Algorithm Explainability — GDPR Art. 22(3)
This document discloses the logic behind Creatorlayer's automated income verification and creditworthiness scoring system, as required by GDPR Article 22(3). Where a lender relies solely on Creatorlayer's automated output to make a credit decision, the lender is also required to give the data subject "meaningful information about the logic involved" (GDPR Art. 13(2)(f) / Art. 14(2)(g)).
What is computed and why
Creatorlayer processes creator-consented platform data to generate a Risk Tape — a structured, encrypted data record that includes:
- Income metrics — monthly revenue history, averages, medians, trends
- Risk metrics — income volatility, maximum drawdown, platform concentration
- Creator Score — a composite 0–100 quality-adjusted score
- Eligibility decision — a rules-based credit eligibility determination
No bank account data, transaction history, or non-consented data is used.
Input data
| Data source | What is collected | Legal basis |
|---|---|---|
| YouTube Analytics API | Monthly estimated revenue (OAuth-consented) | Consent (Art. 6(1)(a)) |
| Stripe Connect | Monthly payout amounts (OAuth-consented) | Consent (Art. 6(1)(a)) |
| Twitch | Subscriber counts, broadcaster type (OAuth-consented) | Consent (Art. 6(1)(a)) |
| TikTok | Follower counts, video engagement (OAuth-consented) | Consent (Art. 6(1)(a)) |
| Patreon | Patron count, monthly pledge total (OAuth-consented) | Consent (Art. 6(1)(a)) |
A creator can connect as many or as few platforms as they choose. Platforms not connected contribute ND3 (Not Available) codes — they do not penalise the creator beyond the data quality impact described below.
Metric computation
1. avg_monthly_revenue
Arithmetic mean of the last 12 months of gross income. Used as the primary income signal for advance sizing.
2. median_monthly_revenue
Median of the last 12 months. Resistant to single-month outliers; used alongside the mean to detect skewed distributions.
3. volatility_cv_12m (Coefficient of Variation)
std / mean over the last 12 months. Measures income stability. A CV of 0.0 means
perfectly stable income; a CV > 1.0 means high unpredictability.
Cap: CV is capped at 5.0 when monthly mean < €100 to prevent extreme values for very small creators.
4. max_drawdown_pct_36m
Peak-to-trough drawdown over up to 36 months. Formula:
max((peak - current) / peak) for each month where peak is the running maximum
up to that point.
5. platform_concentration_index (HHI)
Herfindahl-Hirschman Index: Σ (share_i²) across platforms. A value of 1.0 means
100% single-platform dependence; 0.25 means equal distribution across four platforms.
6. yoy_growth_pct
(avg_last12m - avg_prior12m) / avg_prior12m × 100. Only computed when 24+ months
of data are available. Null otherwise.
7. income_trend_slope_pct
Least-squares linear regression slope over the last 3 months, expressed as a percentage of the 3-month average. Positive = growing, negative = declining.
8. seasonal_adjustment_flag
True when removing high-season calendar months reduces the CV by ≥ 20% (relative). This signals that volatility is largely seasonal rather than structural.
Creator Score
The Creator Score (0–100) is a weighted composite that synthesises the Risk Profile metrics into a single signal. Higher is better.
| Component | Weight | Source field |
|---|---|---|
| Income stability | 35% | 1 - volatility_cv_12m (normalised) |
| Drawdown resilience | 25% | 1 - max_drawdown_pct_36m |
| Platform diversification | 20% | 1 - platform_concentration_index |
| Track record | 10% | track_record_months / 36 (capped) |
| Data quality | 10% | data_quality.overall_score / 100 |
Scores are further adjusted by risk-tier caps: a subprime creator's score is
hard-capped at 45, and ineligible at 30, regardless of the component calculation.
Note: The exact weights are subject to change as the model is calibrated. All
changes are versioned via the risk_version field on the Risk Profile.
Eligibility decision (RBF product)
The eligibility engine applies deterministic policy rules — there is no machine-learning
model involved. The rules as of rp_1.0.0:
| Condition | Risk Tier | Eligible |
|---|---|---|
track_record_months < 6 | ineligible | No |
CV ≤ 0.25 AND drawdown ≤ 0.40 | prime | Yes |
CV ≤ 0.50 AND drawdown ≤ 0.60 | standard | Yes |
| Otherwise | subprime | No |
Advance sizing (when eligible)
| Tier | Max advance | Revenue share | Payback cap |
|---|---|---|---|
| Prime | avg_monthly × 12 × 0.35 | 15% | 1.3× |
| Standard | avg_monthly × 12 × 0.25 | 10% | 1.5× |
These parameters are set by the lender's product configuration and may be customised per-lender. The values above are Creatorlayer reference defaults.
Data quality and ND codes
Not all fields are available for every creator. Missing or unavailable fields are recorded using the ESMA ND (Not Determined) convention:
| Code | Meaning |
|---|---|
| ND1 | Not applicable — field does not apply to this creator or platform |
| ND2 | Not collected — field applies but was not gathered in this verification run |
| ND3 | Not available — data was sought but unavailable (e.g. platform API error) |
| ND4 | Not disclosed — creator or platform declined to share |
A data_quality.overall_score of 100 indicates a fully complete tape. Lower scores
reflect missing fields and reduce the Creator Score via the data quality component.
Human review
Creators have the right to request human review of any eligibility decision made on the basis of this automated output. To exercise this right, contact the lender directly. Lenders using Creatorlayer are contractually required to provide a human review mechanism (GDPR Art. 22(3)).
Creatorlayer does not make final lending decisions — it provides structured risk data to lenders who make the final credit determination.
Versioning
Every Risk Tape carries a risk_version field (e.g. "rp_1.0.0") that identifies
the algorithm version used to generate it. When the algorithm changes materially, the
version is incremented and migration guidance is published in the changelog.
Last updated: April 2026. Algorithm version: rp_1.0.0.