Design principles
Cost asymmetry
Every point must cost something real: money (gas, payments) or time (wallet age, sustained activity). Signals that can be mass-produced for free earn little or nothing. This is what makes the score expensive to farm.
Diminishing returns
Every signal is capped, and volume-based signals (transactions, gas, followers) are damped logarithmically or in tiers. The 1,000th transaction is worth far less than the 10th. Grinding one signal cannot dominate the score.
Nothing is purchasable
There is no way to buy points directly. Paid actions on SOCI4L (premium, custom slugs, donations) count because they are costly signals, not because they are payments to us. Their weights are capped like everything else.
One social account, one wallet
A verified social account can be bound to exactly one wallet, enforced at the database level. The same X or GitHub account can never vouch for a second address.
Transparent categories, evolving parameters
The signal categories, caps, and principles on this page are public and stable. Exact curve parameters may be tuned over time to counter gaming; material changes will be noted on this page.
Signal categories
On-chain signals apply to any address; no SOCI4L account needed. Profile and social signals require claiming the wallet's profile, which is where the strongest humanity evidence (verified social bindings) comes from.
On-chain signals: any address
Time since the address’s first C-Chain transaction. Accrues steadily and saturates after a few years. Old wallets cannot be mass-produced; age is the single hardest signal to fake.
Outgoing transaction count, logarithmically damped: the earliest transactions count most, and it takes exponentially more activity to keep gaining. Raw volume alone cannot dominate the score.
Cumulative gas the address has paid on the C-Chain, logarithmically damped. Gas is a direct, unrecoverable cost, the purest economic signal available on-chain.
Profile signals: claimed profiles
Wallet ownership proven by signature, the gate that unlocks every profile and social signal.
Basic identity completeness.
Basic identity completeness.
Each linked social profile adds a little, capped.
Each link on the profile adds a little, capped.
Social & economic signals
OAuth-verified accounts (X, GitHub, …) bound one-to-one to the wallet. The strongest humanity signal: an aged, active social account cannot be conjured retroactively.
Real on-chain tips sent to other profiles through the SOCI4L donation contract: a paid, timestamped economic signal that cannot be back-dated.
Tiered with steep diminishing returns: early followers count most, later ones far less. Uncapped but heavily damped, so follower farming barely moves the score.
Tiers
The S4 vector
One additive number is easy to read but easy to game, and it can't answer a B2B question like “is this an airdrop bot?” vs “is this a DeFi power user?”. So SOCI4L computes a four-dimension vector and a 0–100 composite. The composite is now the public headline score; the four dimensions are what B2B consumers threshold on. The original v1 number is kept internally for history and back-compatibility.
Driven mostly by verified social bindings and wallet age, the signals hardest to manufacture retroactively.
Transaction volume and recency, damped so raw count alone can’t carry it. Silence decays the dimension over time.
Follower standing, shrunk by graph-anomaly heuristics that flag follow rings, bursts and thin-follower armies.
Costly, timestamped actions on a log scale: gas paid, premium, a paid slug, donations. None can be back-dated.
Each address also carries a confidence value: signal classes we don't collect yet (behavioral naturalness, deep graph structure) lower confidence rather than being silently scored as zero. Low data means low confidence, never an unfairly low score.
The formulation
For readers who want the model itself. The forms below are public; the exact coefficients (weights, saturation scales, anomaly thresholds) stay private so the score can't be reduced to a farming checklist (see Known limitations).
The headline score is a convex blend of the four dimensions, humanity-led (ω_humanity > ω_social > ω_economic > ω_activity). The weights sum to 1; their exact values are withheld.
Each dimension is a weighted sum of its own normalized signals. Signal classes we don't collect yet are renormalized away, never scored as zero.
Volume signals saturate: the n-th transaction (or unit of gas) is worth roughly 1/n of the first, so grinding one axis decays fast. The scale s (where 'full' sits) is withheld.
Confidence is mean data-coverage across four signal classes (on-chain, social, graph, behavioral). Behavioral telemetry isn't wired in yet, so confidence is honestly capped below 1; low data lowers confidence, not your score.
Detected graph anomalies F (follow rings, follower bursts, thin-follower armies) lower our confidence and damp the social axis: floored, never a hard ban. The detection thresholds are withheld: publishing them is an evasion recipe.
The defensibility argument: accrued cost is time (wallet age, sustained activity) plus money (gas, premium, paid slug, donations). To reach a veteran's score, a fresh wallet must accrue comparable cost, and the time component can't be bought, only waited out. SOCI4L measures accrued cost; it never sells it.
These invariants aren't just stated; they're enforced as automated tests that gate any change to the model: a fresh wallet with a pasted bio can never outrank a multi-year, active, verified wallet, and anomaly flags always lower confidence. The math runs, it isn't marketing.
Verified Human #N
Separately from the score, an address earns a permanent sequential Verified Human number the first time it is seen with at least one OAuth-verified social account and real on-chain history. Numbers record honest chronology; they are assigned in qualification order, are never revoked or reassigned, and there is no cap. Earlier numbers simply mean earlier proof, not purchased status.
Data sources & freshness
- On-chain signals are read from the Avalanche C-Chain (RPC + public explorer data) and cached for 24 hours.
- Profile, social, and follower signals are read live from SOCI4L's first-party data.
- Daily snapshots record every profile's score over time; the time series itself is a future signal (sudden jumps are suspicious).
- Gas and history calculations scan up to the first 10,000 transactions of an address; beyond that, totals are conservative undercounts, and the transaction count itself remains exact.
Known limitations (v1)
Honesty over marketing: the current score does not yet read staking positions, contract diversity, NFT hold duration, or deep social-graph structure, and we do not publish a sybil catch-rate we haven't measured on a real address list. The headline number is now the S4 composite (0–100) shown above; the original flat v1 score is kept internally for history and back-compatibility. Weights are deliberately conservative while the on-chain signal set grows, and exact parameters stay private to keep the score expensive to game. When the scoring model changes materially, this page and its version change with it.