docs/resource_schema.md

Resource Schema

Every entry in data/resources.yml is a YAML mapping. Only a few fields are required; the rest are optional and added when known. The validator (scripts/validate_catalog.rb) enforces the required fields and the format of the optional ones it understands.

Required fields

Field	Type	Notes
name	string	Unique across the catalog. Matches the README entry.
kind	enum	One of dataset, benchmark, model, toolkit, survey, challenge, collection.
status	enum	One of open, request, benchmark, partial, watch. See status_policy.md.
url	http(s) URL	Primary official link (project page, GitHub, Hugging Face, arXiv, OpenReview, university page, Zenodo).
tasks	list of strings	Non-empty. Reuse tokens from taxonomy.md where possible.

Common optional fields

Field	Type	Notes
year	integer	Release/publication year (2000–next year).
released	YYYY or YYYY-MM	First public/arXiv release date; the exact month is taken from the arXiv id where available, otherwise the year. Shown as the README "Released" column.
scope	enum	egocentric (default if absent) or adjacent.
scale	string	Scale/signal summary. Use numbers from the official source; do not inflate.
modalities	list of strings	Reuse modality tokens from taxonomy.md.

Link fields (validated as http(s) URLs when present)

Field	Notes
paper	Paper/arXiv/OpenReview link.
project	Project or dataset homepage.
code	Code repository.
data	Direct data/download link (e.g., Hugging Face, Zenodo DOI).
benchmark	Benchmark/challenge page.
leaderboard	Public leaderboard.
mirror	Unofficial mirror (clearly labeled as such).
access_url	Access/license request page.
license_url	Canonical license text.

Provenance and access fields (free text unless noted)

Field	Type	Notes
license	string	License identifier (e.g., cc-by-nc-4.0, mit, other).
access	string	Access friction note (e.g., "form + approval").
verified_at	date YYYY-MM-DD	When access/status was last confirmed.
release_note	string	What is still missing or pending for watch/partial.
scope_note	string	Why a resource is in or out of egocentric scope.
modality_notes	string	Extra modality detail not captured by tokens.
raw_video_dependency	string	Whether the resource depends on another dataset's raw video.
citation_key	string	Stable BibTeX-style key.
created_by	string	Authors/creators.
released_by	string	Releasing org/channel.
publisher	string	Hosting platform.
derived_from	string	Parent dataset this is built on.
milestone_origin	string	Short institution, lab, company, project, or consortium label shown on milestone cards when the entry has milestone.

Minimal example

- name: EgoNight
  kind: benchmark
  year: 2025
  status: open
  url: https://insait-institute.github.io/EgoNight/
  paper: https://arxiv.org/abs/2510.06218
  code: https://github.com/dehezhang2/EgoNight
  scale: "EgoNight-VQA (3,658 QA over 90 videos, 12 QA types) plus day-night retrieval and depth"
  modalities: [egocentric-video, qa, depth]
  tasks: [egocentric-vqa, egocentric-depth, retrieval]
  scope_note: "Low-light / nighttime robustness benchmark."
  verified_at: "2026-06-15"
  citation_key: egonight_2025

Validation

ruby scripts/validate_catalog.rb

Errors fail the build (missing required fields, bad enum values, malformed URLs or dates). Unknown task/modality tokens produce a non-fatal warning so the vocabulary can be consolidated in data/taxonomy.yml.