Reputational system of participants

1) Why does the ecosystem need a reputation

• predictable limits, routing priorities, and credits/penalties;
• fewer controversial cases and faster reconciliation;
• fair RevShare (contribution to × quality) with transparent logic.

2) Ontology and units

Сущности: `participantId`, `role` (operator/studio/psp/kyc/affiliate/etc.) , `jurisdiction`, `trustTier`, `score`, `decayAt`.

• SLO/SRE: integration uptime, p95/p99 API, bus lag, MTTR, webhook delivery.
• Attribution/quality: accuracy/timeliness of postbacks, controversy, share of duplicates.
• Compliance/security: PD incidents, sanctions fouls, SoD/keys/rotation.
• RG (responsible game): triggers, guardrails observance, correct segmentation.
• Finance: chargebacks, correctness of reports, timely payments, Cost-to-Serve.
• Audit/tracing: completeness of trace/logs, "time per trace packet."
• Partnership: Speed of incident response, quality of communications (SLA responses).

All definitions are fixed in the metrics directory (versions, windows, owners).

3) Scoring model

3. 1 Basic formula

[
Score = \sum_k w_k \cdot N_k - \sum_m p_m \cdot V_m
]

where (N_k) - normalized positive signals (SLO, attribution accuracy, etc.), (V_m) - violations (PD incidents, RG fouls, disputes), (w_k,p_m) - weights approved by the ecosystem council.

3. 2 Quality "as multiplier"

[
Q = w_{slo}! \cdot! SLO + w_{attr}! \cdot! ATTR + w_{rg}! \cdot! RG + w_{sec}! \cdot! SEC
]

• in RevShare network splits: (share_i\propto CT_i\cdot Q_i);
• in traffic prioritization/limits/credit lines.

3. 3 Attenuation and incidents

Exponential decay for events older than the window (for example, 90 days).
Error budget: auto-malus → exceeded before recovery.
Amnesty: based on RCA and conformance test results.

4) Trust Tiers and Rights

Automatic transitions between levels - according to the schedule and/or when the rules are triggered.

5) Application of reputation in operations

Traffic routing: prioritization of nodes with high'Q ', auto-cooling at p95/errors.
Quotas and capping: dynamic limits of campaigns, rates, APM routes.
RevShare splits: adjustment of shares by the quality multiplier.
Credit lines and NET conditions: decrease/increase in holdings.

Access to experiments: "pilots first - T3/T4."

Showcases/catalogs: sorting providers/affiliates by Score with badges.

6) Provability and protection against manipulation

Data oracles: signed summaries (SLO/finance/attribution) with 'traceId' and versioned formulas.
WORM audit: immutable logs of recalculations, changes in weights and initial events.
Anti-Sybil: KYP/KYB, payment/contract connectivity, mirror node detector.
Anomaly-detection: bursts of ideal metrics, geo/ASN mismatch, unnatural lags.

7) Dashboards and scorecards

• Score, Tier, trend, contribution to NetRev, SLO/ATTR/RG/SEC sub-scoring;
• controversial cases, sanctions/bonuses, SLA "time for a trace package," Tier forecast.

Ecosystem Operational Overview: Ranking by Role/Region, Tier Distribution, Risk Map, Heatmap SLO, List of Nodes on "Cooling."

SLO showcases: freshness ≤ 1-5 s, p95 render ≤ 1. 5–2. 0 s, uptime ≥ 99. 9%.

8) Sanctions and bonus policies

Auto-bonus: stable SLO> target N periods → increased limit/decreased hold.
Auto-malus: RG or PD-incident → temporary pause, lower Tier, RevShare recalculation.
Grace-period: short-term failure without consistency - soft penalty/warning.
Recidivism-curve: increasing penalties for repetition within the window.

9) Appeals and transparent changes

Appeal: SLA for parsing, mandatory trace package, sandbox-replay calculation.
Public changelogs: versions of weights/formulas, date of entry, A/B effects.
Justified exception: temporary relaxation of rules with TTL and automatic withdrawal.

10) Rights/policies integration

• 'trustTier ∈ {T1.. T4} ',' auditReadiness = true/false ',' rgComplianceScore≥X '- access conditions to limits, liquidity pools, campaigns;
• deny-gardrails are not removed by reputation (deny-first).

11) Anti-patterns

One total score without breakdown: it is impossible to understand the source of the problem.
"Drawn" weights without versions/changelogs: controversy and mistrust.
Ignoring RG/compliance: high traffic masks the risk of fines.
Punishments without a recovery window: "sticky" sanctions → network stagnation.
Offset stories and manual exports without signatures: holes and takes.
SPOF-estimator: central service without N + 1/DR.

12) Implementation checklist

1. Catalog of signals: definitions, owners, windows, sources.
2. Formulas/weights: semver, changelogs, sandbox validation.
3. Oracles and WORM audits: signed summaries, tracing.
4. Showcases and APIs: scorkard, rating, filters, event webhooks.
5. Tier/limits policies: up/down auto rules, stop buttons.
6. Appeals: Package template, SLA, settlement replay.
7. Anti-manipulation: Sybil-detection, anomalies, honey A/B tests.
8. Integrations: routing, RevShare, credit policy, campaigns.
9. DR/chaos: Regular evaluator/oracle failure drills.

13) Maturity Roadmap

v1 (Foundation): basic Score + Tier, SLO/ATTR signals, manual reviews.
v2 (Integration): auto-limits/priorities, sanctions/bonuses, public scorecards.
v3 (Automation): predictive risk rate, dynamic weights, feedback-control on routing.
v4 (Networked Governance): federated exchange of reputation between chains, DAO rules of weights and transparent treasuries.

14) Success metrics

Risk/quality: reduction of disputes, MTTR incidents, share of nodes with SLO ≥ goals.
Business: uplift CR/FTD/ARPU/LTV by prioritizing quality nodes.
Compliance/RG: PD incidents = 0, decrease in RG flags/1k active.
Economics: Cost-to-Serve ↓, share of auto-solutions ↑, predictability of ↑ cache.
Partnership: time to provide a trace package, the share of T3/T4 in turnover.

Brief Summary

The reputation system is a framework of network trust: formalized signals, formula versions and provable calculation, automatic levels and limits, integration into routing and RevShare, appeals and transparency. Make reputation a protocol attribute (not an opinion) - and the ecosystem will start to scale faster, safer and more profitable.