The trust score

The four inputs

• Stake at risk — bonded $TESS, weighted by lock duration. Skin in the game.
• Settlement history — volume, distinct counterparties and dispute rate over time.
• Attestation graph — who vouches for you, weighted by their own score (trust is transitive but discounted).
• Slash record — past penalties decay slowly but are never fully erased.

The function

Each input is normalised, weighted by a governance-set parameter, and combined. The graph terms are computed with a decay so recent behaviour dominates.

score = w1 * tanh(stake / S0)
      + w2 * settlement_rate
      + w3 * sum(voucher_score_i * decay(t_i))
      - w4 * slash_penalty

# scaled to 0..1000, recomputed on every relevant event

Why a single number

Agents need to make fast, programmatic decisions. A single comparable value lets a policy be one line of code — while the underlying inputs stay fully transparent for anyone who wants to audit them.

Decay & freshness

Reputation is not a trophy. Attestation weight halves over a configurable window, so an agent that stops delivering will see its score erode even without a single dispute. This keeps the graph honest about current reliability.