Cornix Crypto Signal Marketplace Metrics: Review Sample Size Audit

What was measured

Measurement version cornix-marketplace-snapshot-v1 uses one unique CSR provider slug as the record unit and selects only rows whose exact sourceKind is Cornix supported group marketplace. It does not scrape a current Cornix page, contact reviewers, validate accounts, or reconstruct signals. All findings describe the committed 2026-07-05 snapshot.

Stored group IDs are not unique record keys

The 85 rows contain 83 distinct cornixGroupId values. 2 values occur on more than one CSR slug and cover 4 records. The audit does not publish or use the affected provider names as a ranking or accusation. A repeated source ID is a strong join candidate, but a merge decision still needs the current group route, displayed product, operator, timeline, and source history to reconcile.

This field also explains why the record denominator and source-entity denominator must stay separate. The report analyzes 85 CSR records because every other metric is stored per row. It reports 83 group IDs separately and does not pretend those are 83 verified providers, 83 active groups, or 83 independent businesses.

Most records have very small reviewer denominators

The four bucket rows reconcile to all 85 records and their stored reviewer counts sum to 331. That sum is a sum of row-level denominators, not a count of unique people: the snapshot has no reviewer identifiers, deduplication key, review dates, verified-customer flag, or cross-group overlap measure. The nearest-rank reviewer-count distribution is minimum 0, median 1, 75th percentile 4, 90th percentile 15, and maximum 39.

69 records have fewer than five reviewers, including 39 with zero. A star value backed by one or a few stored reviews can move sharply when another review arrives. The audit therefore reports denominators beside star summaries rather than ranking rows by the displayed average.

Star summaries require the reviewer context

The unweighted and reviewer-weighted means are close in this frozen cohort, but neither is a provider recommendation. An unweighted mean lets a one-review row count as much as a 39-review row. A weighted mean assumes the stored denominators are comparable and non-overlapping, which the export cannot establish. The useful conclusion is narrower: reviewer sample size is visible, frequently small, and should accompany any quotation of a marketplace star value.

The weekly activity field is derived from the monthly field

For each row, the audit calculates cornixLastMonthSignals * 3 / 13 and compares it with cornixAverageWeeklySignals. 43 rows are exactly equal in JavaScript’s binary-number representation. All 85 are within 1e-9; the maximum absolute floating-point residual is 1.1368683772161603e-13. The 69 distinct monthly values therefore create the weekly values through one deterministic transformation.

This is a data-lineage observation, not a criticism or misconduct finding. A weekly average can be a convenient display. It simply must not be cited as independent confirmation of the monthly figure. Neither field says whether the items were executable trade alerts, duplicates, updates, manual notes, profitable outcomes, losses, open positions, or signals a follower actually received.

Offer flags and exchange labels are descriptors, not terms or safety evidence

The offer rows reconcile to 85 records. A true or false flag describes the frozen export only. It cannot establish whether a trial remains available, whether payment occurs through Cornix or another party, whether a service is free after exchange or referral requirements, or whether the current terms match the snapshot.

Compatibility breadth does not measure permissions, order types, latency, rejected orders, partial fills, fees, slippage, liquidation controls, API-key scope, or current support. Verify those conditions against the current exchange and tool documentation before connecting an account.

Public verification and historical internal workflow cross-check

Before paying or copying, resolve the current operator and official route, preserve commercial terms, request a dated and complete signal archive, reconstruct all eligible signals including losses and open positions, and test follower-side execution with fees and slippage. Marketplace stars, activity counts, access flags, and compatibility labels can inform questions. They cannot answer those questions on their own.

Official context sources

These sources explain why identity, disclosure, complete performance evidence, and resistance to urgency matter. They do not validate any record in the CSR dataset.

• ESMA: finfluencer factsheet

  Transparency, accuracy, paid-promotion disclosure, and recommendation boundaries.

• CFTC: commodity trading systems sold on the internet

  Hypothetical results, fees, slippage, loss capacity, and performance-claim limitations.

Continue the transparency research

• Where the 3,200 Crypto Provider Records Came From

  Source-family and field-completeness analysis for the 3,200-record CSR provider inventory, including TGStat, RealTelegram, Cornix, manual seeds, and gaps.

• Crypto Signal Provider Subscriber Counts: Distribution and Source Bias

  Original distribution analysis of 1,244 stored subscriber observations across 3,200 CSR provider records, including quantiles, source coverage, skew, and non-claims.

• Which Crypto Signal Provider Metadata Fields Appear Together?

  Cross-tab analysis of direct routes, aliases, subscriber observations, and explicit keys across 3,200 provider records, with seven observed field patterns.

• Crypto Signal Provider Country and Language Metadata Audit

  Audit country and language labels across 3,200 crypto signal provider records, including source defaults, non-global coverage, and jurisdiction limits.

• Crypto Signal Provider Transparency Report 2026

  Original analysis of 3,200 crypto signal provider candidate records: platform concentration, binary verification, historical workflow, sources, and missing evidence.