Open source · Framework projectsDownload library

Good engineering.
Made to be reused.

Make a task explicit. Inspect a dataset with clear rules. Start with small, complete engineering units that can be read, run and improved.

Explore the projects
task.graphLocal run illustration
01
Read datasource
02
Validatevalidate
03
Aggregateaggregate
04
Reportreport
3 synthetic records2 regional totals
Source, tests, documentation and a license. Together.JavaScript / JSON / Markdown

An open starting point,
in code you can understand.

These independent references focus on two foundations often rebuilt from scratch: task dependencies and data quality. Run them offline without an account, an API key or third-party dependencies.

They are starting points for extension, not the production ID Axis engine or HuanYu model weights. Each README explains its implementation and limitations.

Start with the foundations.

02 reference projects

Task Graph Kit

Make task dependencies explicit.

Task ordering, bounded retries, failure propagation and structured run records. A small graph runner you can read and execute.

  • Dependencies complete before consumers
  • Failures are recorded; dependent work is blocked
  • Opt-in retries for explicitly idempotent work
Download sourceZIP · 4.5 KB
example.mjs
import { runGraph } from './graph.mjs';
const report = await runGraph([
 {id:'source',run:()=>[{region:'North',amount:120},{region:'North',amount:80},{region:'South',amount:150}]},
 {id:'validate',dependsOn:['source'],run:({source})=>{if(source.some(r=>!Number.isFinite(r.amount)))throw new Error('Invalid amount');return source;}},
 {id:'aggregate',dependsOn:['validate'],run:({validate})=>validate.reduce((a,r)=>({...a,[r.region]:(a[r.region]??0)+r.amount}),{})},
 {id:'report',dependsOn:['aggregate'],run:({aggregate})=>({unit:'CNY',synthetic:true,totals:aggregate})},
]);
console.log(JSON.stringify(report,null,2));

Explicit dependencies organise work. Structured outputs connect the next step.

sample-invalid.jsonl3 issues found
02regionRequired value missing
02amountType mismatch
02idDuplicate key
JSONL → JSON

Turn “something is wrong with this data” into specific, actionable findings.

Dataset Quality Kit

Locate data issues by row.

Check required JSONL fields, types and unique keys against explicit rules. Receive a machine-readable issue report.

Valid and invalid examples are bundled. Findings omit source values and can feed an automated workflow. Extend more complex rules from this transparent base.

Download sourceZIP · 4.7 KB

Read it.
Run it.
Make it better.

A useful reference does not hide its decisions. It makes every behaviour explainable.

  1. 01

    Read the scope

    Start with the README: inputs, runtime and what is intentionally absent.

  2. 02

    Reproduce the example

    Run the bundled synthetic data, then inspect failing cases and test output.

  3. 03

    Extend with a question

    Change one clear behaviour and add tests and documentation that demonstrate it.

Let one piece of work
become the next starting point.

Browse all downloadsLearn how to contribute

These reference kits use the MIT license. See the LICENSE included in each archive for the exact terms.

鲁ICP备2024109755号-2
Drag to move. Right-click, touch and hold, or press Shift+F10 to choose a corner.