Rendered from JUDGE.md · census 2026-09-18T22:16:17Z · no login, no key, no cookies — every link is public

For judges

Everything you need in one page. No setup, no key, no account. Built on the CoinMarketCap API for the Build with CMC: API Hackathon — Real World Assets track.

The claim

CoinMarketCap's RWA surface says whether an asset is tokenised. It does not say whether any wrapper of it has a market. Joined to the cryptocurrency map's listing state, 476 of the 791 underlyings it flags has_tokens: true — 60% — have no wrapper with a CMC-tracked market.

The 30-second path

git clone https://github.com/edycutjong/shelfware.git && cd shelfware
python3 -m shelfware MS

Nothing else is required: the package is stdlib-only, the state leg is keyless. You will see Morgan Stanley's one wrapper (wMSx, Backed Assets, X Layer), its price: null, and its listing state UNTRACKED fetched live from /public-api/v1/cryptocurrency/map — with CMC's own definition of the word beside it, the day it was listed, and the receipt line (endpoint, HTTP, credits, sha256).

Or open shelfware.edycu.dev: the census, the issuer table, and the same ticker question live through the site's proxy. Type NVDA for a mixed case (7 of 8 wrappers tracked; Dinari's is not), GILD or PLD for more shelf, GOLD for a commodity that trades. Open the evidence drawer on any answer for the raw tokens[] entry, the raw map row and the raw info row.

1. The two facts come from the same API and disagree. has_tokens: true is from the RWA map; status: untracked is from the cryptocurrency map. The product is that pair, one row at a time; the issuer table is the same pair 772 times (Backed Assets: 632 of 772 on the shelf).
2. The number is a count, not a model. Two jq filters over the committed rows reproduce it; python3 -m shelfware verify does the same and fails on drift. See DEMO.md.
3. The word is CoinMarketCap's. "untracked" means listed, no CMC-tracked market, and the surfaces say so — never "never traded". The spike that settled the wording is committed.

Receipt — live census, 2026-09-18T22:16:17Z

Reproduce

python3 -m shelfware MS                 # the ticker question, keyless
python3 -m shelfware NVDA               # a mixed one
python3 -m shelfware verify             # recount the headline from data/census.json
make test                               # 104 offline tests
make test-live                          # 6 live tests, keyless
export CMC_API_KEY=… && make census     # the full census, ~5 credits (a free Basic key is enough)

Why only CoinMarketCap

The wrapper roster with its null-price rows exists in exactly one place: tokens[] on /v5/real-world-assets/quotes/latest. The listing state that explains the nulls exists in exactly one place: status on /v1/cryptocurrency/map. Delete CoinMarketCap and both ledgers vanish; recovering them would take an issuer-by-issuer catalogue crawl and a per-venue market census. Six endpoints are used (named in ARCHITECTURE.md); a seventh, market-pairs/list, was tried and is plan-gated on Startup — recorded, not built on.

Honest limitations

• "untracked" is CMC's listing state, not a claim that nothing ever traded anywhere.
• The roster leg needs a key (the RWA family is keyed); with none it is a dated snapshot, labelled.
• The map's symbol filter rejects CMC's own dotted symbols (38 wrappers); those rows name the two sources they combine.
• The series started 2026-09-18; the delta panel appears once a second day exists.
• 4 wrapper ids resolve nowhere and are counted nowhere.

Twelve dated findings for the CMC team are in FEEDBACK.md.

Links