Methodology — what was checked, what was not, and how to reproduce it.

1. What was crawled

• Nifty Gateway product index. The unauthenticated search endpoint at search-api.niftygateway.com/api/v2/indexes/products/search, swept with every two-letter (aa through zz) and two-digit query term and deduplicated by contract address. Used to build the project catalogue at /data/nifty-gateway-catalog.json.
• Ethereum mainnet RPC. Public RPC endpoint https://ethereum-rpc.publicnode.com. Used to (a) call selector 0xbe772d4c (niftyRegistryContract()) on each contract and (b) fetch deployed bytecode for each contract for literal scanning. No archive node required.
• Internet Archive · Wayback Machine. Two captures of the Nifty Gateway Terms of Use page: 24 July 2023 (capturing the "Last revised: December 27, 2022" revision) and 29 December 2025 (capturing the "Last revised: January 18, 2024" revision). Print-to-PDF copies are mirrored locally in /docs/ so the dossier remains readable if either capture moves.
• Closure-notice capture. Wayback capture dated 20 March 2026 of the same TOS page, preserving the closure banner that announces the 23 April 2026 platform shutdown and the 24 January 2026 transition to withdrawals-only mode.
• Delaware Division of Corporations. Public corporate-entity database for Nifty Gateway LLC, used for state-of-formation, formation date, and the franchise-tax / annual-report obligations applicable to a Delaware LLC.
• Contemporaneous press. Trade publications (CoinDesk, The Block, Decrypt, Artnet) and corporate press releases used only as corroboration for dated events (acquisition, shutdown announcement, withdrawals-only mode). No press claim is treated as primary evidence; press citations support claims that are independently verifiable from other sources.

2. What was sampled

• The 10,090-contract universe. The set of distinct on-chain BuilderInstance addresses surfaced by the alphabet-sweep above. The platform's own meta object reports approximately 10,128 collections; the alphabet-sweep recovers 10,090. The remaining ~38 collections are not surfaced by any two-character substring used in the sweep and are excluded.
• Registry-gate scan: all 10,090 contracts. Every contract in the universe was probed by both the niftyRegistryContract() selector call and the bytecode-literal scan. Sample = population. No sampling bias possible at this stage.
• Terms-of-Use clause subset. The /their-own-words page quotes 14 specific clauses from the TOS (§ 2, § 4, § 11, § 12, § 17, § 18, § 22, § 23, § 26, § 27, § 28). These are not all of the TOS; they are the clauses materially relevant to the four-layer technical-dependency model on /trapped. The full document is preserved as a downloadable PDF in /docs/ so the reader can verify nothing material was omitted.

3. What was directly observed

Anything tagged kind: observed in the claims index. Includes: the on-chain bytecode of each contract, the response of selector 0xbe772d4c, the response of tokenURI() for sampled contracts, the verbatim text of the captured Terms of Use, the closure banner on niftygateway.com, and the corporate-entity record at the Delaware Division of Corporations. Each observed claim names the exact endpoint, capture date, or document.

4. What was inferred, and from what

Anything tagged kind: inferred in the claims index. Three claims fall in this category. Each names the derivation explicitly. Aggregate primary revenue (NG-FIN-001) is the deduped sum of (units sold × lowest minted price) per contract across the catalogued cohort, treated as a strict lower bound. The annual breakdown (NG-FIN-004) and the year-by-year registry-gate decline (NG-TECH-005) are direct consequences of the same data plus a date filter. No model, no projection, no extrapolation.

5. What was explicitly NOT verified

• Internal Nifty Gateway / Gemini Trust Co. accounting. Only the platform and its parent hold the complete sales ledger. The dossier never claims to know what they internally recorded, what they were paid in fees, what they paid out to artists, or what they remitted in tax.
• Off-chain primary settlements. Drops settled outside the on-chain mint (most notably the Beeple Spring Collection auctions, where the catalog records primaryUSD: null) are excluded from the aggregate. Their independently corroborated totals appear in the dossier as labelled, separately-tagged figures, not folded into NG-FIN-001.
• Per-token transfer-gate behaviour at any future moment. The registry-gate scan establishes the necessary conditions for the standard NG transfer-gate pattern; it is not a formal proof that any specific contract's transfer function defers to the registry on every call. Per-token certainty requires reading that specific contract's source or disassembly.
• Future setBaseURI() actions. The dossier documents that the function exists and is owner-callable. It makes no prediction about whether any specific contract owner will or will not call it, or what URL they will set if they do.
• Intent. The dossier makes no claim about anyone's intent — neither the platform's, nor the parent's, nor any individual's. NG-EDIT-001 is the single editorial framing in the index, and it is explicitly tagged as such.

6. Known limitations

• The platform's product index drives the contract universe; ~38 collections (~0.4% of the platform's own count) are not recoverable by alphabet-sweep and are excluded.
• Aggregate primary revenue (NG-FIN-001) is a strict lower bound: 33% of contracts have no recoverable USD figure in the platform's index. The true total is very likely higher.
• USD figures published by the platform's index are drop-time prices; ETH-denominated drops were converted by the platform at a rate the platform set. The dossier does not re-price at any other rate.
• The Wayback captures of the Terms of Use are themselves snapshots; intermediate revisions between 27 December 2022 and 18 January 2024 are not analysed.
• The closure banner at /docs/ngc-2026-04-23-shutdown.png was captured on 20 March 2026; the dossier does not assert whether the banner's exact wording has changed since.

7. Dataset versioning

• /data/nifty-gateway-catalog.json · 10,090 contracts · last enriched on the date in the file's _meta.lastEnrichedAt field.
• /data/registry-gate-scan.json · 10,090 contracts · last scanned on the date in the file's _meta.scannedAt field.
• /data/claims.json · atomic claims index · version recorded in _meta.version; per-claim version and lastVerified dates track individual updates.
• Withdrawals or revisions to claims are tracked via supersedes / supersededBy pointers in the JSON; IDs are not recycled.

8. Reproducibility

Anyone can repeat any step in this methodology with the public inputs named above and a few hours of compute. The on-chain checks require only a public Ethereum RPC and the JSON-RPC spec. The product-index crawl requires only a polite rate limit and the alphabet sweep described in section 1. The Wayback captures are dereferenceable URLs; the PDFs in /docs/ are byte-identical to those captures. The corporate-entity record is free to look up. The atomic-claims index at /data/claims.json contains, for every numerical claim in the dossier, the exact file, field, and method needed to recompute it. Independent verification, contradiction, or correction is welcomed; the point of the index is to make it cheap.