Atlas Trust

Trust in Atlas is explicit, verifiable, and temporary. Verified humans allocate trust on each other across specific topics. Allocations mature on a rolling timeline; superseded versions drop off automatically. Witnessing makes the whole process publicly verifiable.

In real life, trust is specific — you trust different people on different things. Atlas brings that distinction online: trust on climate science doesn't make you trusted on software security.

Trust Allocation

Every verified participant holds 100 trust allocations — a continuous budget you distribute across people and topics. Each allocation pairs a target identity with a topic string and a numeric score. Re-allocating publishes a new envelope that supersedes the previous version once it matures.

{
  "@context": "https://schema.org",
  "@type": "ReviewAction",
  "additionalType": "atlas:trustAllocation",
  "object": {
    "@type": "Person",
    "identifier": {
      "@type": "PropertyValue",
      "propertyID": "atlas:hash",
      "value": "<target-identity-hash>"
    }
  },
  "resultReview": {
    "@type": "Review",
    "reviewAspect": "photography",
    "reviewRating": {
      "@type": "Rating",
      "ratingValue": 15
    }
  }
}

Allocations are published as ReviewAction envelopes with additionalType: "atlas:trustAllocation". They become valid only after being witnessed by the network.

Square-Root Scaling

Attestation power is the sum of square roots of each individual allocation received. This means distributed trust always beats concentrated trust — even when the same total points are spent.

Same total investment, dramatically different power. One person giving 100 points produces only 10 power, while 100 people giving 1 point each produces 100. Getting many independent verified humans to allocate trust is far harder than convincing one — the square root makes broad consensus the only path to real influence.

Competences

Trust allocations are materialized into competence scores — per-identity, per-topic aggregates that registries attach to every PresentedEnvelope. Clients use competences to filter and rank content by the author's demonstrated expertise.

{
  "topic": "climate",
  "score": 12,
  "identityHash": "b3a1f8..."
}

Rolling Recognition

Trust allocations don't take effect instantly — and they don't last forever. Every allocation carries its own publication timestamp and matures individually. Three independent parameters, all governed by Legislation, shape the rhythm:

• maturationPeriod — how long an allocation waits before counting (30 days by default). This is the anti-Sybil defense: anyone has the maturation window to flag a fraudulent allocation before it counts.
• snapshotInterval — how often the recognized trust graph is re-measured for governance reads (~daily by default). Controls measurement stability, not security.

// Each allocation matures individually after the maturation period
// (default 30 days, configurable via Legislation)

Published t=0  →  Maturing (~30 days)  →  Recognized at t = T_mat

// Revisions supersede the previous allocation once the new one matures
// Revocations propagate within one snapshot interval (daily by default)

// Attestation Power = Σ √(each mature allocation)
// 4 people × 25 points = 4 × √25 = 20  >  1 person × 100 = √100 = 10

The delay prevents last-minute manipulation and gives the network time to verify every allocation individually. Revisions only take effect after they themselves mature, so power transitions are gradual — never a synchronized spike, never with a stale calendar lag. Revocations propagate within one snapshot interval.

Negative Trust

Trust isn't only positive. You can allocate negative trust on bad actors — a public signal that starves them of attestation power and makes their content filterable across the network.

When an identity's integrity score exceeds the negative standard threshold (square root of the second-highest integrity scorer), the identity is banned from issuing TradeAction envelopes and excluded from burn calculations.

Negative trust expires over time as the underlying trust envelopes reach their one-year validity limit. Changed behavior leads to natural recovery — more forgiving than permanent bans, more honest than shadow-banning.