CONTRIBUTE

Help build the platform

EvoKernel Spec is the open knowledge graph mapping any model onto any hardware. Today: 28 vendors · 39 accelerators · 19 models · 22 measured cases. Every number cites a source. Three contributor tracks below.

tier: official
🏷

Vendor official data

You represent NVIDIA / AMD / Huawei / Cambricon / etc.

Add or correct authoritative numbers for your products. Cite a whitepaper or datasheet; reviewers verify your association via vendor PR channel. Your contributions get the vendor-attested label.

Submit vendor claim →
tier: estimated → official
🛠

Community data correction

Anyone with a public source — a paper, blog, GitHub repo, MLPerf result.

Fix a wrong number. Add a missing field. Replace a 404 evidence URL. Same review path as vendor claims; tier assigned per source-type matrix in our policy.

Report a correction →
tier: measured

Reproducible deployment case

You ran a real model on real hardware and have logs to prove it.

A measured case enters the Calibration Map immediately and shifts every Tier 1 prediction for that hardware. The single highest-leverage contribution.

Submit a case →

Contribution lifecycle

  1. 1/5

    Pick the matching track above and open an issue (or fork → PR directly).

  2. 2/5

    Maintainers verify within 72 hours (vendor) / 7 days (community / case).

  3. 3/5

    Once approved, the YAML lands in `data/`; CI re-validates on every commit.

  4. 4/5

    Next build picks up the change automatically — `/showcase`, `/pricing`, calculator all refresh.

  5. 5/5

    Tagged release attributes you in `CONTRIBUTORS.md` and the GitHub Release notes.

Data tiering — what each label means

Every quantitative claim carries one of three tiers. The tier reflects evidence strength for that specific number, not the source document overall.

🏷

tier: official

Vendor-attested. Whitepapers, datasheets, MLPerf submissions signed by the manufacturer.

tier: measured

Independently observed on real hardware. Reproducible cases with logs + commit hash + scripts.

⚠️

tier: estimated

Derived, inferred, extrapolated. Calibrated against reality but not directly observed.

Full policy: docs/DATA-TIERING.md →

Skip the issue, send a PR

For experienced contributors: fork, branch, edit YAML, push, open PR. The issue template is just a guided form — power users save time by working directly in YAML.

# Add a new hardware card
git clone https://github.com/ying-wen/evokernel-spec
cd evokernel-spec
pnpm install
# create data/hardware/{vendor}/{slug}.yaml
# fill schema (see schemas/hardware.ts)
pnpm validate           # zod + cross-ref check
pnpm test               # unit tests
./launch.sh             # local preview before PR
git checkout -b add-{slug}
git add data/hardware/{vendor}/{slug}.yaml
git commit -s -m "data: add {slug} hardware"
git push -u origin add-{slug}
gh pr create

Detailed dev guide: docs/DEVELOPMENT.md →

Recognition & community