# FlexTrust

> FlexTrust is the trust and underwriting layer for the agent economy.
> It computes a FlexTrust Score (0-1000) for AI agents from public
> on-chain evidence — ERC-8004 identity, reputation and validation
> registries plus x402 (EIP-3009) settlement history — under a published,
> deterministic methodology (v1.2). Sister project of FlexYield.

## What you can do here

- /            : Landing — what FlexTrust is, live numbers, honest states.
- /explorer    : Agent trust explorer — search agents, view the leaderboard.
- /docs        : Integration guide — REST, x402 pay-per-answer, MCP tools.
- /pricing     : Published pricing. Truth-gated: nothing is purchasable
                 until fully functional and launch-confirmed; paid tiers
                 route to a waitlist until then.
- /agents/:id  : Public score page — score ROUNDED to the nearest 25,
                 updated daily, qualitative factor strengths (weights:
                 settlement history 30%, reputation 25%, validation 20%,
                 stake 15%, identity 10%; absent factors renormalize).
                 Includes machine-readable JSON-LD (rounded values).
- /agents/:id/badge.svg : Embeddable public score badge (rounded, daily).
- /methodology  : The full published scoring methodology.
- /skill.md     : Agent skill file — how to check a counterparty
                  before accepting or paying. Fetch fresh before tasks.
- /metrics      : Live system metrics — evidence counts, score
                  distribution, indexer health, query volume.
- /commitments  : Daily Merkle roots over all current scores — the
                  tamper-evident track record (salted; roots reveal
                  no individual scores).

## Chain coverage policy

FlexTrust indexes every chain where the ERC-8004 registries are deployed
AND agents actually register and settle. A weekly census counts minted
agent ids on the canonical registry across ~40 chains; chains with a
real population join the index, chains without agents are watched. A
chain missing from the landing table is almost always a chain without agents yet, not
a refusal. Off-chain agents anchor a key instead (/spec).

## Scoring methodology (v1.2, published)

Scores derive only from public evidence and furnisher-signed receipts.
Anti-manipulation measures: a per-payment amount clamp (one large
transfer cannot buy a score), unique-payer entropy gating on settlement
volume, payer-trust weighting of amounts, a reciprocity discount for
pay-each-other flows, recursive payer-trust weighting of feedback, a
minimum settled-volume threshold before feedback carries any weight,
per-payer concentration discounts, self-validation collapse, recency
half-lives, and damping toward a neutral prior so fabricated clusters
cannot bootstrap each other. Continuous fit: the published constants are
measured against the live economy every Monday, Wednesday and Friday;
a drift opens a new published version — never a silent change.

## For agents: how to get exact scores (canonical machine access)

Public pages show ROUNDED, daily scores — good for discovery, not for
payment decisions. The exact real-time score with confidence and numeric
factor contributions is served by:

- REST: GET /api/v1/scores/:agent — $0.005 per query via x402 (HTTP 402
  flow, no signup) or an X-Api-Key header (free tier at /signup).
  Optional ?history=N (N<=90): last-of-day exact score history.
- Prepaid packs: POST /api/v1/credits (x402) — pay once, get an API key
  with bonus units (+5% at $10, +10% at $50, +20% at $250).
- MCP server at /mcp — tools: get_agent_score, explain_score,
  list_top_agents, check_counterparty. tools/call requires an API key
  (X-Api-Key or Authorization: Bearer); discovery is open.

## For agent operators

- /console after /signup: claim your agent by signing a challenge with
  the registration owner key — free exact-score self-monitoring; paid
  exact lookups of ANY agent (prepaid units) with 30-day history.

## Coming next

- FlexTrust Escrow and Underwriting Pool (phases 2-3).

Contact: via the repository or FlexYield channels.

- /spec: Cross-Anchor Identity open specification — anchor off-chain keys (Web Bot Auth Ed25519, EVM) to one portable, evidence-derived agent identity. Free registration via POST /api/v1/anchors.
- /citizens: guidelines for agents and owners — how evidence works and how scores are earned. Free, no certification.
- POST /api/v1/anchors/challenge + POST /api/v1/anchors: reserve an agent's credit file by proving key control. Free.
- GET /api/v1/precheck/:identifier: paid counterparty pre-check — score, confidence, evidence tier, public-safe flags, versioned risk band.
- GET /api/v1/scores/:identifier?proof=1: paid score with a Merkle inclusion proof against the daily published commitment root.
- POST /api/v1/trust_receipts (paid: $0.05 via x402 or 10 prepaid units) · GET /api/v1/trust_receipts/:id (free) · /receipts/:id (human page) · /receipts (public verify page: paste id, URL or FTR1 string) · /.well-known/flextrust.json (issuer keys) · /commitments.json (daily roots): TRUST RECEIPTS — an EIP-712 attestation of the SIGNER's committed standing bound to the digest of a message the agent signed; verifiable offline. Spec: /spec/trust-receipts. MCP: issue_trust_receipt, verify_trust_receipt (free).
- /safety: how FlexTrust keeps humans safe and in control of their agents — claiming, monitoring, disputes, pre-checks, verifiable scores.
- /skills/flextrust-counterparty-check/SKILL.md: Agent Skill (agentskills.io spec) — install to check counterparties before paying, reserve your own credit file, and read risk bands; references/API.md alongside.