Méthodologie AUROS Activity Index

Dimensions

• Transfer frequency (weight 0.28) — 7 / 30 / 90 day transfer counts.
• Market depth (weight 0.22) — 30d volume vs float.
• Holder dispersion (weight 0.22) — active wallets vs top-holder concentration.
• Bid-ask spread (weight 0.13) — skipped when no secondary book (null is not zero).
• Temporal trend (weight 0.15) — vs prior edition; skipped on first edition.

Global score = weight-normalized mean of available dimensions only. Each dimension is stamped seed, live overlay, edition, or n/a so desks see what moved.

Overlays live

• Depth — CoinGecko market cap + 24h volume × 30 (proxy) when a coingeckoId resolves.
• Tokentx / top-10 — Allowlisted EVM ERC-20 via seeded contract, contracts map, or CoinGecko platforms. Preference: Ethereum first. Truncated samples are discarded for transfers. Ethplorer top-10 fallback is Ethereum-only.

• Ethereum
• Polygon
• Base
• Arbitrum One

• Rows without overlays stay seed_indicative. Mixed rows show mixed_public plus a live chip (chain · depth · tokentx · top10 · wallets). Non-EVM rails are not claimed.

Enveloppe de score

Each asset also shows a leave-one-dimension-out band: how far the global would move if one input dropped. That is a disagreement envelope, adjacent to conformal prediction — not a statistically guaranteed 95% interval. See /technology/methodology.

Badges

• Active — global ≥ 70
• Low activity — 40–69
• Dormant — global < 40

Badge changes vs the prior persisted edition appear on /activity with the previous edition id and date. Cite only listed moves — never invent an overtake.

Anti-manipulation (inputs réservés)

Live wash detection stays off (detectionLive: false). An allowlisted EVM tokentx sample (Ethereum, Polygon, Base, or Arbitrum via Etherscan API V2) may publish self-transfers (from = to), distinct token values, and wallets/tx ratios, and may raise sample-hint suspects when thresholds trip — still not a certified wash verdict. Unique counterparties in that sample are not a census. CoinGecko depth is not wash detection. No “Verified Active” claim from these counts. Off-chain dormancy is explained later by Settlement Reality (HITL, not a mint oracle).

Qualité des données

Two qualities can appear in the same edition. mixed_public means a public overlay landed: CoinGecko depth (24h volume × 30 / market cap) and/or tokentx transfer counts on an allowlisted EVM chain (Ethereum · Polygon · Base · Arbitrum) plus unique from/to in that sample when a contract resolves from seed or CoinGecko platforms and the page window is complete (truncated transfer samples are discarded). Chain preference: Ethereum first, then Polygon → Base → Arbitrum. Unique counterparties are not a full holder census. A public top-10 holder list (Etherscan V2; Ethplorer on Ethereum only) may overlay concentration % when it responds — never invented. seed_indicative means no public overlay. Sample counts (self-transfers, unique values) and sample-hint suspects appear when a tokentx sample is attached — detectionLive stays false. When the window is complete we also publish a raw uniqueWallets / transfers ratio for analysts — still not a wash verdict. Sequential editions persist so trend can activate across two snapshots. Non-EVM rails (Solana, etc.) are not claimed live.