusion": true,
"word_count_ok": true
} | {
"total": 5,
"validity_rate": 1,
"problems": []
} | {
"fidelity_score": 3,
"supported_claims": 12,
"unsupported_claims": 8,
"notes": "The page contains several accurate details about the repository structure and functionality, but also includes some incorrect claims about specific file contents and implementation details that don't match the actual source code."... |
sw-coder-14b-v3 | markedjs/marked | architecture_overview | stop | # Marked.js Architecture
## Overview
Marked.js is a markdown parser written in TypeScript that converts markdown text to HTML. The architecture is designed around a lexer-parser-renderer pattern with extensibility through hooks and extensions. The core components are the Lexer, Parser, Renderer, and Hooks system, whi... | {
"passed": 6,
"total": 8,
"has_h1": true,
"has_h2": true,
"has_h3": true,
"has_mermaid": true,
"has_citations": false,
"has_table": false,
"has_conclusion": true,
"word_count_ok": true
} | {
"total": 0,
"validity_rate": 0,
"problems": []
} | {
"fidelity_score": 3,
"supported_claims": 12,
"unsupported_claims": 8,
"notes": "Many claims are partially accurate but miss key implementation details like the actual class names (_Lexer, _Parser) and the specific hook method signatures, while some architectural elements like the 'marked' function entry point... |
sw-coder-14b-v3 | markedjs/marked | data_flow | stop | # Marked.js System Overview
## Overview
Marked.js is a markdown parser written in TypeScript that converts markdown text into HTML. The system follows a three-stage processing pipeline: **Lexing** (tokenization), **Parsing** (compiling), and **Rendering**. The system is designed to be extensible through hooks, extens... | {
"passed": 7,
"total": 8,
"has_h1": true,
"has_h2": true,
"has_h3": true,
"has_mermaid": true,
"has_citations": true,
"has_table": false,
"has_conclusion": true,
"word_count_ok": true
} | {
"total": 5,
"validity_rate": 0.8,
"problems": [
{
"file": "src/Hooks.ts",
"start": 73,
"end": 73,
"error": "line-out-of-bounds(69)"
}
]
} | {
"fidelity_score": 3,
"supported_claims": 12,
"unsupported_claims": 8,
"notes": "Many claims are partially accurate but some key technical details are incorrect or misleading, such as the entry point being the 'marked' function (which is not in the provided context), and the sequence diagram showing hooks as a... |
sw-coder-14b-v3 | date-fns/date-fns | architecture_overview | length | # date-fns Architecture Overview
## Overview
The `date-fns` library is a comprehensive, modern JavaScript date utility library that provides a modular, functional approach to date manipulation. The codebase is organized into several packages, each serving specific functionalities related to date formatting, parsing, ... | {
"passed": 5,
"total": 8,
"has_h1": true,
"has_h2": true,
"has_h3": true,
"has_mermaid": true,
"has_citations": false,
"has_table": false,
"has_conclusion": false,
"word_count_ok": true
} | {
"total": 0,
"validity_rate": 0,
"problems": []
} | {
"fidelity_score": 2,
"supported_claims": 8,
"unsupported_claims": 12,
"notes": "The wiki page makes several claims about the date-fns architecture that are not supported by the provided context, including incorrect component interactions, entry points, and design decisions, while correctly identifying some pa... |
sw-coder-14b-v3 | date-fns/date-fns | data_flow | stop | # date-fns System Overview
## Overview
date-fns is a popular JavaScript library for date and time manipulation. The system is organized into several packages, including core functionality, timezone handling, and UTC date utilities. The core package provides the main date manipulation functions, while the tz and utc p... | {
"passed": 6,
"total": 8,
"has_h1": true,
"has_h2": true,
"has_h3": true,
"has_mermaid": true,
"has_citations": false,
"has_table": false,
"has_conclusion": true,
"word_count_ok": true
} | {
"total": 0,
"validity_rate": 0,
"problems": []
} | {
"fidelity_score": 3,
"supported_claims": 12,
"unsupported_claims": 8,
"notes": "Many claims about parsing mechanisms, locale support, and timezone handling are either inaccurate or not fully supported by the provided code context, particularly regarding the core parsing implementation details and specific tes... |
End of preview.
YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
SemanticWiki-Eval v3
An evaluation harness for repository-level architectural wiki generation with
verifiable citations. Most LLM documentation benchmarks measure format compliance;
this one additionally measures whether every path:line citation actually resolves in
the real repository, and whether the page's claims are factually supported by the code.
Protocol (scaled paired study, primary results)
- Held-out repos: 50 real GitHub repositories (permissive licenses: MIT / Apache-2.0
/ BSD-3-Clause; go 17, rust 16, typescript 13, python 4), discovered via GitHub
search and excluded from the 100-repo training registry. Pinned set:
eval_repo_list.json. The identical set is used for every arm (paired design). - Task: 2 categories per repo (architecture_overview, data_flow) = 100 pages per
model, temperature 0.7 / top_p 0.8, max 4096 output tokens, identical system prompt
and
<START_OF_CONTEXT>code-context format for every arm. - Scoring axes:
format_score- 8 structural checks (H1/H2/H3, Mermaid, citations present, table, conclusion, length window).citation_validity- deterministic: every citedpath:start-endmust exist in the repo snapshot, be in bounds, and cite non-empty lines (sw_pipeline.verify_citations).fidelity_score- LLM judge (Qwen/Qwen3-Coder-30B-A3B-Instruct, temperature 0), 1-5, factual support against the source context.
- Statistics: page-level means with 95% percentile-bootstrap CIs; finetuned-vs-base deltas tested with a paired sign-flip randomization over (repo, category) cells (100 cells per family, 2000 resamples, two-sided p).
Results (50 held-out repos, 100 pages per model)
| Model | Format | Citation validity | Fidelity (1-5) | Citations/page |
|---|---|---|---|---|
| semanticwiki-coder-14b-v3 | 0.823 [.809,.838] | 0.694 [.609,.782] | 3.51 [3.41,3.62] | 7.0 |
| Qwen2.5-Coder-14B-Instruct (base) | 0.784 [.770,.797] | 0.315 [.229,.409] | 3.17 [3.06,3.29] | 3.4 |
| semanticwiki-14b-v3-qwen3 | 0.780 [.761,.797] | 0.417 [.324,.508] | 3.29 [3.19,3.40] | 3.9 |
| Qwen3-14B (base) | 0.830 [.816,.844] | 0.707 [.620,.789] | 4.11 [3.97,4.24] | 7.0 |
Paired finetuned-vs-base deltas (sign-flip test, 100 cells, p < 0.001 for all)
| Family | Axis | Delta (FT - base) | 95% CI |
|---|---|---|---|
| Qwen2.5-Coder-14B | citation validity | +0.379 | [+0.267, +0.488] |
| Qwen2.5-Coder-14B | format | +0.039 | [+0.020, +0.058] |
| Qwen2.5-Coder-14B | fidelity | +0.34 | [+0.20, +0.49] |
| Qwen3-14B | citation validity | -0.291 | [-0.416, -0.165] |
| Qwen3-14B | format | -0.050 | [-0.074, -0.028] |
| Qwen3-14B | fidelity | -0.82 | [-0.98, -0.66] |
Findings
- The same SFT recipe improves one 14B family and significantly degrades the other. On Qwen2.5-Coder-14B, fine-tuning on verified-citation data raised citation validity by +0.379 (0.315 -> 0.694). On Qwen3-14B, the identical recipe lowered validity by -0.291 (0.707 -> 0.417) and fidelity by -0.82 points. Every 95% CI excludes zero (p < 0.001, n=100 paired cells per family). The un-finetuned Qwen3-14B is the strongest model in the study.
- Mechanism hypothesis: SFT helps where the target behavior is absent (the Qwen2.5-Coder base cites sparsely and invalidly, 3.4 cites/page at 0.315) and hurts where it would replace a better existing policy (the Qwen3 base already cites densely and validly, 7.0 cites/page at 0.707). 313 verified examples x 2 epochs appear to have overwritten the base's stronger citation discipline with the teacher's narrower style - consistent with LIMA-style "SFT teaches style" and SFT-memorizes findings, and a counterexample to the assumption that more SFT data on a strong base is never harmful.
- Distillation leaves a validity gap. Teacher dataset pages verify at ~1.0 after repair; students transfer to 0.42-0.69 on unseen repos.
- Category asymmetry: citation validity is systematically higher on data_flow pages than architecture_overview pages (e.g., 0.846 vs 0.542 for coder-14b-v3) - overview pages span the whole repository and offer more opportunities to over-claim.
- Format is saturated and uninformative (0.78-0.83 across all arms): the deterministic citation axis is what separates models, which is the argument for verifiable-citation benchmarks.
An earlier preliminary study (13 held-out repos, python-heavy) is superseded by this one but preserved in git history; its direction agreed on the coder family and on the qwen3 regression.
Limitations
- The judge shares its model family with the Qwen3 student base; note, however, that the regression on Qwen3 is largest on the judge-independent citation axis.
- The held-out set skews toward go/rust/typescript (python 4, java 0).
- Verification checks file existence, line bounds, and non-empty cited lines - not semantic entailment of the claim by the cited lines.
- Single teacher, single LoRA rank, single seed, temperature 0.7.
Files
eval_pages_finetuned.jsonl/eval_pages_baseline.jsonl- all 400 pages with per-page scores.eval_summary_finetuned.json/eval_summary_baseline.json- aggregates, bootstrap CIs, per-category breakdown.eval_stats_paired.json- paired sign-flip tests (the table above).eval_repo_list.json- the pinned 50-repo held-out set.eval_repo_snapshots.tar.gz- held-out repo snapshots for re-verification.- Harness:
GhostScientist/semanticwiki-dataset-v3-scripts/sw_eval.py; citation verifier:sw_pipeline.pyin the same repo.
Provenance
Training data: GhostScientist/semanticwiki-data-v3 (97 real repos, teacher Qwen/Qwen3-Coder-30B-A3B-Instruct, citation-repair loop). Pipeline and training scripts: GhostScientist/semanticwiki-dataset-v3-scripts.
- Downloads last month
- 88