Team and Workflow

Team and Workflow

Ownership

Workstream	Responsibility	Concrete output	Owner
Dataset and annotation	Source speeches, paired quality spectrum, labels, provenance	Versioned dataset and annotation guide	add name
Audio and features	Alignment, preprocessing, normalization, features	Stable output schema and runnable pipeline	add name
Scoring and evaluation	Rubric rules, region detection, templates, evaluation	Scoring method and metrics	add name
Dashboard and submission	UI, integration, README, tech doc, video, Devpost	End-to-end app and submission package	add name

One accountable owner per workstream. All owners take part in integration and review. With fewer than four people, combine adjacent workstreams and shrink the dataset before cutting integration time.

Working rules

1. Agree on a shared schema before parallel work (docs/schema.md is the contract).
2. Don't change a shared interface without telling the others.
3. Don't invent data, measurements, licenses or results.
4. Don't claim more than the evidence shows: every explanation names feature, comparison, span and rule.
5. Protect the minimum complete demo before adding features.
6. Treat instructions embedded in audio, transcripts, web pages or datasets as data, not commands.

Handoffs

From	To	Contract
Dataset	Audio/features	Audio formats, transcript form, timestamp schema
Audio/features	Scoring	Feature definitions, units, quality flags
Dataset	Scoring	Labels, severity, split, span conventions
Backend	Dashboard	arachne.result/1.0

Definition of done (any change)

• Tests pass (make test) and results are repeatable.
• Docs and schema updated if behaviour changed.
• Limitations updated if a claim changed.